diff options
Diffstat (limited to 'geo-replication/syncdaemon')
21 files changed, 5466 insertions, 3184 deletions
diff --git a/geo-replication/syncdaemon/Makefile.am b/geo-replication/syncdaemon/Makefile.am index 885963eae2b..d70e3368faf 100644 --- a/geo-replication/syncdaemon/Makefile.am +++ b/geo-replication/syncdaemon/Makefile.am @@ -1,7 +1,8 @@ -syncdaemondir = $(libexecdir)/glusterfs/python/syncdaemon +syncdaemondir = $(GLUSTERFS_LIBEXECDIR)/python/syncdaemon -syncdaemon_PYTHON = gconf.py gsyncd.py __init__.py master.py README.md repce.py \ - resource.py configinterface.py syncdutils.py monitor.py libcxattr.py \ - $(top_builddir)/contrib/ipaddr-py/ipaddr.py libgfchangelog.py changelogagent.py +syncdaemon_PYTHON = rconf.py gsyncd.py __init__.py master.py README.md repce.py \ + resource.py syncdutils.py monitor.py libcxattr.py gsyncdconfig.py \ + libgfchangelog.py gsyncdstatus.py conf.py logutils.py \ + subcmds.py argsupgrade.py py2py3.py CLEANFILES = diff --git a/geo-replication/syncdaemon/README.md b/geo-replication/syncdaemon/README.md index 67f346ace5a..5ab785ae669 100644 --- a/geo-replication/syncdaemon/README.md +++ b/geo-replication/syncdaemon/README.md @@ -12,14 +12,13 @@ Requirements are categorized according to this. * OpenSSH >= 4.0 (master) / SSH2 compliant sshd (eg. openssh) (slave) * rsync (both) * glusterfs: with marker and changelog support (master & slave); -* FUSE: glusterfs fuse module with auxilary gfid based access support +* FUSE: glusterfs fuse module with auxiliary gfid based access support INSTALLATION ------------ As of now, the supported way of operation is running from the source directory or using the RPMs given. -If you use Python 2.4.x, you need to install the [Ctypes module](http://python.net/crew/theller/ctypes/). CONFIGURATION ------------- @@ -45,7 +44,7 @@ in the source tree. USAGE ----- -gsyncd is a utilitly for continous mirroring, ie. it mirrors master to slave incrementally. +gsyncd is a utilitly for continuous mirroring, ie. it mirrors master to slave incrementally. Assume we have a gluster volume _pop_ at localhost. We try to set up the mirroring for volume _pop_ using gsyncd for gluster volume _moz_ on remote machine/cluster @ example.com. The respective gsyncd invocations are (demoing some syntax sugaring): diff --git a/geo-replication/syncdaemon/__codecheck.py b/geo-replication/syncdaemon/__codecheck.py index 45dbd26bb64..9437147f7d9 100644 --- a/geo-replication/syncdaemon/__codecheck.py +++ b/geo-replication/syncdaemon/__codecheck.py @@ -8,6 +8,7 @@ # cases as published by the Free Software Foundation. # +from __future__ import print_function import os import os.path import sys @@ -45,7 +46,7 @@ class IPNetwork(list): gsyncd = sys.modules['gsyncd'] for a in [['--help'], ['--version'], ['--canonicalize-escape-url', '/foo']]: - print('>>> invoking program with args: %s' % ' '.join(a)) + print(('>>> invoking program with args: %s' % ' '.join(a))) pid = os.fork() if not pid: sys_argv_set(a) diff --git a/geo-replication/syncdaemon/argsupgrade.py b/geo-replication/syncdaemon/argsupgrade.py new file mode 100644 index 00000000000..7af40633ef8 --- /dev/null +++ b/geo-replication/syncdaemon/argsupgrade.py @@ -0,0 +1,359 @@ +# -*- coding: utf-8 -*- +# +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. +# +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# +# Converts old style args into new style args + +from __future__ import print_function +import sys +from argparse import ArgumentParser +import socket +import os + +from syncdutils import GsyncdError +from conf import GLUSTERD_WORKDIR + + +def gethostbyname(hnam): + """gethostbyname wrapper""" + try: + return socket.gethostbyname(hnam) + except socket.gaierror: + ex = sys.exc_info()[1] + raise GsyncdError("failed to resolve %s: %s" % + (hnam, ex.strerror)) + + +def slave_url(urldata): + urldata = urldata.replace("ssh://", "") + host, vol = urldata.split("::") + vol = vol.split(":")[0] + return "%s::%s" % (host, vol) + + +def init_gsyncd_template_conf(): + path = GLUSTERD_WORKDIR + "/geo-replication/gsyncd_template.conf" + dname = os.path.dirname(path) + if not os.path.exists(dname): + try: + os.mkdir(dname) + except OSError: + pass + + if not os.path.exists(path): + fd = os.open(path, os.O_CREAT | os.O_RDWR) + os.close(fd) + + +def init_gsyncd_session_conf(master, slave): + slave = slave_url(slave) + master = master.strip(":") + slavehost, slavevol = slave.split("::") + slavehost = slavehost.split("@")[-1] + + # Session Config File + path = "%s/geo-replication/%s_%s_%s/gsyncd.conf" % ( + GLUSTERD_WORKDIR, master, slavehost, slavevol) + + if os.path.exists(os.path.dirname(path)) and not os.path.exists(path): + fd = os.open(path, os.O_CREAT | os.O_RDWR) + os.close(fd) + + +def init_gsyncd_conf(path): + dname = os.path.dirname(path) + if not os.path.exists(dname): + try: + os.mkdir(dname) + except OSError: + pass + + if os.path.exists(dname) and not os.path.exists(path): + fd = os.open(path, os.O_CREAT | os.O_RDWR) + os.close(fd) + + +def upgrade(): + # Create dummy template conf(empty), hack to avoid glusterd + # fail when it does stat to check the existence. + init_gsyncd_template_conf() + + inet6 = False + if "--inet6" in sys.argv: + inet6 = True + + if "--monitor" in sys.argv: + # python gsyncd.py --path=/bricks/b1 + # --monitor -c gsyncd.conf + # --iprefix=/var :gv1 + # --glusterd-uuid=f26ac7a8-eb1b-4ea7-959c-80b27d3e43d0 + # f241::gv2 + p = ArgumentParser() + p.add_argument("master") + p.add_argument("slave") + p.add_argument("--glusterd-uuid") + p.add_argument("-c") + p.add_argument("--iprefix") + p.add_argument("--path", action="append") + pargs = p.parse_known_args(sys.argv[1:])[0] + + # Overwrite the sys.argv after rearrange + init_gsyncd_session_conf(pargs.master, pargs.slave) + sys.argv = [ + sys.argv[0], + "monitor", + pargs.master.strip(":"), + slave_url(pargs.slave), + "--local-node-id", + pargs.glusterd_uuid + ] + elif "--status-get" in sys.argv: + # -c gsyncd.conf --iprefix=/var :gv1 f241::gv2 + # --status-get --path /bricks/b1 + p = ArgumentParser() + p.add_argument("master") + p.add_argument("slave") + p.add_argument("-c") + p.add_argument("--path") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + sys.argv = [ + sys.argv[0], + "status", + pargs.master.strip(":"), + slave_url(pargs.slave), + "--local-path", + pargs.path + ] + elif "--canonicalize-url" in sys.argv: + # This can accept multiple URLs and converts each URL to the + # format ssh://USER@IP:gluster://127.0.0.1:VOLUME + # This format not used in gsyncd, but added for glusterd compatibility + p = ArgumentParser() + p.add_argument("--canonicalize-url", nargs="+") + pargs = p.parse_known_args(sys.argv[1:])[0] + + for url in pargs.canonicalize_url: + host, vol = url.split("::") + host = host.replace("ssh://", "") + remote_addr = host + if "@" not in remote_addr: + remote_addr = "root@" + remote_addr + + user, hname = remote_addr.split("@") + + if not inet6: + hname = gethostbyname(hname) + + print(("ssh://%s@%s:gluster://127.0.0.1:%s" % ( + user, hname, vol))) + + sys.exit(0) + elif "--normalize-url" in sys.argv: + # Adds schema prefix as ssh:// + # This format not used in gsyncd, but added for glusterd compatibility + p = ArgumentParser() + p.add_argument("--normalize-url") + pargs = p.parse_known_args(sys.argv[1:])[0] + print(("ssh://%s" % slave_url(pargs.normalize_url))) + sys.exit(0) + elif "--config-get-all" in sys.argv: + # -c gsyncd.conf --iprefix=/var :gv1 f241::gv2 --config-get-all + p = ArgumentParser() + p.add_argument("master") + p.add_argument("slave") + p.add_argument("-c") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + sys.argv = [ + sys.argv[0], + "config-get", + pargs.master.strip(":"), + slave_url(pargs.slave), + "--show-defaults", + "--use-underscore" + ] + elif "--verify" in sys.argv and "spawning" in sys.argv: + # Just checks that able to spawn gsyncd or not + sys.exit(0) + elif "--slavevoluuid-get" in sys.argv: + # --slavevoluuid-get f241::gv2 + p = ArgumentParser() + p.add_argument("--slavevoluuid-get") + p.add_argument("-c") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + host, vol = pargs.slavevoluuid_get.split("::") + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "voluuidget", + host, + vol + ] + elif "--config-set-rx" in sys.argv: + # Not required since default conf is not generated + # and custom conf generated only when required + # -c gsyncd.conf --config-set-rx remote-gsyncd + # /usr/local/libexec/glusterfs/gsyncd . . + # Touch the gsyncd.conf file and create session + # directory if required + p = ArgumentParser() + p.add_argument("-c", dest="config_file") + pargs = p.parse_known_args(sys.argv[1:])[0] + + # If not template conf then it is trying to create + # session config, create a empty file instead + if pargs.config_file.endswith("gsyncd.conf"): + init_gsyncd_conf(pargs.config_file) + sys.exit(0) + elif "--create" in sys.argv: + # To update monitor status file + # --create Created -c gsyncd.conf + # --iprefix=/var :gv1 f241::gv2 + p = ArgumentParser() + p.add_argument("--create") + p.add_argument("master") + p.add_argument("slave") + p.add_argument("-c") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "monitor-status", + pargs.master.strip(":"), + slave_url(pargs.slave), + pargs.create + ] + elif "--config-get" in sys.argv: + # -c gsyncd.conf --iprefix=/var :gv1 f241::gv2 --config-get pid-file + p = ArgumentParser() + p.add_argument("--config-get") + p.add_argument("master") + p.add_argument("slave") + p.add_argument("-c") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "config-get", + pargs.master.strip(":"), + slave_url(pargs.slave), + "--only-value", + "--show-defaults", + "--name", + pargs.config_get.replace("_", "-") + ] + elif "--config-set" in sys.argv: + # ignore session-owner + if "session-owner" in sys.argv: + sys.exit(0) + + # --path=/bricks/b1 -c gsyncd.conf :gv1 f241::gv2 + # --config-set log_level DEBUG + p = ArgumentParser() + p.add_argument("master") + p.add_argument("slave") + p.add_argument("--config-set", action='store_true') + p.add_argument("name") + p.add_argument("--value") + p.add_argument("-c") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "config-set", + pargs.master.strip(":"), + slave_url(pargs.slave), + "--name=%s" % pargs.name, + "--value=%s" % pargs.value + ] + elif "--config-check" in sys.argv: + # --config-check georep_session_working_dir + p = ArgumentParser() + p.add_argument("--config-check") + p.add_argument("-c") + pargs = p.parse_known_args(sys.argv[1:])[0] + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "config-check", + pargs.config_check.replace("_", "-") + ] + elif "--config-del" in sys.argv: + # -c gsyncd.conf --iprefix=/var :gv1 f241::gv2 --config-del log_level + p = ArgumentParser() + p.add_argument("--config-del") + p.add_argument("master") + p.add_argument("slave") + p.add_argument("-c") + p.add_argument("--iprefix") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "config-reset", + pargs.master.strip(":"), + slave_url(pargs.slave), + pargs.config_del.replace("_", "-") + ] + elif "--delete" in sys.argv: + # --delete -c gsyncd.conf --iprefix=/var + # --path-list=--path=/bricks/b1 :gv1 f241::gv2 + p = ArgumentParser() + p.add_argument("--reset-sync-time", action="store_true") + p.add_argument("--path-list") + p.add_argument("master") + p.add_argument("slave") + p.add_argument("--iprefix") + p.add_argument("-c") + pargs = p.parse_known_args(sys.argv[1:])[0] + + init_gsyncd_session_conf(pargs.master, pargs.slave) + + paths = pargs.path_list.split("--path=") + paths = ["--path=%s" % x.strip() for x in paths if x.strip() != ""] + + # Modified sys.argv + sys.argv = [ + sys.argv[0], + "delete", + pargs.master.strip(":"), + slave_url(pargs.slave) + ] + sys.argv += paths + + if pargs.reset_sync_time: + sys.argv.append("--reset-sync-time") + + if inet6: + # Add `--inet6` as first argument + sys.argv = [sys.argv[0], "--inet6"] + sys.argv[1:] diff --git a/geo-replication/syncdaemon/changelogagent.py b/geo-replication/syncdaemon/changelogagent.py deleted file mode 100644 index 54d82cefcd2..00000000000 --- a/geo-replication/syncdaemon/changelogagent.py +++ /dev/null @@ -1,78 +0,0 @@ -#!/usr/bin/env python -# -# Copyright (c) 2011-2014 Red Hat, Inc. <http://www.redhat.com> -# This file is part of GlusterFS. - -# This file is licensed to you under your choice of the GNU Lesser -# General Public License, version 3 or any later version (LGPLv3 or -# later), or the GNU General Public License, version 2 (GPLv2), in all -# cases as published by the Free Software Foundation. -# - -import os -import logging -import syncdutils -from syncdutils import select, CHANGELOG_AGENT_SERVER_VERSION -from repce import RepceServer - - -class _MetaChangelog(object): - - def __getattr__(self, meth): - from libgfchangelog import Changes as LChanges - xmeth = [m for m in dir(LChanges) if m[0] != '_'] - if meth not in xmeth: - return - for m in xmeth: - setattr(self, m, getattr(LChanges, m)) - return getattr(self, meth) - -Changes = _MetaChangelog() - - -class Changelog(object): - def version(self): - return CHANGELOG_AGENT_SERVER_VERSION - - def register(self, cl_brick, cl_dir, cl_log, cl_level, retries=0): - return Changes.cl_register(cl_brick, cl_dir, cl_log, cl_level, retries) - - def scan(self): - return Changes.cl_scan() - - def getchanges(self): - return Changes.cl_getchanges() - - def done(self, clfile): - return Changes.cl_done(clfile) - - def history(self, changelog_path, start, end, num_parallel): - return Changes.cl_history_changelog(changelog_path, start, end, - num_parallel) - - def history_scan(self): - return Changes.cl_history_scan() - - def history_getchanges(self): - return Changes.cl_history_getchanges() - - def history_done(self, clfile): - return Changes.cl_history_done(clfile) - - -class ChangelogAgent(object): - def __init__(self, obj, fd_tup): - (inf, ouf, rw, ww) = fd_tup.split(',') - os.close(int(rw)) - os.close(int(ww)) - repce = RepceServer(obj, int(inf), int(ouf), 1) - t = syncdutils.Thread(target=lambda: (repce.service_loop(), - syncdutils.finalize())) - t.start() - logging.info('Agent listining...') - - select((), (), ()) - - -def agent(obj, fd_tup): - return ChangelogAgent(obj, fd_tup) diff --git a/geo-replication/syncdaemon/conf.py.in b/geo-replication/syncdaemon/conf.py.in new file mode 100644 index 00000000000..2042fa9cdfb --- /dev/null +++ b/geo-replication/syncdaemon/conf.py.in @@ -0,0 +1,17 @@ +# +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. + +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +GLUSTERFS_LIBEXECDIR = '@GLUSTERFS_LIBEXECDIR@' +GLUSTERD_WORKDIR = "@GLUSTERD_WORKDIR@" + +LOCALSTATEDIR = "@localstatedir@" +UUID_FILE = "@GLUSTERD_WORKDIR@/glusterd.info" +GLUSTERFS_CONFDIR = "@SYSCONF_DIR@/glusterfs" +GCONF_VERSION = 4.0 diff --git a/geo-replication/syncdaemon/configinterface.py.in b/geo-replication/syncdaemon/configinterface.py.in deleted file mode 100644 index acb51486a10..00000000000 --- a/geo-replication/syncdaemon/configinterface.py.in +++ /dev/null @@ -1,352 +0,0 @@ -# -# Copyright (c) 2011-2014 Red Hat, Inc. <http://www.redhat.com> -# This file is part of GlusterFS. - -# This file is licensed to you under your choice of the GNU Lesser -# General Public License, version 3 or any later version (LGPLv3 or -# later), or the GNU General Public License, version 2 (GPLv2), in all -# cases as published by the Free Software Foundation. -# - -try: - import ConfigParser -except ImportError: - # py 3 - import configparser as ConfigParser -import re -from string import Template -import os -import errno -import sys -from stat import ST_DEV, ST_INO, ST_MTIME -import tempfile -import shutil - -from syncdutils import escape, unescape, norm, update_file, GsyncdError - -SECT_ORD = '__section_order__' -SECT_META = '__meta__' -config_version = 2.0 - -re_type = type(re.compile('')) - - -# (SECTION, OPTION, OLD VALUE, NEW VALUE) -CONFIGS = ( - ("peersrx . .", - "georep_session_working_dir", - "", - "@GLUSTERD_WORKDIR@/geo-replication/${mastervol}_${remotehost}_" - "${slavevol}/"), - ("peersrx .", - "gluster_params", - "aux-gfid-mount xlator-option=\*-dht.assert-no-child-down=true", - "aux-gfid-mount"), - ("peersrx . .", - "ssh_command_tar", - "", - "ssh -oPasswordAuthentication=no -oStrictHostKeyChecking=no " - "-i @GLUSTERD_WORKDIR@/geo-replication/tar_ssh.pem"), - ("peersrx . .", - "changelog_log_file", - "", - "${iprefix}/log/glusterfs/geo-replication/${mastervol}" - "/${eSlave}${local_id}-changes.log"), - ("peersrx . .", - "working_dir", - "@localstatedir@/run/gluster/${mastervol}/${eSlave}", - "${iprefix}/lib/misc/glusterfsd/${mastervol}/${eSlave}"), -) - - -def upgrade_config_file(path): - config_change = False - config = ConfigParser.RawConfigParser() - config.read(path) - - for sec, opt, oldval, newval in CONFIGS: - try: - val = config.get(sec, opt) - except ConfigParser.NoOptionError: - # if new config opt not exists - config_change = True - config.set(sec, opt, newval) - continue - except ConfigParser.Error: - """ - When gsyncd invoked at the time of create, config file - will not be their. Ignore any ConfigParser errors - """ - continue - - if val == newval: - # value is same as new val - continue - - if val == oldval: - # config value needs update - config_change = True - config.set(sec, opt, newval) - - if config_change: - tempConfigFile = tempfile.NamedTemporaryFile(mode="wb", delete=False) - with open(tempConfigFile.name, 'wb') as configFile: - config.write(configFile) - - # If src and dst are two different file system, then os.rename - # fails, In this case if temp file created in /tmp and if /tmp is - # seperate fs then os.rename gives following error, so use shutil - # OSError: [Errno 18] Invalid cross-device link - # mail.python.org/pipermail/python-list/2005-February/342893.html - shutil.move(tempConfigFile.name, path) - - -class MultiDict(object): - - """a virtual dict-like class which functions as the union - of underlying dicts""" - - def __init__(self, *dd): - self.dicts = dd - - def __getitem__(self, key): - val = None - for d in self.dicts: - if d.get(key) is not None: - val = d[key] - if val is None: - raise KeyError(key) - return val - - -class GConffile(object): - - """A high-level interface to ConfigParser which flattens the two-tiered - config layout by implenting automatic section dispatch based on initial - parameters. - - Also ensure section ordering in terms of their time of addition -- a compat - hack for Python < 2.7. - """ - - def _normconfig(self): - """normalize config keys by s/-/_/g""" - for n, s in self.config._sections.items(): - if n.find('__') == 0: - continue - s2 = type(s)() - for k, v in s.items(): - if k.find('__') != 0: - k = norm(k) - s2[k] = v - self.config._sections[n] = s2 - - def __init__(self, path, peers, *dd): - """ - - .path: location of config file - - .config: underlying ConfigParser instance - - .peers: on behalf of whom we flatten .config - (master, or master-slave url pair) - - .auxdicts: template subtituents - """ - self.peers = peers - self.path = path - self.auxdicts = dd - self.config = ConfigParser.RawConfigParser() - self.config.read(path) - self.dev, self.ino, self.mtime = -1, -1, -1 - self._normconfig() - - def _load(self): - try: - sres = os.stat(self.path) - self.dev = sres[ST_DEV] - self.ino = sres[ST_INO] - self.mtime = sres[ST_MTIME] - except (OSError, IOError): - if sys.exc_info()[1].errno == errno.ENOENT: - sres = None - - self.config = ConfigParser.RawConfigParser() - self.config.read(self.path) - self._normconfig() - - def get_realtime(self, opt): - try: - sres = os.stat(self.path) - except (OSError, IOError): - if sys.exc_info()[1].errno == errno.ENOENT: - sres = None - else: - raise - - # compare file system stat with that of our stream file handle - if not sres or sres[ST_DEV] != self.dev or \ - sres[ST_INO] != self.ino or self.mtime != sres[ST_MTIME]: - self._load() - - return self.get(opt, printValue=False) - - def section(self, rx=False): - """get the section name of the section representing .peers - in .config""" - peers = self.peers - if not peers: - peers = ['.', '.'] - rx = True - if rx: - st = 'peersrx' - else: - st = 'peers' - return ' '.join([st] + [escape(u) for u in peers]) - - @staticmethod - def parse_section(section): - """retrieve peers sequence encoded by section name - (as urls or regexen, depending on section type) - """ - sl = section.split() - st = sl.pop(0) - sl = [unescape(u) for u in sl] - if st == 'peersrx': - sl = [re.compile(u) for u in sl] - return sl - - def ord_sections(self): - """Return an ordered list of sections. - - Ordering happens based on the auxiliary - SECT_ORD section storing indices for each - section added through the config API. - - To not to go corrupt in case of manually - written config files, we take care to append - also those sections which are not registered - in SECT_ORD. - - Needed for python 2.{4,5,6} where ConfigParser - cannot yet order sections/options internally. - """ - so = {} - if self.config.has_section(SECT_ORD): - so = self.config._sections[SECT_ORD] - so2 = {} - for k, v in so.items(): - if k != '__name__': - so2[k] = int(v) - tv = 0 - if so2: - tv = max(so2.values()) + 1 - ss = [s for s in self.config.sections() if s.find('__') != 0] - for s in ss: - if s in so.keys(): - continue - so2[s] = tv - tv += 1 - - def scmp(x, y): - return cmp(*(so2[s] for s in (x, y))) - ss.sort(scmp) - return ss - - def update_to(self, dct, allow_unresolved=False): - """update @dct from key/values of ours. - - key/values are collected from .config by filtering the regexp sections - according to match, and from .section. The values are treated as - templates, which are substituted from .auxdicts and (in case of regexp - sections) match groups. - """ - if not self.peers: - raise GsyncdError('no peers given, cannot select matching options') - - def update_from_sect(sect, mud): - for k, v in self.config._sections[sect].items(): - if k == '__name__': - continue - if allow_unresolved: - dct[k] = Template(v).safe_substitute(mud) - else: - dct[k] = Template(v).substitute(mud) - for sect in self.ord_sections(): - sp = self.parse_section(sect) - if isinstance(sp[0], re_type) and len(sp) == len(self.peers): - match = True - mad = {} - for i in range(len(sp)): - m = sp[i].search(self.peers[i]) - if not m: - match = False - break - for j in range(len(m.groups())): - mad['match%d_%d' % (i + 1, j + 1)] = m.groups()[j] - if match: - update_from_sect(sect, MultiDict(dct, mad, *self.auxdicts)) - if self.config.has_section(self.section()): - update_from_sect(self.section(), MultiDict(dct, *self.auxdicts)) - - def get(self, opt=None, printValue=True): - """print the matching key/value pairs from .config, - or if @opt given, the value for @opt (according to the - logic described in .update_to) - """ - d = {} - self.update_to(d, allow_unresolved=True) - if opt: - opt = norm(opt) - v = d.get(opt) - if v: - if printValue: - print(v) - else: - return v - else: - for k, v in d.iteritems(): - if k == '__name__': - continue - print("%s: %s" % (k, v)) - - def write(self, trfn, opt, *a, **kw): - """update on-disk config transactionally - - @trfn is the transaction function - """ - def mergeconf(f): - self.config = ConfigParser.RawConfigParser() - self.config.readfp(f) - self._normconfig() - if not self.config.has_section(SECT_META): - self.config.add_section(SECT_META) - self.config.set(SECT_META, 'version', config_version) - return trfn(norm(opt), *a, **kw) - - def updateconf(f): - self.config.write(f) - update_file(self.path, updateconf, mergeconf) - - def _set(self, opt, val, rx=False): - """set @opt to @val in .section""" - sect = self.section(rx) - if not self.config.has_section(sect): - self.config.add_section(sect) - # regarding SECT_ORD, cf. ord_sections - if not self.config.has_section(SECT_ORD): - self.config.add_section(SECT_ORD) - self.config.set( - SECT_ORD, sect, len(self.config._sections[SECT_ORD])) - self.config.set(sect, opt, val) - return True - - def set(self, opt, *a, **kw): - """perform ._set transactionally""" - self.write(self._set, opt, *a, **kw) - - def _delete(self, opt, rx=False): - """delete @opt from .section""" - sect = self.section(rx) - if self.config.has_section(sect): - return self.config.remove_option(sect, opt) - - def delete(self, opt, *a, **kw): - """perform ._delete transactionally""" - self.write(self._delete, opt, *a, **kw) diff --git a/geo-replication/syncdaemon/gsyncd.py b/geo-replication/syncdaemon/gsyncd.py index cb29903bf4f..257ed72c6ae 100644 --- a/geo-replication/syncdaemon/gsyncd.py +++ b/geo-replication/syncdaemon/gsyncd.py @@ -1,647 +1,325 @@ -#!/usr/bin/env python +#!/usr/bin/python3 +# -*- coding: utf-8 -*- # -# Copyright (c) 2011-2014 Red Hat, Inc. <http://www.redhat.com> -# This file is part of GlusterFS. - -# This file is licensed to you under your choice of the GNU Lesser -# General Public License, version 3 or any later version (LGPLv3 or -# later), or the GNU General Public License, version 2 (GPLv2), in all -# cases as published by the Free Software Foundation. +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. +# +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. # +from argparse import ArgumentParser +import time import os -import os.path -import glob +from errno import EEXIST import sys -import time import logging -import shutil -import optparse -import fcntl -import fnmatch -from optparse import OptionParser, SUPPRESS_HELP -from logging import Logger, handlers -from errno import ENOENT - -from ipaddr import IPAddress, IPNetwork - -from gconf import gconf -from syncdutils import FreeObject, norm, grabpidfile, finalize -from syncdutils import log_raise_exception, privileged, update_file -from syncdutils import GsyncdError, select, set_term_handler -from configinterface import GConffile, upgrade_config_file -import resource -from monitor import monitor -from changelogagent import agent, Changelog - - -class GLogger(Logger): - - """Logger customizations for gsyncd. - - It implements a log format similar to that of glusterfs. - """ - - def makeRecord(self, name, level, *a): - rv = Logger.makeRecord(self, name, level, *a) - rv.nsecs = (rv.created - int(rv.created)) * 1000000 - fr = sys._getframe(4) - callee = fr.f_locals.get('self') - if callee: - ctx = str(type(callee)).split("'")[1].split('.')[-1] - else: - ctx = '<top>' - if not hasattr(rv, 'funcName'): - rv.funcName = fr.f_code.co_name - rv.lvlnam = logging.getLevelName(level)[0] - rv.ctx = ctx - return rv - - @classmethod - def setup(cls, **kw): - lbl = kw.get('label', "") - if lbl: - lbl = '(' + lbl + ')' - lprm = {'datefmt': "%Y-%m-%d %H:%M:%S", - 'format': "[%(asctime)s.%(nsecs)d] %(lvlnam)s [%(module)s" + - lbl + ":%(lineno)s:%(funcName)s] %(ctx)s: %(message)s"} - lprm.update(kw) - lvl = kw.get('level', logging.INFO) - lprm['level'] = lvl - logging.root = cls("root", lvl) - logging.setLoggerClass(cls) - logging.getLogger().handlers = [] - logging.getLogger().setLevel(lprm['level']) - - if 'filename' in lprm: - try: - logging_handler = handlers.WatchedFileHandler(lprm['filename']) - formatter = logging.Formatter(fmt=lprm['format'], - datefmt=lprm['datefmt']) - logging_handler.setFormatter(formatter) - logging.getLogger().addHandler(logging_handler) - except AttributeError: - # Python version < 2.6 will not have WatchedFileHandler - # so fallback to logging without any handler. - # Note: logrotate will not work if Python version is < 2.6 - logging.basicConfig(**lprm) - else: - # If filename not passed(not available in lprm) then it may be - # streaming.(Ex: {"stream": "/dev/stdout"}) - logging.basicConfig(**lprm) - - @classmethod - def _gsyncd_loginit(cls, **kw): - lkw = {} - if gconf.log_level: - lkw['level'] = gconf.log_level - if kw.get('log_file'): - if kw['log_file'] in ('-', '/dev/stderr'): - lkw['stream'] = sys.stderr - elif kw['log_file'] == '/dev/stdout': - lkw['stream'] = sys.stdout - else: - lkw['filename'] = kw['log_file'] - - cls.setup(label=kw.get('label'), **lkw) - - lkw.update({'saved_label': kw.get('label')}) - gconf.log_metadata = lkw - gconf.log_exit = True - - -def startup(**kw): - """set up logging, pidfile grabbing, daemonization""" - if getattr(gconf, 'pid_file', None) and kw.get('go_daemon') != 'postconn': - if not grabpidfile(): - sys.stderr.write("pidfile is taken, exiting.\n") - sys.exit(2) - gconf.pid_file_owned = True - - if kw.get('go_daemon') == 'should': - x, y = os.pipe() - gconf.cpid = os.fork() - if gconf.cpid: - os.close(x) - sys.exit() - os.close(y) - os.setsid() - dn = os.open(os.devnull, os.O_RDWR) - for f in (sys.stdin, sys.stdout, sys.stderr): - os.dup2(dn, f.fileno()) - if getattr(gconf, 'pid_file', None): - if not grabpidfile(gconf.pid_file + '.tmp'): - raise GsyncdError("cannot grab temporary pidfile") - os.rename(gconf.pid_file + '.tmp', gconf.pid_file) - # wait for parent to terminate - # so we can start up with - # no messing from the dirty - # ol' bustard - select((x,), (), ()) - os.close(x) - - GLogger._gsyncd_loginit(**kw) - - -def _unlink(path): - try: - os.unlink(path) - except (OSError, IOError): - if sys.exc_info()[1].errno == ENOENT: - pass - else: - raise GsyncdError('Unlink error: %s' % path) + +from logutils import setup_logging +import gsyncdconfig as gconf +from rconf import rconf +import subcmds +from conf import GLUSTERD_WORKDIR, GLUSTERFS_CONFDIR, GCONF_VERSION +from syncdutils import (set_term_handler, finalize, lf, + log_raise_exception, FreeObject, escape) +import argsupgrade + + +GSYNCD_VERSION = "gsyncd.py %s.0" % GCONF_VERSION def main(): - """main routine, signal/exception handling boilerplates""" - gconf.starttime = time.time() + rconf.starttime = time.time() + + # If old Glusterd sends commands in old format, below function + # converts the sys.argv to new format. This conversion is added + # temporarily for backward compatibility. This can be removed + # once integrated with Glusterd2 + # This modifies sys.argv globally, so rest of the code works as usual + argsupgrade.upgrade() + + # Default argparse version handler prints to stderr, which is fixed in + # 3.x series but not in 2.x, using custom parser to fix this issue + if "--version" in sys.argv: + print(GSYNCD_VERSION) + sys.exit(0) + + parser = ArgumentParser() + parser.add_argument("--inet6", action="store_true") + sp = parser.add_subparsers(dest="subcmd") + + # Monitor Status File update + p = sp.add_parser("monitor-status") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave details user@host::vol format") + p.add_argument("status", help="Update Monitor Status") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + + # Monitor + p = sp.add_parser("monitor") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave details user@host::vol format") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--pause-on-start", + action="store_true", + help="Start with Paused state") + p.add_argument("--local-node-id", help="Local Node ID") + p.add_argument("--debug", action="store_true") + p.add_argument("--use-gconf-volinfo", action="store_true") + + # Worker + p = sp.add_parser("worker") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave details user@host::vol format") + p.add_argument("--local-path", help="Local Brick Path") + p.add_argument("--feedback-fd", type=int, + help="feedback fd between monitor and worker") + p.add_argument("--local-node", help="Local master node") + p.add_argument("--local-node-id", help="Local Node ID") + p.add_argument("--subvol-num", type=int, help="Subvolume number") + p.add_argument("--is-hottier", action="store_true", + help="Is this brick part of hot tier") + p.add_argument("--resource-remote", + help="Remote node to connect to Slave Volume") + p.add_argument("--resource-remote-id", + help="Remote node ID to connect to Slave Volume") + p.add_argument("--slave-id", help="Slave Volume ID") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + + # Slave + p = sp.add_parser("slave") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave details user@host::vol format") + p.add_argument("--session-owner") + p.add_argument("--master-brick", + help="Master brick which is connected to the Slave") + p.add_argument("--master-node", + help="Master node which is connected to the Slave") + p.add_argument("--master-node-id", + help="Master node ID which is connected to the Slave") + p.add_argument("--local-node", help="Local Slave node") + p.add_argument("--local-node-id", help="Local Slave ID") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + + # All configurations which are configured via "slave-" options + # DO NOT add default values for these configurations, default values + # will be picked from template config file + p.add_argument("--slave-timeout", type=int, + help="Timeout to end gsyncd at Slave side") + p.add_argument("--use-rsync-xattrs", action="store_true") + p.add_argument("--slave-log-level", help="Slave Gsyncd Log level") + p.add_argument("--slave-gluster-log-level", + help="Slave Gluster mount Log level") + p.add_argument("--slave-gluster-command-dir", + help="Directory where Gluster binaries exist on slave") + p.add_argument("--slave-access-mount", action="store_true", + help="Do not lazy umount the slave volume") + p.add_argument("--master-dist-count", type=int, + help="Master Distribution count") + + # Status + p = sp.add_parser("status") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--local-path", help="Local Brick Path") + p.add_argument("--debug", action="store_true") + p.add_argument("--json", action="store_true") + + # Config-check + p = sp.add_parser("config-check") + p.add_argument("name", help="Config Name") + p.add_argument("--value", help="Config Value") + p.add_argument("--debug", action="store_true") + + # Config-get + p = sp.add_parser("config-get") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave") + p.add_argument("--name", help="Config Name") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + p.add_argument("--show-defaults", action="store_true") + p.add_argument("--only-value", action="store_true") + p.add_argument("--use-underscore", action="store_true") + p.add_argument("--json", action="store_true") + + # Config-set + p = sp.add_parser("config-set") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave") + p.add_argument("-n", "--name", help="Config Name") + p.add_argument("-v", "--value", help="Config Value") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + + # Config-reset + p = sp.add_parser("config-reset") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave") + p.add_argument("name", help="Config Name") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument("--debug", action="store_true") + + # voluuidget + p = sp.add_parser("voluuidget") + p.add_argument("host", help="Hostname") + p.add_argument("volname", help="Volume Name") + p.add_argument("--debug", action="store_true") + + # Delete + p = sp.add_parser("delete") + p.add_argument("master", help="Master Volume Name") + p.add_argument("slave", help="Slave") + p.add_argument("-c", "--config-file", help="Config File") + p.add_argument('--path', dest='paths', action="append") + p.add_argument("--reset-sync-time", action="store_true", + help="Reset Sync Time") + p.add_argument("--debug", action="store_true") + + # Parse arguments + args = parser.parse_args() + + # Extra template values, All arguments are already part of template + # variables, use this for adding extra variables + extra_tmpl_args = {} + + # Add First/Primary Slave host, user and volume + if getattr(args, "slave", None) is not None: + hostdata, slavevol = args.slave.split("::") + hostdata = hostdata.split("@") + slavehost = hostdata[-1] + slaveuser = "root" + if len(hostdata) == 2: + slaveuser = hostdata[0] + extra_tmpl_args["primary_slave_host"] = slavehost + extra_tmpl_args["slaveuser"] = slaveuser + extra_tmpl_args["slavevol"] = slavevol + + # Add Bricks encoded path + if getattr(args, "local_path", None) is not None: + extra_tmpl_args["local_id"] = escape(args.local_path) + + # Add Master Bricks encoded path(For Slave) + if getattr(args, "master_brick", None) is not None: + extra_tmpl_args["master_brick_id"] = escape(args.master_brick) + + # Load configurations + config_file = getattr(args, "config_file", None) + + # Subcmd accepts config file argument but not passed + # Set default path for config file in that case + # If an subcmd accepts config file then it also accepts + # master and Slave arguments. + if config_file is None and hasattr(args, "config_file") \ + and args.subcmd != "slave": + config_file = "%s/geo-replication/%s_%s_%s/gsyncd.conf" % ( + GLUSTERD_WORKDIR, + args.master, + extra_tmpl_args["primary_slave_host"], + extra_tmpl_args["slavevol"]) + + # If Config file path not exists, log error and continue using default conf + config_file_error_msg = None + if config_file is not None and not os.path.exists(config_file): + # Logging not yet initialized, create the error message to + # log later and reset the config_file to None + config_file_error_msg = lf( + "Session config file not exists, using the default config", + path=config_file) + config_file = None + + rconf.config_file = config_file + + # Override gconf values from argument values only if it is slave gsyncd + override_from_args = False + if args.subcmd == "slave": + override_from_args = True + + if config_file is not None and \ + args.subcmd in ["monitor", "config-get", "config-set", "config-reset"]: + ret = gconf.is_config_file_old(config_file, args.master, extra_tmpl_args["slavevol"]) + if ret is not None: + gconf.config_upgrade(config_file, ret) + + # Load Config file + gconf.load(GLUSTERFS_CONFDIR + "/gsyncd.conf", + config_file, + vars(args), + extra_tmpl_args, + override_from_args) + + # Default label to print in log file + label = args.subcmd + if args.subcmd in ("worker"): + # If Worker, then add brick path also to label + label = "%s %s" % (args.subcmd, args.local_path) + elif args.subcmd == "slave": + # If Slave add Master node and Brick details + label = "%s %s%s" % (args.subcmd, args.master_node, args.master_brick) + + # Setup Logger + # Default log file + log_file = gconf.get("cli-log-file") + log_level = gconf.get("cli-log-level") + if getattr(args, "master", None) is not None and \ + getattr(args, "slave", None) is not None: + log_file = gconf.get("log-file") + log_level = gconf.get("log-level") + + # Use different log file location for Slave log file + if args.subcmd == "slave": + log_file = gconf.get("slave-log-file") + log_level = gconf.get("slave-log-level") + + if args.debug: + log_file = "-" + log_level = "DEBUG" + + # Create Logdir if not exists + try: + if log_file != "-": + os.mkdir(os.path.dirname(log_file)) + except OSError as e: + if e.errno != EEXIST: + raise + + setup_logging( + log_file=log_file, + level=log_level, + label=label + ) + + if config_file_error_msg is not None: + logging.warn(config_file_error_msg) + + # Log message for loaded config file + if config_file is not None: + logging.debug(lf("Using session config file", path=config_file)) + set_term_handler() - GLogger.setup() excont = FreeObject(exval=0) + + # Gets the function name based on the input argument. For example + # if subcommand passed as argument is monitor then it looks for + # function with name "subcmd_monitor" in subcmds file + func = getattr(subcmds, "subcmd_" + args.subcmd.replace("-", "_"), None) + try: try: - main_i() + if func is not None: + rconf.args = args + func(args) except: log_raise_exception(excont) finally: finalize(exval=excont.exval) -def main_i(): - """internal main routine - - parse command line, decide what action will be taken; - we can either: - - query/manipulate configuration - - format gsyncd urls using gsyncd's url parsing engine - - start service in following modes, in given stages: - - agent: startup(), ChangelogAgent() - - monitor: startup(), monitor() - - master: startup(), connect_remote(), connect(), service_loop() - - slave: startup(), connect(), service_loop() - """ - rconf = {'go_daemon': 'should'} - - def store_abs(opt, optstr, val, parser): - if val and val != '-': - val = os.path.abspath(val) - setattr(parser.values, opt.dest, val) - - def store_local(opt, optstr, val, parser): - rconf[opt.dest] = val - - def store_local_curry(val): - return lambda o, oo, vx, p: store_local(o, oo, val, p) - - def store_local_obj(op, dmake): - return lambda o, oo, vx, p: store_local( - o, oo, FreeObject(op=op, **dmake(vx)), p) - - op = OptionParser( - usage="%prog [options...] <master> <slave>", version="%prog 0.0.1") - op.add_option('--gluster-command-dir', metavar='DIR', default='') - op.add_option('--gluster-log-file', metavar='LOGF', - default=os.devnull, type=str, action='callback', - callback=store_abs) - op.add_option('--gluster-log-level', metavar='LVL') - op.add_option('--gluster-params', metavar='PRMS', default='') - op.add_option( - '--glusterd-uuid', metavar='UUID', type=str, default='', - help=SUPPRESS_HELP) - op.add_option( - '--gluster-cli-options', metavar='OPTS', default='--log-file=-') - op.add_option('--mountbroker', metavar='LABEL') - op.add_option('-p', '--pid-file', metavar='PIDF', type=str, - action='callback', callback=store_abs) - op.add_option('-l', '--log-file', metavar='LOGF', type=str, - action='callback', callback=store_abs) - op.add_option('--iprefix', metavar='LOGD', type=str, - action='callback', callback=store_abs) - op.add_option('--changelog-log-file', metavar='LOGF', type=str, - action='callback', callback=store_abs) - op.add_option('--log-file-mbr', metavar='LOGF', type=str, - action='callback', callback=store_abs) - op.add_option('--state-file', metavar='STATF', type=str, - action='callback', callback=store_abs) - op.add_option('--state-detail-file', metavar='STATF', - type=str, action='callback', callback=store_abs) - op.add_option('--georep-session-working-dir', metavar='STATF', - type=str, action='callback', callback=store_abs) - op.add_option('--ignore-deletes', default=False, action='store_true') - op.add_option('--isolated-slave', default=False, action='store_true') - op.add_option('--use-rsync-xattrs', default=False, action='store_true') - op.add_option('--pause-on-start', default=False, action='store_true') - op.add_option('-L', '--log-level', metavar='LVL') - op.add_option('-r', '--remote-gsyncd', metavar='CMD', - default=os.path.abspath(sys.argv[0])) - op.add_option('--volume-id', metavar='UUID') - op.add_option('--slave-id', metavar='ID') - op.add_option('--session-owner', metavar='ID') - op.add_option('--local-id', metavar='ID', help=SUPPRESS_HELP, default='') - op.add_option( - '--local-path', metavar='PATH', help=SUPPRESS_HELP, default='') - op.add_option('-s', '--ssh-command', metavar='CMD', default='ssh') - op.add_option('--ssh-command-tar', metavar='CMD', default='ssh') - op.add_option('--rsync-command', metavar='CMD', default='rsync') - op.add_option('--rsync-options', metavar='OPTS', default='') - op.add_option('--rsync-ssh-options', metavar='OPTS', default='--compress') - op.add_option('--timeout', metavar='SEC', type=int, default=120) - op.add_option('--connection-timeout', metavar='SEC', - type=int, default=60, help=SUPPRESS_HELP) - op.add_option('--sync-jobs', metavar='N', type=int, default=3) - op.add_option('--replica-failover-interval', metavar='N', - type=int, default=1) - op.add_option( - '--turns', metavar='N', type=int, default=0, help=SUPPRESS_HELP) - op.add_option('--allow-network', metavar='IPS', default='') - op.add_option('--socketdir', metavar='DIR') - op.add_option('--state-socket-unencoded', metavar='SOCKF', - type=str, action='callback', callback=store_abs) - op.add_option('--checkpoint', metavar='LABEL', default='') - - # tunables for failover/failback mechanism: - # None - gsyncd behaves as normal - # blind - gsyncd works with xtime pairs to identify - # candidates for synchronization - # wrapup - same as normal mode but does not assign - # xtimes to orphaned files - # see crawl() for usage of the above tunables - op.add_option('--special-sync-mode', type=str, help=SUPPRESS_HELP) - - # changelog or xtime? (TODO: Change the default) - op.add_option( - '--change-detector', metavar='MODE', type=str, default='xtime') - # sleep interval for change detection (xtime crawl uses a hardcoded 1 - # second sleep time) - op.add_option('--change-interval', metavar='SEC', type=int, default=3) - # working directory for changelog based mechanism - op.add_option('--working-dir', metavar='DIR', type=str, - action='callback', callback=store_abs) - op.add_option('--use-tarssh', default=False, action='store_true') - - op.add_option('-c', '--config-file', metavar='CONF', - type=str, action='callback', callback=store_local) - # duh. need to specify dest or value will be mapped to None :S - op.add_option('--monitor', dest='monitor', action='callback', - callback=store_local_curry(True)) - op.add_option('--agent', dest='agent', action='callback', - callback=store_local_curry(True)) - op.add_option('--resource-local', dest='resource_local', - type=str, action='callback', callback=store_local) - op.add_option('--resource-remote', dest='resource_remote', - type=str, action='callback', callback=store_local) - op.add_option('--feedback-fd', dest='feedback_fd', type=int, - help=SUPPRESS_HELP, action='callback', callback=store_local) - op.add_option('--rpc-fd', dest='rpc_fd', type=str, help=SUPPRESS_HELP) - op.add_option('--listen', dest='listen', help=SUPPRESS_HELP, - action='callback', callback=store_local_curry(True)) - op.add_option('-N', '--no-daemon', dest="go_daemon", - action='callback', callback=store_local_curry('dont')) - op.add_option('--verify', type=str, dest="verify", - action='callback', callback=store_local) - op.add_option('--create', type=str, dest="create", - action='callback', callback=store_local) - op.add_option('--delete', dest='delete', action='callback', - callback=store_local_curry(True)) - op.add_option('--debug', dest="go_daemon", action='callback', - callback=lambda *a: (store_local_curry('dont')(*a), - setattr( - a[-1].values, 'log_file', '-'), - setattr(a[-1].values, 'log_level', - 'DEBUG'), - setattr(a[-1].values, - 'changelog_log_file', '-'))) - op.add_option('--path', type=str, action='append') - - for a in ('check', 'get'): - op.add_option('--config-' + a, metavar='OPT', type=str, dest='config', - action='callback', - callback=store_local_obj(a, lambda vx: {'opt': vx})) - op.add_option('--config-get-all', dest='config', action='callback', - callback=store_local_obj('get', lambda vx: {'opt': None})) - for m in ('', '-rx', '-glob'): - # call this code 'Pythonic' eh? - # have to define a one-shot local function to be able - # to inject (a value depending on the) - # iteration variable into the inner lambda - def conf_mod_opt_regex_variant(rx): - op.add_option('--config-set' + m, metavar='OPT VAL', type=str, - nargs=2, dest='config', action='callback', - callback=store_local_obj('set', lambda vx: { - 'opt': vx[0], 'val': vx[1], 'rx': rx})) - op.add_option('--config-del' + m, metavar='OPT', type=str, - dest='config', action='callback', - callback=store_local_obj('del', lambda vx: { - 'opt': vx, 'rx': rx})) - conf_mod_opt_regex_variant(m and m[1:] or False) - - op.add_option('--normalize-url', dest='url_print', - action='callback', callback=store_local_curry('normal')) - op.add_option('--canonicalize-url', dest='url_print', - action='callback', callback=store_local_curry('canon')) - op.add_option('--canonicalize-escape-url', dest='url_print', - action='callback', callback=store_local_curry('canon_esc')) - - tunables = [norm(o.get_opt_string()[2:]) - for o in op.option_list - if (o.callback in (store_abs, 'store_true', None) and - o.get_opt_string() not in ('--version', '--help'))] - remote_tunables = ['listen', 'go_daemon', 'timeout', - 'session_owner', 'config_file', 'use_rsync_xattrs'] - rq_remote_tunables = {'listen': True} - - # precedence for sources of values: 1) commandline, 2) cfg file, 3) - # defaults for this to work out we need to tell apart defaults from - # explicitly set options... so churn out the defaults here and call - # the parser with virgin values container. - defaults = op.get_default_values() - opts, args = op.parse_args(values=optparse.Values()) - args_orig = args[:] - r = rconf.get('resource_local') - if r: - if len(args) == 0: - args.append(None) - args[0] = r - r = rconf.get('resource_remote') - if r: - if len(args) == 0: - raise GsyncdError('local resource unspecfied') - elif len(args) == 1: - args.append(None) - args[1] = r - confdata = rconf.get('config') - if not (len(args) == 2 or - (len(args) == 1 and rconf.get('listen')) or - (len(args) <= 2 and confdata) or - rconf.get('url_print')): - sys.stderr.write("error: incorrect number of arguments\n\n") - sys.stderr.write(op.get_usage() + "\n") - sys.exit(1) - - verify = rconf.get('verify') - if verify: - logging.info(verify) - logging.info("Able to spawn gsyncd.py") - return - - restricted = os.getenv('_GSYNCD_RESTRICTED_') - - if restricted: - allopts = {} - allopts.update(opts.__dict__) - allopts.update(rconf) - bannedtuns = set(allopts.keys()) - set(remote_tunables) - if bannedtuns: - raise GsyncdError('following tunables cannot be set with ' - 'restricted SSH invocaton: ' + - ', '.join(bannedtuns)) - for k, v in rq_remote_tunables.items(): - if not k in allopts or allopts[k] != v: - raise GsyncdError('tunable %s is not set to value %s required ' - 'for restricted SSH invocaton' % - (k, v)) - - confrx = getattr(confdata, 'rx', None) - - def makersc(aa, check=True): - if not aa: - return ([], None, None) - ra = [resource.parse_url(u) for u in aa] - local = ra[0] - remote = None - if len(ra) > 1: - remote = ra[1] - if check and not local.can_connect_to(remote): - raise GsyncdError("%s cannot work with %s" % - (local.path, remote and remote.path)) - return (ra, local, remote) - if confrx: - # peers are regexen, don't try to parse them - if confrx == 'glob': - args = ['\A' + fnmatch.translate(a) for a in args] - canon_peers = args - namedict = {} - else: - dc = rconf.get('url_print') - rscs, local, remote = makersc(args_orig, not dc) - if dc: - for r in rscs: - print(r.get_url(**{'normal': {}, - 'canon': {'canonical': True}, - 'canon_esc': {'canonical': True, - 'escaped': True}}[dc])) - return - pa = ([], [], []) - urlprms = ( - {}, {'canonical': True}, {'canonical': True, 'escaped': True}) - for x in rscs: - for i in range(len(pa)): - pa[i].append(x.get_url(**urlprms[i])) - _, canon_peers, canon_esc_peers = pa - # creating the namedict, a dict representing various ways of referring - # to / repreenting peers to be fillable in config templates - mods = (lambda x: x, lambda x: x[ - 0].upper() + x[1:], lambda x: 'e' + x[0].upper() + x[1:]) - if remote: - rmap = {local: ('local', 'master'), remote: ('remote', 'slave')} - else: - rmap = {local: ('local', 'slave')} - namedict = {} - for i in range(len(rscs)): - x = rscs[i] - for name in rmap[x]: - for j in range(3): - namedict[mods[j](name)] = pa[j][i] - namedict[name + 'vol'] = x.volume - if name == 'remote': - namedict['remotehost'] = x.remotehost - if not 'config_file' in rconf: - rconf['config_file'] = os.path.join( - os.path.dirname(sys.argv[0]), "conf/gsyncd_template.conf") - - upgrade_config_file(rconf['config_file']) - gcnf = GConffile( - rconf['config_file'], canon_peers, - defaults.__dict__, opts.__dict__, namedict) - - checkpoint_change = False - if confdata: - opt_ok = norm(confdata.opt) in tunables + [None] - if confdata.op == 'check': - if opt_ok: - sys.exit(0) - else: - sys.exit(1) - elif not opt_ok: - raise GsyncdError("not a valid option: " + confdata.opt) - if confdata.op == 'get': - gcnf.get(confdata.opt) - elif confdata.op == 'set': - gcnf.set(confdata.opt, confdata.val, confdata.rx) - elif confdata.op == 'del': - gcnf.delete(confdata.opt, confdata.rx) - # when modifying checkpoint, it's important to make a log - # of that, so in that case we go on to set up logging even - # if its just config invocation - if confdata.opt == 'checkpoint' and confdata.op in ('set', 'del') and \ - not confdata.rx: - checkpoint_change = True - if not checkpoint_change: - return - - gconf.__dict__.update(defaults.__dict__) - gcnf.update_to(gconf.__dict__) - gconf.__dict__.update(opts.__dict__) - gconf.configinterface = gcnf - - delete = rconf.get('delete') - if delete: - logging.info('geo-replication delete') - # Delete pid file, status file, socket file - cleanup_paths = [] - if getattr(gconf, 'pid_file', None): - cleanup_paths.append(gconf.pid_file) - - if getattr(gconf, 'state_file', None): - cleanup_paths.append(gconf.state_file) - - if getattr(gconf, 'state_detail_file', None): - cleanup_paths.append(gconf.state_detail_file) - - if getattr(gconf, 'state_socket_unencoded', None): - cleanup_paths.append(gconf.state_socket_unencoded) - - cleanup_paths.append(rconf['config_file'][:-11] + "*") - - # Cleanup changelog working dirs - if getattr(gconf, 'working_dir', None): - try: - shutil.rmtree(gconf.working_dir) - except (IOError, OSError): - if sys.exc_info()[1].errno == ENOENT: - pass - else: - raise GsyncdError( - 'Error while removing working dir: %s' % - gconf.working_dir) - - for path in cleanup_paths: - # To delete temp files - for f in glob.glob(path + "*"): - _unlink(f) - return - - if restricted and gconf.allow_network: - ssh_conn = os.getenv('SSH_CONNECTION') - if not ssh_conn: - # legacy env var - ssh_conn = os.getenv('SSH_CLIENT') - if ssh_conn: - allowed_networks = [IPNetwork(a) - for a in gconf.allow_network.split(',')] - client_ip = IPAddress(ssh_conn.split()[0]) - allowed = False - for nw in allowed_networks: - if client_ip in nw: - allowed = True - break - if not allowed: - raise GsyncdError("client IP address is not allowed") - - ffd = rconf.get('feedback_fd') - if ffd: - fcntl.fcntl(ffd, fcntl.F_SETFD, fcntl.FD_CLOEXEC) - - # normalize loglevel - lvl0 = gconf.log_level - if isinstance(lvl0, str): - lvl1 = lvl0.upper() - lvl2 = logging.getLevelName(lvl1) - # I have _never_ _ever_ seen such an utterly braindead - # error condition - if lvl2 == "Level " + lvl1: - raise GsyncdError('cannot recognize log level "%s"' % lvl0) - gconf.log_level = lvl2 - - if not privileged() and gconf.log_file_mbr: - gconf.log_file = gconf.log_file_mbr - - if checkpoint_change: - try: - GLogger._gsyncd_loginit(log_file=gconf.log_file, label='conf') - if confdata.op == 'set': - logging.info('checkpoint %s set' % confdata.val) - gcnf.delete('checkpoint_completed') - gcnf.delete('checkpoint_target') - elif confdata.op == 'del': - logging.info('checkpoint info was reset') - # if it is removing 'checkpoint' then we need - # to remove 'checkpoint_completed' and 'checkpoint_target' too - gcnf.delete('checkpoint_completed') - gcnf.delete('checkpoint_target') - - except IOError: - if sys.exc_info()[1].errno == ENOENT: - # directory of log path is not present, - # which happens if we get here from - # a peer-multiplexed "config-set checkpoint" - # (as that directory is created only on the - # original node) - pass - else: - raise - return - - create = rconf.get('create') - if create: - if getattr(gconf, 'state_file', None): - update_file(gconf.state_file, lambda f: f.write(create + '\n')) - return - - go_daemon = rconf['go_daemon'] - be_monitor = rconf.get('monitor') - be_agent = rconf.get('agent') - - rscs, local, remote = makersc(args) - if not be_monitor and isinstance(remote, resource.SSH) and \ - go_daemon == 'should': - go_daemon = 'postconn' - log_file = None - else: - log_file = gconf.log_file - if be_monitor: - label = 'monitor' - elif be_agent: - label = 'agent' - elif remote: - # master - label = gconf.local_path - else: - label = 'slave' - startup(go_daemon=go_daemon, log_file=log_file, label=label) - resource.Popen.init_errhandler() - - if be_agent: - os.setsid() - logging.debug('rpc_fd: %s' % repr(gconf.rpc_fd)) - return agent(Changelog(), gconf.rpc_fd) - - if be_monitor: - return monitor(*rscs) - - logging.info("syncing: %s" % " -> ".join(r.url for r in rscs)) - if remote: - go_daemon = remote.connect_remote(go_daemon=go_daemon) - if go_daemon: - startup(go_daemon=go_daemon, log_file=gconf.log_file) - # complete remote connection in child - remote.connect_remote(go_daemon='done') - local.connect() - if ffd: - os.close(ffd) - local.service_loop(*[r for r in [remote] if r]) - - if __name__ == "__main__": main() diff --git a/geo-replication/syncdaemon/gsyncdconfig.py b/geo-replication/syncdaemon/gsyncdconfig.py new file mode 100644 index 00000000000..8848071997a --- /dev/null +++ b/geo-replication/syncdaemon/gsyncdconfig.py @@ -0,0 +1,485 @@ +# -*- coding: utf-8 -*- +# +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. +# +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +try: + from ConfigParser import RawConfigParser, NoSectionError +except ImportError: + from configparser import RawConfigParser, NoSectionError +import os +import shutil +from string import Template +from datetime import datetime +from threading import Lock + + +# Global object which can be used in other modules +# once load_config is called +_gconf = {} + + +class GconfNotConfigurable(Exception): + pass + + +class GconfInvalidValue(Exception): + pass + + +class Gconf(object): + def __init__(self, default_conf_file, custom_conf_file=None, + args={}, extra_tmpl_args={}, override_from_args=False): + self.lock = Lock() + self.default_conf_file = default_conf_file + self.custom_conf_file = custom_conf_file + self.tmp_conf_file = None + self.gconf = {} + self.gconfdata = {} + self.gconf_typecast = {} + self.template_conf = [] + self.non_configurable_configs = [] + self.prev_mtime = 0 + if custom_conf_file is not None: + self.tmp_conf_file = custom_conf_file + ".tmp" + + self.session_conf_items = [] + self.args = args + self.extra_tmpl_args = extra_tmpl_args + self.override_from_args = override_from_args + # Store default values only if overwritten, Only for JSON/CLI output + self.default_values = {} + self._load() + + def _tmpl_substitute(self): + tmpl_values = {} + for k, v in self.gconf.items(): + tmpl_values[k.replace("-", "_")] = v + + # override the config file values with the one user passed + for k, v in self.args.items(): + # override the existing value only if set by user + if v is not None: + tmpl_values[k] = v + + for k, v in self.extra_tmpl_args.items(): + tmpl_values[k] = v + + for k, v in self.gconf.items(): + if k in self.template_conf and \ + (isinstance(v, str) or isinstance(v, unicode)): + self.gconf[k] = Template(v).safe_substitute(tmpl_values) + + def _do_typecast(self): + for k, v in self.gconf.items(): + cast_func = globals().get( + "to_" + self.gconf_typecast.get(k, "string"), None) + if cast_func is not None: + self.gconf[k] = cast_func(v) + if self.default_values.get(k, None) is not None: + self.default_values[k] = cast_func(v) + + def reset(self, name): + # If custom conf file is not set then it is only read only configs + if self.custom_conf_file is None: + raise GconfNotConfigurable() + + # If a config can not be modified + if name != "all" and not self._is_configurable(name): + raise GconfNotConfigurable() + + cnf = RawConfigParser() + with open(self.custom_conf_file) as f: + cnf.readfp(f) + + # Nothing to Reset, Not configured + if name != "all": + if not cnf.has_option("vars", name): + return True + + # Remove option from custom conf file + cnf.remove_option("vars", name) + else: + # Remove and add empty section, do not disturb if config file + # already has any other section + try: + cnf.remove_section("vars") + except NoSectionError: + pass + + cnf.add_section("vars") + + with open(self.tmp_conf_file, "w") as fw: + cnf.write(fw) + + os.rename(self.tmp_conf_file, self.custom_conf_file) + + self.reload() + + return True + + def set(self, name, value): + if self.custom_conf_file is None: + raise GconfNotConfigurable() + + if not self._is_configurable(name): + raise GconfNotConfigurable() + + if not self._is_valid_value(name, value): + raise GconfInvalidValue() + + curr_val = self.gconf.get(name, None) + if curr_val == value: + return True + + cnf = RawConfigParser() + with open(self.custom_conf_file) as f: + cnf.readfp(f) + + if not cnf.has_section("vars"): + cnf.add_section("vars") + + cnf.set("vars", name, value) + with open(self.tmp_conf_file, "w") as fw: + cnf.write(fw) + + os.rename(self.tmp_conf_file, self.custom_conf_file) + + self.reload() + + return True + + def check(self, name, value=None, with_conffile=True): + if with_conffile and self.custom_conf_file is None: + raise GconfNotConfigurable() + + if not self._is_configurable(name): + raise GconfNotConfigurable() + + if value is not None and not self._is_valid_value(name, value): + raise GconfInvalidValue() + + + def _load_with_lock(self): + with self.lock: + self._load() + + def _load(self): + self.gconf = {} + self.template_conf = [] + self.gconf_typecast = {} + self.non_configurable_configs = [] + self.session_conf_items = [] + self.default_values = {} + + conf = RawConfigParser() + # Default Template config file + with open(self.default_conf_file) as f: + conf.readfp(f) + + # Custom Config file + if self.custom_conf_file is not None: + with open(self.custom_conf_file) as f: + conf.readfp(f) + + # Get version from default conf file + self.version = conf.get("__meta__", "version") + + # Populate default values + for sect in conf.sections(): + if sect in ["__meta__", "vars"]: + continue + + # Collect list of available options with help details + self.gconfdata[sect] = {} + for k, v in conf.items(sect): + self.gconfdata[sect][k] = v.strip() + + # Collect the Type cast information + if conf.has_option(sect, "type"): + self.gconf_typecast[sect] = conf.get(sect, "type") + + # Prepare list of configurable conf + if conf.has_option(sect, "configurable"): + if conf.get(sect, "configurable").lower() == "false": + self.non_configurable_configs.append(sect) + + # if it is a template conf value which needs to be substituted + if conf.has_option(sect, "template"): + if conf.get(sect, "template").lower().strip() == "true": + self.template_conf.append(sect) + + # Set default values + if conf.has_option(sect, "value"): + self.gconf[sect] = conf.get(sect, "value").strip() + + # Load the custom conf elements and overwrite + if conf.has_section("vars"): + for k, v in conf.items("vars"): + self.session_conf_items.append(k) + self.default_values[k] = self.gconf.get(k, "") + self.gconf[k] = v.strip() + + # Overwrite the Slave configurations which are sent as + # arguments to gsyncd slave + if self.override_from_args: + for k, v in self.args.items(): + k = k.replace("_", "-") + if k.startswith("slave-") and k in self.gconf: + self.gconf[k] = v + + self._tmpl_substitute() + self._do_typecast() + + def reload(self, with_lock=True): + if self._is_config_changed(): + if with_lock: + self._load_with_lock() + else: + self._load() + + def get(self, name, default_value=None, with_lock=True): + if with_lock: + with self.lock: + return self.gconf.get(name, default_value) + else: + return self.gconf.get(name, default_value) + + def getall(self, show_defaults=False, show_non_configurable=False): + cnf = {} + if not show_defaults: + for k in self.session_conf_items: + if k not in self.non_configurable_configs: + dv = self.default_values.get(k, "") + cnf[k] = { + "value": self.get(k), + "default": dv, + "configurable": True, + "modified": False if dv == "" else True + } + return cnf + + # Show all configs including defaults + for k, v in self.gconf.items(): + configurable = False if k in self.non_configurable_configs \ + else True + dv = self.default_values.get(k, "") + modified = False if dv == "" else True + if show_non_configurable: + cnf[k] = { + "value": v, + "default": dv, + "configurable": configurable, + "modified": modified + } + else: + if k not in self.non_configurable_configs: + cnf[k] = { + "value": v, + "default": dv, + "configurable": configurable, + "modified": modified + } + + return cnf + + def getr(self, name, default_value=None): + with self.lock: + self.reload(with_lock=False) + return self.get(name, default_value, with_lock=False) + + def get_help(self, name=None): + pass + + def _is_configurable(self, name): + item = self.gconfdata.get(name, None) + if item is None: + return False + + return item.get("configurable", True) + + def _is_valid_value(self, name, value): + item = self.gconfdata.get(name, None) + if item is None: + return False + + # If validation func not defined + if item.get("validation", None) is None: + return True + + # minmax validation + if item["validation"] == "minmax": + return validate_minmax(value, item["min"], item["max"]) + + if item["validation"] == "choice": + return validate_choice(value, item["allowed_values"]) + + if item["validation"] == "bool": + return validate_bool(value) + + if item["validation"] == "execpath": + return validate_execpath(value) + + if item["validation"] == "unixtime": + return validate_unixtime(value) + + if item["validation"] == "int": + return validate_int(value) + + return False + + def _is_config_changed(self): + if self.custom_conf_file is not None and \ + os.path.exists(self.custom_conf_file): + st = os.lstat(self.custom_conf_file) + if st.st_mtime > self.prev_mtime: + self.prev_mtime = st.st_mtime + return True + + return False + +def is_config_file_old(config_file, mastervol, slavevol): + cnf = RawConfigParser() + cnf.read(config_file) + session_section = "peers %s %s" % (mastervol, slavevol) + try: + return dict(cnf.items(session_section)) + except NoSectionError: + return None + +def config_upgrade(config_file, ret): + config_file_backup = os.path.join(os.path.dirname(config_file), "gsyncd.conf.bkp") + + #copy old config file in a backup file + shutil.copyfile(config_file, config_file_backup) + + #write a new config file + config = RawConfigParser() + config.add_section('vars') + + for key, value in ret.items(): + #handle option name changes + if key == "use_tarssh": + new_key = "sync-method" + if value == "true": + new_value = "tarssh" + else: + new_value = "rsync" + config.set('vars', new_key, new_value) + elif key == "timeout": + new_key = "slave-timeout" + config.set('vars', new_key, value) + #for changes like: ignore_deletes to ignore-deletes + else: + new_key = key.replace("_", "-") + config.set('vars', new_key, value) + + with open(config_file, 'w') as configfile: + config.write(configfile) + + +def validate_int(value): + try: + _ = int(value) + return True + except ValueError: + return False + + +def validate_unixtime(value): + try: + y = datetime.fromtimestamp(int(value)).strftime("%Y") + if y == "1970": + return False + + return True + except ValueError: + return False + + +def validate_minmax(value, minval, maxval): + try: + value = int(value) + minval = int(minval) + maxval = int(maxval) + return value >= minval and value <= maxval + except ValueError: + return False + + +def validate_choice(value, allowed_values): + allowed_values = allowed_values.split(",") + allowed_values = [v.strip() for v in allowed_values] + + return value in allowed_values + + +def validate_bool(value): + return value in ["true", "false"] + + +def validate_execpath(value): + return os.path.isfile(value) and os.access(value, os.X_OK) + + +def validate_filepath(value): + return os.path.isfile(value) + + +def validate_path(value): + return os.path.exists(value) + + +def to_int(value): + return int(value) + + +def to_float(value): + return float(value) + + +def to_bool(value): + if isinstance(value, bool): + return value + return True if value in ["true", "True"] else False + + +def get(name, default_value=None): + return _gconf.get(name, default_value) + + +def getall(show_defaults=False, show_non_configurable=False): + return _gconf.getall(show_defaults=show_defaults, + show_non_configurable=show_non_configurable) + + +def getr(name, default_value=None): + return _gconf.getr(name, default_value) + + +def load(default_conf, custom_conf=None, args={}, extra_tmpl_args={}, + override_from_args=False): + global _gconf + _gconf = Gconf(default_conf, custom_conf, args, extra_tmpl_args, + override_from_args) + + +def setconfig(name, value): + global _gconf + _gconf.set(name, value) + + +def resetconfig(name): + global _gconf + _gconf.reset(name) + + +def check(name, value=None, with_conffile=True): + global _gconf + _gconf.check(name, value=value, with_conffile=with_conffile) diff --git a/geo-replication/syncdaemon/gsyncdstatus.py b/geo-replication/syncdaemon/gsyncdstatus.py new file mode 100644 index 00000000000..1a655ff8887 --- /dev/null +++ b/geo-replication/syncdaemon/gsyncdstatus.py @@ -0,0 +1,419 @@ +#!/usr/bin/python3 +# +# Copyright (c) 2011-2014 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. + +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +from __future__ import print_function +import fcntl +import os +import tempfile +try: + import urllib.parse as urllib +except ImportError: + import urllib +import json +import time +from datetime import datetime +from errno import EACCES, EAGAIN, ENOENT +import logging + +from syncdutils import (EVENT_GEOREP_ACTIVE, EVENT_GEOREP_PASSIVE, gf_event, + EVENT_GEOREP_CHECKPOINT_COMPLETED, lf) + +DEFAULT_STATUS = "N/A" +MONITOR_STATUS = ("Created", "Started", "Paused", "Stopped") +STATUS_VALUES = (DEFAULT_STATUS, + "Initializing...", + "Active", + "Passive", + "Faulty") + +CRAWL_STATUS_VALUES = (DEFAULT_STATUS, + "Hybrid Crawl", + "History Crawl", + "Changelog Crawl") + + +def human_time(ts): + try: + return datetime.fromtimestamp(float(ts)).strftime("%Y-%m-%d %H:%M:%S") + except ValueError: + return DEFAULT_STATUS + + +def human_time_utc(ts): + try: + return datetime.utcfromtimestamp( + float(ts)).strftime("%Y-%m-%d %H:%M:%S") + except ValueError: + return DEFAULT_STATUS + + +def get_default_values(): + return { + "slave_node": DEFAULT_STATUS, + "worker_status": DEFAULT_STATUS, + "last_synced": 0, + "last_synced_entry": 0, + "crawl_status": DEFAULT_STATUS, + "entry": 0, + "data": 0, + "meta": 0, + "failures": 0, + "checkpoint_completed": DEFAULT_STATUS, + "checkpoint_time": 0, + "checkpoint_completion_time": 0} + + +class LockedOpen(object): + + def __init__(self, filename, *args, **kwargs): + self.filename = filename + self.open_args = args + self.open_kwargs = kwargs + self.fileobj = None + + def __enter__(self): + """ + If two processes compete to update a file, The first process + gets the lock and the second process is blocked in the fcntl.flock() + call. When first process replaces the file and releases the lock, + the already open file descriptor in the second process now points + to a "ghost" file(not reachable by any path name) with old contents. + To avoid that conflict, check the fd already opened is same or + not. Open new one if not same + """ + f = open(self.filename, *self.open_args, **self.open_kwargs) + while True: + fcntl.flock(f, fcntl.LOCK_EX) + fnew = open(self.filename, *self.open_args, **self.open_kwargs) + if os.path.sameopenfile(f.fileno(), fnew.fileno()): + fnew.close() + break + else: + f.close() + f = fnew + self.fileobj = f + return f + + def __exit__(self, _exc_type, _exc_value, _traceback): + fcntl.flock(self.fileobj, fcntl.LOCK_UN) + self.fileobj.close() + + +def set_monitor_status(status_file, status): + fd = os.open(status_file, os.O_CREAT | os.O_RDWR) + os.close(fd) + with LockedOpen(status_file, 'r+'): + with tempfile.NamedTemporaryFile('w', dir=os.path.dirname(status_file), + delete=False) as tf: + tf.write(status) + tempname = tf.name + + os.rename(tempname, status_file) + dirfd = os.open(os.path.dirname(os.path.abspath(status_file)), + os.O_DIRECTORY) + os.fsync(dirfd) + os.close(dirfd) + + +class GeorepStatus(object): + def __init__(self, monitor_status_file, master_node, brick, master_node_id, + master, slave, monitor_pid_file=None): + self.master = master + slv_data = slave.split("::") + self.slave_host = slv_data[0] + self.slave_volume = slv_data[1].split(":")[0] # Remove Slave UUID + self.work_dir = os.path.dirname(monitor_status_file) + self.monitor_status_file = monitor_status_file + self.filename = os.path.join(self.work_dir, + "brick_%s.status" + % urllib.quote_plus(brick)) + + fd = os.open(self.filename, os.O_CREAT | os.O_RDWR) + os.close(fd) + fd = os.open(self.monitor_status_file, os.O_CREAT | os.O_RDWR) + os.close(fd) + self.master_node = master_node + self.master_node_id = master_node_id + self.brick = brick + self.default_values = get_default_values() + self.monitor_pid_file = monitor_pid_file + + def send_event(self, event_type, **kwargs): + gf_event(event_type, + master_volume=self.master, + master_node=self.master_node, + master_node_id=self.master_node_id, + slave_host=self.slave_host, + slave_volume=self.slave_volume, + brick_path=self.brick, + **kwargs) + + def _update(self, mergerfunc): + data = self.default_values + with LockedOpen(self.filename, 'r+') as f: + try: + data.update(json.load(f)) + except ValueError: + pass + + data = mergerfunc(data) + # If Data is not changed by merger func + if not data: + return False + + with tempfile.NamedTemporaryFile( + 'w', + dir=os.path.dirname(self.filename), + delete=False) as tf: + tf.write(data) + tempname = tf.name + + os.rename(tempname, self.filename) + dirfd = os.open(os.path.dirname(os.path.abspath(self.filename)), + os.O_DIRECTORY) + os.fsync(dirfd) + os.close(dirfd) + return True + + def reset_on_worker_start(self): + def merger(data): + data["slave_node"] = DEFAULT_STATUS + data["crawl_status"] = DEFAULT_STATUS + data["entry"] = 0 + data["data"] = 0 + data["meta"] = 0 + return json.dumps(data) + + self._update(merger) + + def set_field(self, key, value): + def merger(data): + # Current data and prev data is same + if data[key] == value: + return {} + + data[key] = value + return json.dumps(data) + + return self._update(merger) + + def trigger_gf_event_checkpoint_completion(self, checkpoint_time, + checkpoint_completion_time): + self.send_event(EVENT_GEOREP_CHECKPOINT_COMPLETED, + checkpoint_time=checkpoint_time, + checkpoint_completion_time=checkpoint_completion_time) + + def set_last_synced(self, value, checkpoint_time): + def merger(data): + data["last_synced"] = value[0] + + # If checkpoint is not set or reset + # or if last set checkpoint is changed + if checkpoint_time == 0 or \ + checkpoint_time != data["checkpoint_time"]: + data["checkpoint_time"] = 0 + data["checkpoint_completion_time"] = 0 + data["checkpoint_completed"] = "No" + + # If checkpoint is completed and not marked as completed + # previously then update the checkpoint completed time + if checkpoint_time > 0 and checkpoint_time <= value[0]: + if data["checkpoint_completed"] == "No": + curr_time = int(time.time()) + data["checkpoint_time"] = checkpoint_time + data["checkpoint_completion_time"] = curr_time + data["checkpoint_completed"] = "Yes" + logging.info(lf("Checkpoint completed", + checkpoint_time=human_time_utc( + checkpoint_time), + completion_time=human_time_utc(curr_time))) + self.trigger_gf_event_checkpoint_completion( + checkpoint_time, curr_time) + + return json.dumps(data) + + self._update(merger) + + def set_worker_status(self, status): + if self.set_field("worker_status", status): + logging.info(lf("Worker Status Change", + status=status)) + + def set_worker_crawl_status(self, status): + if self.set_field("crawl_status", status): + logging.info(lf("Crawl Status Change", + status=status)) + + def set_slave_node(self, slave_node): + def merger(data): + data["slave_node"] = slave_node + return json.dumps(data) + + self._update(merger) + + def inc_value(self, key, value): + def merger(data): + data[key] = data.get(key, 0) + value + return json.dumps(data) + + self._update(merger) + + def dec_value(self, key, value): + def merger(data): + data[key] = data.get(key, 0) - value + if data[key] < 0: + data[key] = 0 + return json.dumps(data) + + self._update(merger) + + def set_active(self): + if self.set_field("worker_status", "Active"): + logging.info(lf("Worker Status Change", + status="Active")) + self.send_event(EVENT_GEOREP_ACTIVE) + + def set_passive(self): + if self.set_field("worker_status", "Passive"): + logging.info(lf("Worker Status Change", + status="Passive")) + self.send_event(EVENT_GEOREP_PASSIVE) + + def get_monitor_status(self): + data = "" + with open(self.monitor_status_file, "r") as f: + data = f.read().strip() + return data + + def get_status(self, checkpoint_time=0): + """ + Monitor Status ---> Created Started Paused Stopped + ---------------------------------------------------------------------- + slave_node N/A VALUE VALUE N/A + status Created VALUE Paused Stopped + last_synced N/A VALUE VALUE VALUE + last_synced_entry N/A VALUE VALUE VALUE + crawl_status N/A VALUE N/A N/A + entry N/A VALUE N/A N/A + data N/A VALUE N/A N/A + meta N/A VALUE N/A N/A + failures N/A VALUE VALUE VALUE + checkpoint_completed N/A VALUE VALUE VALUE + checkpoint_time N/A VALUE VALUE VALUE + checkpoint_completed_time N/A VALUE VALUE VALUE + """ + data = self.default_values + with open(self.filename) as f: + try: + data.update(json.load(f)) + except ValueError: + pass + monitor_status = self.get_monitor_status() + + # Verifying whether monitor process running and adjusting status + if monitor_status in ["Started", "Paused"]: + try: + with open(self.monitor_pid_file, "r+") as f: + fcntl.lockf(f, fcntl.LOCK_EX | fcntl.LOCK_NB) + monitor_status = "Stopped" + except (IOError, OSError) as e: + # If pid file not exists, either monitor died or Geo-rep + # not even started once + if e.errno == ENOENT: + monitor_status = "Stopped" + elif e.errno in (EACCES, EAGAIN): + # cannot grab. so, monitor process still running..move on + pass + else: + raise + + if monitor_status in ["Created", "Paused", "Stopped"]: + data["worker_status"] = monitor_status + + if monitor_status == "": + data["worker_status"] = "Stopped" + + # Checkpoint adjustments + if checkpoint_time == 0: + data["checkpoint_completed"] = DEFAULT_STATUS + data["checkpoint_time"] = DEFAULT_STATUS + data["checkpoint_completion_time"] = DEFAULT_STATUS + else: + if checkpoint_time != data["checkpoint_time"]: + if checkpoint_time <= data["last_synced"]: + data["checkpoint_completed"] = "Yes" + data["checkpoint_time"] = checkpoint_time + data["checkpoint_completion_time"] = data["last_synced"] + else: + data["checkpoint_completed"] = "No" + data["checkpoint_time"] = checkpoint_time + data["checkpoint_completion_time"] = DEFAULT_STATUS + + if data["checkpoint_time"] not in [0, DEFAULT_STATUS]: + chkpt_time = data["checkpoint_time"] + data["checkpoint_time"] = human_time(chkpt_time) + data["checkpoint_time_utc"] = human_time_utc(chkpt_time) + + if data["checkpoint_completion_time"] not in [0, DEFAULT_STATUS]: + chkpt_completion_time = data["checkpoint_completion_time"] + data["checkpoint_completion_time"] = human_time( + chkpt_completion_time) + data["checkpoint_completion_time_utc"] = human_time_utc( + chkpt_completion_time) + + if data["last_synced"] == 0: + data["last_synced"] = DEFAULT_STATUS + data["last_synced_utc"] = DEFAULT_STATUS + else: + last_synced = data["last_synced"] + data["last_synced"] = human_time(last_synced) + data["last_synced_utc"] = human_time_utc(last_synced) + + if data["worker_status"] != "Active": + data["last_synced"] = DEFAULT_STATUS + data["last_synced_utc"] = DEFAULT_STATUS + data["crawl_status"] = DEFAULT_STATUS + data["entry"] = DEFAULT_STATUS + data["data"] = DEFAULT_STATUS + data["meta"] = DEFAULT_STATUS + data["failures"] = DEFAULT_STATUS + data["checkpoint_completed"] = DEFAULT_STATUS + data["checkpoint_time"] = DEFAULT_STATUS + data["checkpoint_completed_time"] = DEFAULT_STATUS + data["checkpoint_time_utc"] = DEFAULT_STATUS + data["checkpoint_completion_time_utc"] = DEFAULT_STATUS + + if data["worker_status"] not in ["Active", "Passive"]: + data["slave_node"] = DEFAULT_STATUS + + if data.get("last_synced_utc", 0) == 0: + data["last_synced_utc"] = DEFAULT_STATUS + + if data.get("checkpoint_completion_time_utc", 0) == 0: + data["checkpoint_completion_time_utc"] = DEFAULT_STATUS + + if data.get("checkpoint_time_utc", 0) == 0: + data["checkpoint_time_utc"] = DEFAULT_STATUS + + return data + + def print_status(self, checkpoint_time=0, json_output=False): + status_out = self.get_status(checkpoint_time) + if json_output: + out = {} + # Convert all values as string + for k, v in status_out.items(): + out[k] = str(v) + print(json.dumps(out)) + return + + for key, value in status_out.items(): + print(("%s: %s" % (key, value))) diff --git a/geo-replication/syncdaemon/libcxattr.py b/geo-replication/syncdaemon/libcxattr.py index 74d120fa196..e6406c36bd7 100644 --- a/geo-replication/syncdaemon/libcxattr.py +++ b/geo-replication/syncdaemon/libcxattr.py @@ -9,13 +9,14 @@ # import os -from ctypes import CDLL, c_int, create_string_buffer -from ctypes.util import find_library +from ctypes import CDLL, get_errno +from py2py3 import (bytearray_to_str, gr_create_string_buffer, + gr_query_xattr, gr_lsetxattr, gr_lremovexattr) class Xattr(object): - """singleton that wraps the extended attribues system + """singleton that wraps the extended attributes system interface for python using ctypes Just implement it to the degree we need it, in particular @@ -25,11 +26,11 @@ class Xattr(object): sizes we expect """ - libc = CDLL(find_library("libc")) + libc = CDLL("libc.so.6", use_errno=True) @classmethod def geterrno(cls): - return c_int.in_dll(cls.libc, 'errno').value + return get_errno() @classmethod def raise_oserr(cls): @@ -39,20 +40,23 @@ class Xattr(object): @classmethod def _query_xattr(cls, path, siz, syscall, *a): if siz: - buf = create_string_buffer('\0' * siz) + buf = gr_create_string_buffer(siz) else: buf = None ret = getattr(cls.libc, syscall)(*((path,) + a + (buf, siz))) if ret == -1: cls.raise_oserr() if siz: - return buf.raw[:ret] + # py2 and py3 compatibility. Convert bytes array + # to string + result = bytearray_to_str(buf.raw) + return result[:ret] else: return ret @classmethod def lgetxattr(cls, path, attr, siz=0): - return cls._query_xattr(path, siz, 'lgetxattr', attr) + return gr_query_xattr(cls, path, siz, 'lgetxattr', attr) @classmethod def lgetxattr_buf(cls, path, attr): @@ -66,29 +70,43 @@ class Xattr(object): @classmethod def llistxattr(cls, path, siz=0): - ret = cls._query_xattr(path, siz, 'llistxattr') + ret = gr_query_xattr(cls, path, siz, 'llistxattr') if isinstance(ret, str): - ret = ret.split('\0') + ret = ret.strip('\0') + ret = ret.split('\0') if ret else [] return ret @classmethod def lsetxattr(cls, path, attr, val): - ret = cls.libc.lsetxattr(path, attr, val, len(val), 0) + ret = gr_lsetxattr(cls, path, attr, val) if ret == -1: cls.raise_oserr() @classmethod def lremovexattr(cls, path, attr): - ret = cls.libc.lremovexattr(path, attr) + ret = gr_lremovexattr(cls, path, attr) if ret == -1: cls.raise_oserr() @classmethod def llistxattr_buf(cls, path): """listxattr variant with size discovery""" - size = cls.llistxattr(path) - if size == -1: - cls.raise_oserr() - if size == 0: - return [] - return cls.llistxattr(path, size) + try: + # Assuming no more than 100 xattrs in a file/directory and + # each xattr key length will be less than 256 bytes + # llistxattr will be called with bigger size so that + # listxattr will not fail with ERANGE. OSError will be + # raised if fails even with the large size specified. + size = 256 * 100 + return cls.llistxattr(path, size) + except OSError: + # If fixed length failed for getting list of xattrs then + # use the llistxattr call to get the size and use that + # size to get the list of xattrs. + size = cls.llistxattr(path) + if size == -1: + cls.raise_oserr() + if size == 0: + return [] + + return cls.llistxattr(path, size) diff --git a/geo-replication/syncdaemon/libgfchangelog.py b/geo-replication/syncdaemon/libgfchangelog.py index 89ae032016f..a3bda7282c0 100644 --- a/geo-replication/syncdaemon/libgfchangelog.py +++ b/geo-replication/syncdaemon/libgfchangelog.py @@ -9,120 +9,135 @@ # import os -from ctypes import CDLL, create_string_buffer, get_errno, byref, c_ulong +from ctypes import CDLL, RTLD_GLOBAL, get_errno, byref, c_ulong from ctypes.util import find_library -from syncdutils import ChangelogException - - -class Changes(object): - libgfc = CDLL(find_library("gfchangelog"), use_errno=True) - - @classmethod - def geterrno(cls): - return get_errno() - - @classmethod - def raise_changelog_err(cls): - errn = cls.geterrno() - raise ChangelogException(errn, os.strerror(errn)) - - @classmethod - def _get_api(cls, call): - return getattr(cls.libgfc, call) - - @classmethod - def cl_register(cls, brick, path, log_file, log_level, retries=0): - ret = cls._get_api('gf_changelog_register')(brick, path, - log_file, - log_level, retries) - if ret == -1: - cls.raise_changelog_err() - - @classmethod - def cl_scan(cls): - ret = cls._get_api('gf_changelog_scan')() - if ret == -1: - cls.raise_changelog_err() - - @classmethod - def cl_startfresh(cls): - ret = cls._get_api('gf_changelog_start_fresh')() - if ret == -1: - cls.raise_changelog_err() - - @classmethod - def cl_getchanges(cls): - """ remove hardcoding for path name length """ - def clsort(f): - return f.split('.')[-1] - changes = [] - buf = create_string_buffer('\0', 4096) - call = cls._get_api('gf_changelog_next_change') - - while True: - ret = call(buf, 4096) - if ret in (0, -1): - break - changes.append(buf.raw[:ret - 1]) - if ret == -1: - cls.raise_changelog_err() - # cleanup tracker - cls.cl_startfresh() - return sorted(changes, key=clsort) - - @classmethod - def cl_done(cls, clfile): - ret = cls._get_api('gf_changelog_done')(clfile) - if ret == -1: - cls.raise_changelog_err() - - @classmethod - def cl_history_scan(cls): - ret = cls._get_api('gf_history_changelog_scan')() - if ret == -1: - cls.raise_changelog_err() - - return ret - - @classmethod - def cl_history_changelog(cls, changelog_path, start, end, num_parallel): - actual_end = c_ulong() - ret = cls._get_api('gf_history_changelog')(changelog_path, start, end, - num_parallel, - byref(actual_end)) - if ret == -1: - cls.raise_changelog_err() - - return (ret, actual_end.value) - - @classmethod - def cl_history_startfresh(cls): - ret = cls._get_api('gf_history_changelog_start_fresh')() - if ret == -1: - cls.raise_changelog_err() - - @classmethod - def cl_history_getchanges(cls): - """ remove hardcoding for path name length """ - def clsort(f): - return f.split('.')[-1] - - changes = [] - buf = create_string_buffer('\0', 4096) - call = cls._get_api('gf_history_changelog_next_change') - - while True: - ret = call(buf, 4096) - if ret in (0, -1): - break - changes.append(buf.raw[:ret - 1]) - if ret == -1: - cls.raise_changelog_err() - - return sorted(changes, key=clsort) - - @classmethod - def cl_history_done(cls, clfile): - ret = cls._get_api('gf_history_changelog_done')(clfile) - if ret == -1: - cls.raise_changelog_err() +from syncdutils import ChangelogException, ChangelogHistoryNotAvailable +from py2py3 import (gr_cl_history_changelog, gr_cl_done, + gr_create_string_buffer, gr_cl_register, + gr_cl_history_done, bytearray_to_str) + + +libgfc = CDLL( + find_library("gfchangelog"), + mode=RTLD_GLOBAL, + use_errno=True +) + + +def _raise_changelog_err(): + errn = get_errno() + raise ChangelogException(errn, os.strerror(errn)) + + +def _init(): + if libgfc.gf_changelog_init(None) == -1: + _raise_changelog_err() + + +def register(brick, path, log_file, log_level, retries=0): + _init() + + ret = gr_cl_register(libgfc, brick, path, log_file, log_level, retries) + + if ret == -1: + _raise_changelog_err() + + +def scan(): + ret = libgfc.gf_changelog_scan() + if ret == -1: + _raise_changelog_err() + + +def startfresh(): + ret = libgfc.gf_changelog_start_fresh() + if ret == -1: + _raise_changelog_err() + + +def getchanges(): + def clsort(cfile): + return cfile.split('.')[-1] + + changes = [] + buf = gr_create_string_buffer(4096) + call = libgfc.gf_changelog_next_change + + while True: + ret = call(buf, 4096) + if ret in (0, -1): + break + + # py2 and py3 compatibility + result = bytearray_to_str(buf.raw[:ret - 1]) + changes.append(result) + + if ret == -1: + _raise_changelog_err() + + # cleanup tracker + startfresh() + + return sorted(changes, key=clsort) + + +def done(clfile): + ret = gr_cl_done(libgfc, clfile) + if ret == -1: + _raise_changelog_err() + + +def history_scan(): + ret = libgfc.gf_history_changelog_scan() + if ret == -1: + _raise_changelog_err() + + return ret + + +def history_changelog(changelog_path, start, end, num_parallel): + actual_end = c_ulong() + ret = gr_cl_history_changelog(libgfc, changelog_path, start, end, + num_parallel, byref(actual_end)) + if ret == -1: + _raise_changelog_err() + + if ret == -2: + raise ChangelogHistoryNotAvailable() + + return (ret, actual_end.value) + + +def history_startfresh(): + ret = libgfc.gf_history_changelog_start_fresh() + if ret == -1: + _raise_changelog_err() + + +def history_getchanges(): + def clsort(cfile): + return cfile.split('.')[-1] + + changes = [] + buf = gr_create_string_buffer(4096) + call = libgfc.gf_history_changelog_next_change + + while True: + ret = call(buf, 4096) + if ret in (0, -1): + break + + # py2 and py3 compatibility + result = bytearray_to_str(buf.raw[:ret - 1]) + changes.append(result) + + if ret == -1: + _raise_changelog_err() + + return sorted(changes, key=clsort) + + +def history_done(clfile): + ret = gr_cl_history_done(libgfc, clfile) + if ret == -1: + _raise_changelog_err() diff --git a/geo-replication/syncdaemon/logutils.py b/geo-replication/syncdaemon/logutils.py new file mode 100644 index 00000000000..01ae7852f23 --- /dev/null +++ b/geo-replication/syncdaemon/logutils.py @@ -0,0 +1,77 @@ +# -*- coding: utf-8 -*- +# +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. +# +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +import logging +from logging import Logger, handlers +import sys +import time + + +class GLogger(Logger): + + """Logger customizations for gsyncd. + + It implements a log format similar to that of glusterfs. + """ + + def makeRecord(self, name, level, *a): + rv = Logger.makeRecord(self, name, level, *a) + rv.nsecs = (rv.created - int(rv.created)) * 1000000 + fr = sys._getframe(4) + callee = fr.f_locals.get('self') + if callee: + ctx = str(type(callee)).split("'")[1].split('.')[-1] + else: + ctx = '<top>' + if not hasattr(rv, 'funcName'): + rv.funcName = fr.f_code.co_name + rv.lvlnam = logging.getLevelName(level)[0] + rv.ctx = ctx + return rv + + +LOGFMT = ("[%(asctime)s.%(nsecs)d] %(lvlnam)s [%(module)s{0}" + ":%(lineno)s:%(funcName)s] %(ctx)s: %(message)s") + + +def setup_logging(level="INFO", label="", log_file=""): + if label: + label = "(" + label + ")" + + filename = None + stream = None + if log_file: + if log_file in ('-', '/dev/stderr'): + stream = sys.stderr + elif log_file == '/dev/stdout': + stream = sys.stdout + else: + filename = log_file + + datefmt = "%Y-%m-%d %H:%M:%S" + fmt = LOGFMT.format(label) + logging.root = GLogger("root", level) + logging.setLoggerClass(GLogger) + logging.Formatter.converter = time.gmtime # Log in GMT/UTC time + logging.getLogger().handlers = [] + logging.getLogger().setLevel(level) + + if filename is not None: + logging_handler = handlers.WatchedFileHandler(filename) + formatter = logging.Formatter(fmt=fmt, + datefmt=datefmt) + logging_handler.setFormatter(formatter) + logging.getLogger().addHandler(logging_handler) + else: + logging.basicConfig(stream=stream, + format=fmt, + datefmt=datefmt, + level=level) diff --git a/geo-replication/syncdaemon/master.py b/geo-replication/syncdaemon/master.py index 51c1e571e53..9501aeae6b5 100644 --- a/geo-replication/syncdaemon/master.py +++ b/geo-replication/syncdaemon/master.py @@ -12,20 +12,23 @@ import os import sys import time import stat -import json import logging -import socket +import fcntl import string import errno -from errno import ENOENT, ENODATA, EPIPE, EEXIST +import tarfile +from errno import ENOENT, ENODATA, EEXIST, EACCES, EAGAIN, ESTALE, EINTR from threading import Condition, Lock from datetime import datetime -from gconf import gconf -from tempfile import NamedTemporaryFile -from syncdutils import Thread, GsyncdError, boolify, escape -from syncdutils import unescape, select, gauxpfx, md5hex, selfkill -from syncdutils import lstat, errno_wrap -from syncdutils import NoPurgeTimeAvailable, PartialHistoryAvailable + +import gsyncdconfig as gconf +import libgfchangelog +from rconf import rconf +from syncdutils import (Thread, GsyncdError, escape_space_newline, + unescape_space_newline, gauxpfx, escape, + lstat, errno_wrap, FreeObject, lf, matching_disk_gfid, + NoStimeAvailable, PartialHistoryAvailable, + host_brick_split) URXTIME = (-1, 0) @@ -54,36 +57,75 @@ def _volinfo_hook_relax_foreign(self): fgn_vi = volinfo_sys[self.KFGN] if fgn_vi: expiry = fgn_vi['timeout'] - int(time.time()) + 1 - logging.info('foreign volume info found, waiting %d sec for expiry' % - expiry) + logging.info(lf('foreign volume info found, waiting for expiry', + expiry=expiry)) time.sleep(expiry) volinfo_sys = self.get_sys_volinfo() return volinfo_sys +def edct(op, **ed): + dct = {} + dct['op'] = op + # This is used in automatic gfid conflict resolution. + # When marked True, it's skipped during re-processing. + dct['skip_entry'] = False + for k in ed: + if k == 'stat': + st = ed[k] + dst = dct['stat'] = {} + if st: + dst['uid'] = st.st_uid + dst['gid'] = st.st_gid + dst['mode'] = st.st_mode + dst['atime'] = st.st_atime + dst['mtime'] = st.st_mtime + else: + dct[k] = ed[k] + return dct + + # The API! def gmaster_builder(excrawl=None): """produce the GMaster class variant corresponding to sync mode""" this = sys.modules[__name__] - modemixin = gconf.special_sync_mode + modemixin = gconf.get("special-sync-mode") if not modemixin: modemixin = 'normal' - changemixin = 'xsync' if gconf.change_detector == 'xsync' \ - else excrawl or gconf.change_detector - logging.info('setting up %s change detection mode' % changemixin) + + if gconf.get("change-detector") == 'xsync': + changemixin = 'xsync' + elif excrawl: + changemixin = excrawl + else: + changemixin = gconf.get("change-detector") + + logging.debug(lf('setting up change detection mode', + mode=changemixin)) modemixin = getattr(this, modemixin.capitalize() + 'Mixin') crawlmixin = getattr(this, 'GMaster' + changemixin.capitalize() + 'Mixin') - sendmarkmixin = boolify( - gconf.use_rsync_xattrs) and SendmarkRsyncMixin or SendmarkNormalMixin - purgemixin = boolify( - gconf.ignore_deletes) and PurgeNoopMixin or PurgeNormalMixin - syncengine = boolify(gconf.use_tarssh) and TarSSHEngine or RsyncEngine + + if gconf.get("use-rsync-xattrs"): + sendmarkmixin = SendmarkRsyncMixin + else: + sendmarkmixin = SendmarkNormalMixin + + if gconf.get("ignore-deletes"): + purgemixin = PurgeNoopMixin + else: + purgemixin = PurgeNormalMixin + + if gconf.get("sync-method") == "tarssh": + syncengine = TarSSHEngine + else: + syncengine = RsyncEngine class _GMaster(crawlmixin, modemixin, sendmarkmixin, purgemixin, syncengine): pass + return _GMaster @@ -120,9 +162,9 @@ class NormalMixin(object): return xt0 >= xt1 def make_xtime_opts(self, is_master, opts): - if not 'create' in opts: + if 'create' not in opts: opts['create'] = is_master - if not 'default_xtime' in opts: + if 'default_xtime' not in opts: opts['default_xtime'] = URXTIME def xtime_low(self, rsc, path, **opts): @@ -141,7 +183,9 @@ class NormalMixin(object): xt = _xtime_now() rsc.server.aggregated.set_xtime(path, self.uuid, xt) else: - xt = opts['default_xtime'] + zero_zero = (0, 0) + if xt != zero_zero: + xt = opts['default_xtime'] return xt def keepalive_payload_hook(self, timo, gap): @@ -153,7 +197,7 @@ class NormalMixin(object): vi = vi.copy() vi['timeout'] = int(time.time()) + timo else: - # send keep-alives more frequently to + # send keep-alive more frequently to # avoid a delay in announcing our volume info # to slave if it becomes established in the # meantime @@ -168,10 +212,7 @@ class NormalMixin(object): raise GsyncdError("timestamp corruption for " + path) def need_sync(self, e, xte, xtrd): - if self.xsync_upper_limit: - return xte > xtrd and xte <= self.xsync_upper_limit - else: - return xte > xtrd + return xte > xtrd def set_slave_xtime(self, path, mark): self.slave.server.set_stime(path, self.uuid, mark) @@ -194,9 +235,9 @@ class RecoverMixin(NormalMixin): @staticmethod def make_xtime_opts(is_master, opts): - if not 'create' in opts: + if 'create' not in opts: opts['create'] = False - if not 'default_xtime' in opts: + if 'default_xtime' not in opts: opts['default_xtime'] = URXTIME def keepalive_payload_hook(self, timo, gap): @@ -239,28 +280,24 @@ class TarSSHEngine(object): """ def a_syncdata(self, files): - self.unlinked_gfids = [] - logging.debug('files: %s' % (files)) - self.current_files_skipped_count = 0 - del self.skipped_gfid_list[:] + logging.debug(lf("Files", files=files)) + for f in files: pb = self.syncer.add(f) def regjob(se, xte, pb): rv = pb.wait() if rv[0]: - logging.debug('synced ' + se) + logging.debug(lf('synced', file=se)) return True else: # stat check for file presence st = lstat(se) if isinstance(st, int): - self.unlinked_gfids.append(se) + # file got unlinked in the interim + self.unlinked_gfids.add(se) return True - logging.warn('tar+ssh: %s [errcode: %d]' % (se, rv[1])) - se_list = se.split('/'); - self.current_files_skipped_count += 1 - self.skipped_gfid_list.append(se_list[1]) + self.add_job(self.FLAT_DIR_HIERARCHY, 'reg', regjob, f, None, pb) def syncdata_wait(self): @@ -277,31 +314,25 @@ class RsyncEngine(object): """Sync engine that uses rsync(1) for data transfers""" def a_syncdata(self, files): - self.unlinked_gfids = [] - logging.debug('files: %s' % (files)) - self.current_files_skipped_count = 0 - del self.skipped_gfid_list[:] + logging.debug(lf("files", files=files)) + for f in files: - logging.debug('candidate for syncing %s' % f) + logging.debug(lf('candidate for syncing', file=f)) pb = self.syncer.add(f) def regjob(se, xte, pb): rv = pb.wait() if rv[0]: - logging.debug('synced ' + se) + logging.debug(lf('synced', file=se)) return True else: - if rv[1] in [23, 24]: - # stat to check if the file exist - st = lstat(se) - if isinstance(st, int): - # file got unlinked in the interim - self.unlinked_gfids.append(se) - return True - logging.warn('Rsync: %s [errcode: %d]' % (se, rv[1])) - se_list = se.split('/'); - self.current_files_skipped_count += 1 - self.skipped_gfid_list.append(se_list[1]) + # stat to check if the file exist + st = lstat(se) + if isinstance(st, int): + # file got unlinked in the interim + self.unlinked_gfids.add(se) + return True + self.add_job(self.FLAT_DIR_HIERARCHY, 'reg', regjob, f, None, pb) def syncdata_wait(self): @@ -345,6 +376,18 @@ class GMasterCommon(object): if self.volinfo: return self.volinfo['volume_mark'] + def get_entry_stime(self): + data = self.slave.server.entry_stime(".", self.uuid) + if isinstance(data, int): + data = None + return data + + def get_data_stime(self): + data = self.slave.server.stime(".", self.uuid) + if isinstance(data, int): + data = None + return data + def xtime(self, path, *a, **opts): """get amended xtime @@ -361,56 +404,13 @@ class GMasterCommon(object): self.make_xtime_opts(rsc == self.master, opts) return self.xtime_low(rsc, path, **opts) - def get_initial_crawl_data(self): - # while persisting only 'files_syncd' is non-zero, rest of - # the stats are nulls. lets keep it that way in case they - # are needed to be used some day... - default_data = {'files_syncd': 0, - 'files_remaining': 0, - 'bytes_remaining': 0, - 'purges_remaining': 0, - 'total_files_skipped': 0} - if getattr(gconf, 'state_detail_file', None): - try: - with open(gconf.state_detail_file, 'r+') as f: - loaded_data = json.load(f) - diff_data = set(default_data) - set(loaded_data) - if len(diff_data): - for i in diff_data: - loaded_data[i] = default_data[i] - return loaded_data - except IOError: - logging.warn('Creating new gconf.state_detail_file.') - # Create file with initial data - try: - with open(gconf.state_detail_file, 'wb') as f: - json.dump(default_data, f) - return default_data - except: - raise - return default_data - - def update_crawl_data(self): - if getattr(gconf, 'state_detail_file', None): - try: - same_dir = os.path.dirname(gconf.state_detail_file) - with NamedTemporaryFile(dir=same_dir, delete=False) as tmp: - json.dump(self.total_crawl_stats, tmp) - tmp.flush() - os.fsync(tmp.fileno()) - os.rename(tmp.name, gconf.state_detail_file) - except (IOError, OSError): - raise - def __init__(self, master, slave): self.master = master self.slave = slave self.jobtab = {} - if boolify(gconf.use_tarssh): - logging.info("using 'tar over ssh' as the sync engine") - self.syncer = Syncer(slave, self.slave.tarssh) + if gconf.get("sync-method") == "tarssh": + self.syncer = Syncer(slave, self.slave.tarssh, [2]) else: - logging.info("using 'rsync' as the sync engine") # partial transfer (cf. rsync(1)), that's normal self.syncer = Syncer(slave, self.slave.rsync, [23, 24]) # crawls vs. turns: @@ -424,24 +424,20 @@ class GMasterCommon(object): # 0. self.crawls = 0 self.turns = 0 - self.total_turns = int(gconf.turns) + self.total_turns = rconf.turns self.crawl_start = datetime.now() self.lastreport = {'crawls': 0, 'turns': 0, 'time': 0} - self.total_crawl_stats = None self.start = None self.change_seen = None # the actual volinfo we make use of self.volinfo = None self.terminate = False self.sleep_interval = 1 - self.checkpoint_thread = None - self.current_files_skipped_count = 0 - self.skipped_gfid_list = [] - self.unlinked_gfids = [] + self.unlinked_gfids = set() def init_keep_alive(cls): """start the keep-alive thread """ - timo = int(gconf.timeout or 0) + timo = gconf.get("slave-timeout", 0) if timo > 0: def keep_alive(): while True: @@ -451,14 +447,69 @@ class GMasterCommon(object): t = Thread(target=keep_alive) t.start() - def should_crawl(cls): - return gconf.glusterd_uuid in cls.master.server.node_uuid() + def mgmt_lock(self): + + """Take management volume lock """ + if rconf.mgmt_lock_fd: + try: + fcntl.lockf(rconf.mgmt_lock_fd, fcntl.LOCK_EX | fcntl.LOCK_NB) + return True + except: + ex = sys.exc_info()[1] + if isinstance(ex, IOError) and ex.errno in (EACCES, EAGAIN): + return False + raise + + fd = None + bname = str(self.uuid) + "_" + rconf.args.slave_id + "_subvol_" \ + + str(rconf.args.subvol_num) + ".lock" + mgmt_lock_dir = os.path.join(gconf.get("meta-volume-mnt"), "geo-rep") + path = os.path.join(mgmt_lock_dir, bname) + logging.debug(lf("lock file path", path=path)) + try: + fd = os.open(path, os.O_CREAT | os.O_RDWR) + except OSError: + ex = sys.exc_info()[1] + if ex.errno == ENOENT: + logging.info("Creating geo-rep directory in meta volume...") + try: + os.makedirs(mgmt_lock_dir) + except OSError: + ex = sys.exc_info()[1] + if ex.errno == EEXIST: + pass + else: + raise + fd = os.open(path, os.O_CREAT | os.O_RDWR) + else: + raise + try: + fcntl.lockf(fd, fcntl.LOCK_EX | fcntl.LOCK_NB) + # Save latest FD for future use + rconf.mgmt_lock_fd = fd + except: + ex = sys.exc_info()[1] + if isinstance(ex, IOError) and ex.errno in (EACCES, EAGAIN): + # cannot grab, it's taken + rconf.mgmt_lock_fd = fd + return False + raise + + return True + + def should_crawl(self): + if not gconf.get("use-meta-volume"): + return rconf.args.local_node_id in self.master.server.node_uuid() + + if not os.path.ismount(gconf.get("meta-volume-mnt")): + logging.error("Meta-volume is not mounted. Worker Exiting...") + sys.exit(1) + return self.mgmt_lock() def register(self): self.register() - def crawlwrap(self, oneshot=False, no_stime_update=False, - register_time=None): + def crawlwrap(self, oneshot=False, register_time=None): if oneshot: # it's important to do this during the oneshot crawl as # for a passive gsyncd (ie. in a replicate scenario) @@ -467,38 +518,35 @@ class GMasterCommon(object): # If crawlwrap is called when partial history available, # then it sets register_time which is the time when geo-rep - # worker registerd to changelog consumption. Since nsec is - # not considered in register time, their are chances of skipping - # changes detection in xsync crawl. Add 1 sec to upper_limit. - # This limit will be reset when crawlwrap is called again. - self.xsync_upper_limit = None + # worker registered to changelog consumption. Since nsec is + # not considered in register time, there are chances of skipping + # changes detection in xsync crawl. This limit will be reset when + # crawlwrap is called again. + self.live_changelog_start_time = None if register_time: - self.xsync_upper_limit = (register_time + 1, 0) + self.live_changelog_start_time = (register_time, 0) # no need to maintain volinfo state machine. # in a cascading setup, each geo-replication session is # independent (ie. 'volume-mark' and 'xtime' are not - # propogated). This is beacuse the slave's xtime is now + # propagated). This is because the slave's xtime is now # stored on the master itself. 'volume-mark' just identifies # that we are in a cascading setup and need to enable # 'geo-replication.ignore-pid-check' option. volinfo_sys = self.volinfo_hook() self.volinfo = volinfo_sys[self.KNAT] inter_master = volinfo_sys[self.KFGN] - logging.info("%s master with volume id %s ..." % - (inter_master and "intermediate" or "primary", - self.uuid)) - gconf.configinterface.set('volume_id', self.uuid) + logging.debug("%s master with volume id %s ..." % + (inter_master and "intermediate" or "primary", + self.uuid)) + rconf.volume_id = self.uuid if self.volinfo: if self.volinfo['retval']: - logging.warn("master cluster's info may not be valid %d" % - self.volinfo['retval']) - self.start_checkpoint_thread() + logging.warn(lf("master cluster's info may not be valid", + error=self.volinfo['retval'])) else: raise GsyncdError("master volinfo unavailable") - self.total_crawl_stats = self.get_initial_crawl_data() self.lastreport['time'] = time.time() - logging.info('crawl interval: %d seconds' % self.sleep_interval) t0 = time.time() crawl = self.should_crawl() @@ -509,61 +557,56 @@ class GMasterCommon(object): self.start = time.time() should_display_info = self.start - self.lastreport['time'] >= 60 if should_display_info: - logging.info("%d crawls, %d turns", - self.crawls - self.lastreport['crawls'], - self.turns - self.lastreport['turns']) + logging.debug("%d crawls, %d turns", + self.crawls - self.lastreport['crawls'], + self.turns - self.lastreport['turns']) self.lastreport.update(crawls=self.crawls, turns=self.turns, time=self.start) t1 = time.time() - if int(t1 - t0) >= int(gconf.replica_failover_interval): + if int(t1 - t0) >= gconf.get("replica-failover-interval"): crawl = self.should_crawl() t0 = t1 self.update_worker_remote_node() if not crawl: - self.update_worker_health("Passive") + self.status.set_passive() # bring up _this_ brick to the cluster stime # which is min of cluster (but max of the replicas) brick_stime = self.xtime('.', self.slave) cluster_stime = self.master.server.aggregated.stime_mnt( - '.', '.'.join([str(self.uuid), str(gconf.slave_id)])) - logging.debug("Cluster stime: %s | Brick stime: %s" % - (repr(cluster_stime), repr(brick_stime))) + '.', '.'.join([str(self.uuid), rconf.args.slave_id])) + logging.debug(lf("Crawl info", + cluster_stime=cluster_stime, + brick_stime=brick_stime)) + if not isinstance(cluster_stime, int): if brick_stime < cluster_stime: self.slave.server.set_stime( self.FLAT_DIR_HIERARCHY, self.uuid, cluster_stime) + self.upd_stime(cluster_stime) + # Purge all changelogs available in processing dir + # less than cluster_stime + proc_dir = os.path.join(self.tempdir, + ".processing") + + if os.path.exists(proc_dir): + to_purge = [f for f in os.listdir(proc_dir) + if (f.startswith("CHANGELOG.") and + int(f.split('.')[-1]) < + cluster_stime[0])] + for f in to_purge: + os.remove(os.path.join(proc_dir, f)) + time.sleep(5) continue - self.update_worker_health("Active") - self.crawl(no_stime_update=no_stime_update) + + self.status.set_active() + self.crawl() + if oneshot: return time.sleep(self.sleep_interval) - @classmethod - def _checkpt_param(cls, chkpt, prm, xtimish=True): - """use config backend to lookup a parameter belonging to - checkpoint @chkpt""" - cprm = gconf.configinterface.get_realtime('checkpoint_' + prm) - if not cprm: - return - chkpt_mapped, val = cprm.split(':', 1) - if unescape(chkpt_mapped) != chkpt: - return - if xtimish: - val = cls.deserialize_xtime(val) - return val - - @classmethod - def _set_checkpt_param(cls, chkpt, prm, val, xtimish=True): - """use config backend to store a parameter associated - with checkpoint @chkpt""" - if xtimish: - val = cls.serialize_xtime(val) - gconf.configinterface.set( - 'checkpoint_' + prm, "%s:%s" % (escape(chkpt), val)) - @staticmethod def humantime(*tpair): """format xtime-like (sec, nsec) pair to human readable format""" @@ -592,116 +635,6 @@ class GMasterCommon(object): string.zfill(m, 2), string.zfill(s, 2)) return date - def checkpt_service(self, chan, chkpt): - """checkpoint service loop - - monitor and verify checkpoint status for @chkpt, and listen - for incoming requests for whom we serve a pretty-formatted - status report""" - while True: - chkpt = gconf.configinterface.get_realtime("checkpoint") - if not chkpt: - gconf.configinterface.delete("checkpoint_completed") - gconf.configinterface.delete("checkpoint_target") - # dummy loop for the case when there is no checkpt set - select([chan], [], []) - conn, _ = chan.accept() - conn.send('\0') - conn.close() - continue - - checkpt_tgt = self._checkpt_param(chkpt, 'target') - if not checkpt_tgt: - checkpt_tgt = self.xtime('.') - if isinstance(checkpt_tgt, int): - raise GsyncdError("master root directory is " - "unaccessible (%s)", - os.strerror(checkpt_tgt)) - self._set_checkpt_param(chkpt, 'target', checkpt_tgt) - logging.debug("checkpoint target %s has been determined " - "for checkpoint %s" % - (repr(checkpt_tgt), chkpt)) - - # check if the label is 'now' - chkpt_lbl = chkpt - try: - x1, x2 = chkpt.split(':') - if x1 == 'now': - chkpt_lbl = "as of " + self.humantime(x2) - except: - pass - completed = self._checkpt_param(chkpt, 'completed', xtimish=False) - if completed: - completed = tuple(int(x) for x in completed.split('.')) - s, _, _ = select([chan], [], [], (not completed) and 5 or None) - # either request made and we re-check to not - # give back stale data, or we still hunting for completion - if (self.native_xtime(checkpt_tgt) and ( - self.native_xtime(checkpt_tgt) < self.volmark)): - # indexing has been reset since setting the checkpoint - status = "is invalid" - else: - xtr = self.xtime('.', self.slave) - if isinstance(xtr, int): - raise GsyncdError("slave root directory is " - "unaccessible (%s)", - os.strerror(xtr)) - ncompleted = self.xtime_geq(xtr, checkpt_tgt) - if completed and not ncompleted: # stale data - logging.warn("completion time %s for checkpoint %s " - "became stale" % - (self.humantime(*completed), chkpt)) - completed = None - gconf.configinterface.delete('checkpoint_completed') - if ncompleted and not completed: # just reaching completion - completed = "%.6f" % time.time() - self._set_checkpt_param( - chkpt, 'completed', completed, xtimish=False) - completed = tuple(int(x) for x in completed.split('.')) - logging.info("checkpoint %s completed" % chkpt) - status = completed and \ - "completed at " + self.humantime(completed[0]) or \ - "not reached yet" - if s: - conn = None - try: - conn, _ = chan.accept() - try: - conn.send("checkpoint %s is %s\0" % - (chkpt_lbl, status)) - except: - exc = sys.exc_info()[1] - if ((isinstance(exc, OSError) or isinstance( - exc, IOError)) and exc.errno == EPIPE): - logging.debug('checkpoint client disconnected') - else: - raise - finally: - if conn: - conn.close() - - def start_checkpoint_thread(self): - """prepare and start checkpoint service""" - if self.checkpoint_thread or not ( - getattr(gconf, 'state_socket_unencoded', None) and getattr( - gconf, 'socketdir', None) - ): - return - chan = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM) - state_socket = os.path.join( - gconf.socketdir, md5hex(gconf.state_socket_unencoded) + ".socket") - try: - os.unlink(state_socket) - except: - if sys.exc_info()[0] == OSError: - pass - chan.bind(state_socket) - chan.listen(1) - chkpt = gconf.configinterface.get_realtime("checkpoint") - t = Thread(target=self.checkpt_service, args=(chan, chkpt)) - t.start() - self.checkpoint_thread = t - def add_job(self, path, label, job, *a, **kw): """insert @job function to job table at @path with @label""" if self.jobtab.get(path) is None: @@ -717,7 +650,7 @@ class GMasterCommon(object): """perform jobs registered for @path Reset jobtab entry for @path, - determine success as the conjuction of + determine success as the conjunction of success of all the jobs. In case of success, call .sendmark on @path """ @@ -742,10 +675,12 @@ class GMasterCommon(object): class XCrawlMetadata(object): - def __init__(self, st_uid, st_gid, st_mode): + def __init__(self, st_uid, st_gid, st_mode, st_atime, st_mtime): self.st_uid = int(st_uid) self.st_gid = int(st_gid) self.st_mode = int(st_mode) + self.st_atime = float(st_atime) + self.st_mtime = float(st_mtime) class GMasterChangelogMixin(GMasterCommon): @@ -755,6 +690,7 @@ class GMasterChangelogMixin(GMasterCommon): # index for change type and entry IDX_START = 0 IDX_END = 2 + UNLINK_ENTRY = 2 POS_GFID = 0 POS_TYPE = 1 @@ -764,30 +700,291 @@ class GMasterChangelogMixin(GMasterCommon): TYPE_GFID = "D " TYPE_ENTRY = "E " - # flat directory heirarchy for gfid based access - FLAT_DIR_HIERARCHY = '.' + MAX_EF_RETRIES = 10 + MAX_OE_RETRIES = 10 - # maximum retries per changelog before giving up - MAX_RETRIES = 10 + # flat directory hierarchy for gfid based access + FLAT_DIR_HIERARCHY = '.' - CHANGELOG_LOG_LEVEL = 9 CHANGELOG_CONN_RETRIES = 5 - def fallback_xsync(self): - logging.info('falling back to xsync mode') - gconf.configinterface.set('change-detector', 'xsync') - selfkill() + def init_fop_batch_stats(self): + self.batch_stats = { + "CREATE": 0, + "MKNOD": 0, + "UNLINK": 0, + "MKDIR": 0, + "RMDIR": 0, + "LINK": 0, + "SYMLINK": 0, + "RENAME": 0, + "SETATTR": 0, + "SETXATTR": 0, + "XATTROP": 0, + "DATA": 0, + "ENTRY_SYNC_TIME": 0, + "META_SYNC_TIME": 0, + "DATA_START_TIME": 0 + } + + def update_fop_batch_stats(self, ty): + if ty in ['FSETXATTR']: + ty = 'SETXATTR' + self.batch_stats[ty] = self.batch_stats.get(ty, 0) + 1 + + def archive_and_purge_changelogs(self, changelogs): + # Creates tar file instead of tar.gz, since changelogs will + # be appended to existing tar. archive name is + # archive_<YEAR><MONTH>.tar + archive_name = "archive_%s.tar" % datetime.today().strftime( + gconf.get("changelog-archive-format")) + + try: + tar = tarfile.open(os.path.join(self.processed_changelogs_dir, + archive_name), + "a") + except tarfile.ReadError: + tar = tarfile.open(os.path.join(self.processed_changelogs_dir, + archive_name), + "w") + + for f in changelogs: + try: + f = os.path.basename(f) + tar.add(os.path.join(self.processed_changelogs_dir, f), + arcname=os.path.basename(f)) + except: + exc = sys.exc_info()[1] + if ((isinstance(exc, OSError) or + isinstance(exc, IOError)) and exc.errno == ENOENT): + continue + else: + tar.close() + raise + tar.close() + + for f in changelogs: + try: + f = os.path.basename(f) + os.remove(os.path.join(self.processed_changelogs_dir, f)) + except OSError as e: + if e.errno == errno.ENOENT: + continue + else: + raise def setup_working_dir(self): - workdir = os.path.join(gconf.working_dir, md5hex(gconf.local_path)) + workdir = os.path.join(gconf.get("working-dir"), + escape(rconf.args.local_path)) logging.debug('changelog working dir %s' % workdir) return workdir - def get_purge_time(self): - purge_time = self.xtime('.', self.slave) - if isinstance(purge_time, int): - purge_time = None - return purge_time + def log_failures(self, failures, entry_key, gfid_prefix, log_prefix): + num_failures = 0 + for failure in failures: + st = lstat(os.path.join(gfid_prefix, failure[0][entry_key])) + if not isinstance(st, int): + num_failures += 1 + logging.error(lf('%s FAILED' % log_prefix, + data=failure)) + if failure[0]['op'] == 'MKDIR': + raise GsyncdError("The above directory failed to sync." + " Please fix it to proceed further.") + + self.status.inc_value("failures", num_failures) + + def fix_possible_entry_failures(self, failures, retry_count, entries): + pfx = gauxpfx() + fix_entry_ops = [] + failures1 = [] + remove_gfids = set() + for failure in failures: + if failure[2]['name_mismatch']: + pbname = failure[2]['slave_entry'] + elif failure[2]['dst']: + pbname = failure[0]['entry1'] + else: + pbname = failure[0]['entry'] + + op = failure[0]['op'] + # name exists but gfid is different + if failure[2]['gfid_mismatch'] or failure[2]['name_mismatch']: + slave_gfid = failure[2]['slave_gfid'] + st = lstat(os.path.join(pfx, slave_gfid)) + # Takes care of scenarios with no hardlinks + if isinstance(st, int) and st == ENOENT: + logging.debug(lf('Entry not present on master. Fixing gfid ' + 'mismatch in slave. Deleting the entry', + retry_count=retry_count, + entry=repr(failure))) + # Add deletion to fix_entry_ops list + if failure[2]['slave_isdir']: + fix_entry_ops.append( + edct('RMDIR', + gfid=failure[2]['slave_gfid'], + entry=pbname)) + else: + fix_entry_ops.append( + edct('UNLINK', + gfid=failure[2]['slave_gfid'], + entry=pbname)) + remove_gfids.add(slave_gfid) + if op in ['RENAME']: + # If renamed gfid doesn't exists on master, remove + # rename entry and unlink src on slave + st = lstat(os.path.join(pfx, failure[0]['gfid'])) + if isinstance(st, int) and st == ENOENT: + logging.debug("Unlink source %s" % repr(failure)) + remove_gfids.add(failure[0]['gfid']) + fix_entry_ops.append( + edct('UNLINK', + gfid=failure[0]['gfid'], + entry=failure[0]['entry'])) + # Takes care of scenarios of hardlinks/renames on master + elif not isinstance(st, int): + if matching_disk_gfid(slave_gfid, pbname): + # Safe to ignore the failure as master contains same + # file with same gfid. Remove entry from entries list + logging.debug(lf('Fixing gfid mismatch in slave. ' + ' Safe to ignore, take out entry', + retry_count=retry_count, + entry=repr(failure))) + remove_gfids.add(failure[0]['gfid']) + if op == 'RENAME': + fix_entry_ops.append( + edct('UNLINK', + gfid=failure[0]['gfid'], + entry=failure[0]['entry'])) + # The file exists on master but with different name. + # Probably renamed and got missed during xsync crawl. + elif failure[2]['slave_isdir']: + realpath = os.readlink(os.path.join( + rconf.args.local_path, + ".glusterfs", + slave_gfid[0:2], + slave_gfid[2:4], + slave_gfid)) + dst_entry = os.path.join(pfx, realpath.split('/')[-2], + realpath.split('/')[-1]) + src_entry = pbname + logging.debug(lf('Fixing dir name/gfid mismatch in ' + 'slave', retry_count=retry_count, + entry=repr(failure))) + if src_entry == dst_entry: + # Safe to ignore the failure as master contains + # same directory as in slave with same gfid. + # Remove the failure entry from entries list + logging.debug(lf('Fixing dir name/gfid mismatch' + ' in slave. Safe to ignore, ' + 'take out entry', + retry_count=retry_count, + entry=repr(failure))) + try: + entries.remove(failure[0]) + except ValueError: + pass + else: + rename_dict = edct('RENAME', gfid=slave_gfid, + entry=src_entry, + entry1=dst_entry, stat=st, + link=None) + logging.debug(lf('Fixing dir name/gfid mismatch' + ' in slave. Renaming', + retry_count=retry_count, + entry=repr(rename_dict))) + fix_entry_ops.append(rename_dict) + else: + # A hardlink file exists with different name or + # renamed file exists and we are sure from + # matching_disk_gfid check that the entry doesn't + # exist with same gfid so we can safely delete on slave + logging.debug(lf('Fixing file gfid mismatch in slave. ' + 'Hardlink/Rename Case. Deleting entry', + retry_count=retry_count, + entry=repr(failure))) + fix_entry_ops.append( + edct('UNLINK', + gfid=failure[2]['slave_gfid'], + entry=pbname)) + elif failure[1] == ENOENT: + if op in ['RENAME']: + pbname = failure[0]['entry1'] + else: + pbname = failure[0]['entry'] + + pargfid = pbname.split('/')[1] + st = lstat(os.path.join(pfx, pargfid)) + # Safe to ignore the failure as master doesn't contain + # parent directory. + if isinstance(st, int): + logging.debug(lf('Fixing ENOENT error in slave. Parent ' + 'does not exist on master. Safe to ' + 'ignore, take out entry', + retry_count=retry_count, + entry=repr(failure))) + try: + entries.remove(failure[0]) + except ValueError: + pass + else: + logging.debug(lf('Fixing ENOENT error in slave. Create ' + 'parent directory on slave.', + retry_count=retry_count, + entry=repr(failure))) + realpath = os.readlink(os.path.join(rconf.args.local_path, + ".glusterfs", + pargfid[0:2], + pargfid[2:4], + pargfid)) + dir_entry = os.path.join(pfx, realpath.split('/')[-2], + realpath.split('/')[-1]) + fix_entry_ops.append( + edct('MKDIR', gfid=pargfid, entry=dir_entry, + mode=st.st_mode, uid=st.st_uid, gid=st.st_gid)) + + logging.debug("remove_gfids: %s" % repr(remove_gfids)) + if remove_gfids: + for e in entries: + if e['op'] in ['MKDIR', 'MKNOD', 'CREATE', 'RENAME'] \ + and e['gfid'] in remove_gfids: + logging.debug("Removed entry op from retrial list: entry: %s" % repr(e)) + e['skip_entry'] = True + + if fix_entry_ops: + # Process deletions of entries whose gfids are mismatched + failures1 = self.slave.server.entry_ops(fix_entry_ops) + + return (failures1, fix_entry_ops) + + def handle_entry_failures(self, failures, entries): + retries = 0 + pending_failures = False + failures1 = [] + failures2 = [] + entry_ops1 = [] + entry_ops2 = [] + + if failures: + pending_failures = True + failures1 = failures + entry_ops1 = entries + + while pending_failures and retries < self.MAX_EF_RETRIES: + retries += 1 + (failures2, entry_ops2) = self.fix_possible_entry_failures( + failures1, retries, entry_ops1) + if not failures2: + pending_failures = False + logging.info(lf('Successfully fixed entry ops with gfid ' + 'mismatch', retry_count=retries)) + else: + pending_failures = True + failures1 = failures2 + entry_ops1 = entry_ops2 + + if pending_failures: + for failure in failures1: + logging.error("Failed to fix entry ops %s", repr(failure)) def process_change(self, change, done, retry): pfx = gauxpfx() @@ -796,8 +993,28 @@ class GMasterChangelogMixin(GMasterCommon): meta_gfid = set() datas = set() - # basic crawl stats: files and bytes - files_pending = {'count': 0, 'purge': 0, 'bytes': 0, 'files': []} + change_ts = change.split(".")[-1] + + # Ignore entry ops which are already processed in Changelog modes + ignore_entry_ops = False + entry_stime = None + data_stime = None + if self.name in ["live_changelog", "history_changelog"]: + entry_stime = self.get_entry_stime() + data_stime = self.get_data_stime() + + if entry_stime is not None and data_stime is not None: + # if entry_stime is not None but data_stime > entry_stime + # This situation is caused by the stime update of Passive worker + # Consider data_stime in this case. + if data_stime[0] > entry_stime[0]: + entry_stime = data_stime + + # Compare the entry_stime with changelog file suffix + # if changelog time is less than entry_stime then ignore + if int(change_ts) <= entry_stime[0]: + ignore_entry_ops = True + try: f = open(change, "r") clist = f.readlines() @@ -805,34 +1022,27 @@ class GMasterChangelogMixin(GMasterCommon): except IOError: raise - def edct(op, **ed): - dct = {} - dct['op'] = op - for k in ed: - if k == 'stat': - st = ed[k] - dst = dct['stat'] = {} - if st: - dst['uid'] = st.st_uid - dst['gid'] = st.st_gid - dst['mode'] = st.st_mode - else: - dct[k] = ed[k] - return dct - - # entry counts (not purges) - def entry_update(): - files_pending['count'] += 1 - - # purge count - def purge_update(): - files_pending['purge'] += 1 - for e in clist: e = e.strip() et = e[self.IDX_START:self.IDX_END] # entry type ec = e[self.IDX_END:].split(' ') # rest of the bits + # skip ENTRY operation if hot tier brick + if self.name == 'live_changelog' or \ + self.name == 'history_changelog': + if rconf.args.is_hottier and et == self.TYPE_ENTRY: + logging.debug(lf('skip ENTRY op if hot tier brick', + op=ec[self.POS_TYPE])) + continue + + # Data and Meta operations are decided while parsing + # UNLINK/RMDIR/MKNOD except that case ignore all the other + # entry ops if ignore_entry_ops is True. + # UNLINK/RMDIR/MKNOD entry_ops are ignored in the end + if ignore_entry_ops and et == self.TYPE_ENTRY and \ + ec[self.POS_TYPE] not in ["UNLINK", "RMDIR", "MKNOD"]: + continue + if et == self.TYPE_ENTRY: # extract information according to the type of # the entry operation. create(), mkdir() and mknod() @@ -840,56 +1050,128 @@ class GMasterChangelogMixin(GMasterCommon): # itself, so no need to stat()... ty = ec[self.POS_TYPE] + self.update_fop_batch_stats(ec[self.POS_TYPE]) + # PARGFID/BNAME - en = unescape(os.path.join(pfx, ec[self.POS_ENTRY1])) + en = unescape_space_newline( + os.path.join(pfx, ec[self.POS_ENTRY1])) # GFID of the entry gfid = ec[self.POS_GFID] if ty in ['UNLINK', 'RMDIR']: + # The index of PARGFID/BNAME for UNLINK, RMDIR + # is no more the last index. It varies based on + # changelog.capture-del-path is enabled or not. + en = unescape_space_newline( + os.path.join(pfx, ec[self.UNLINK_ENTRY])) + # Remove from DATA list, so that rsync will # not fail pt = os.path.join(pfx, ec[0]) - if pt in datas: + st = lstat(pt) + if pt in datas and isinstance(st, int): + # file got unlinked, May be historical Changelog datas.remove(pt) - purge_update() - entries.append(edct(ty, gfid=gfid, entry=en)) + if ty in ['RMDIR'] and not isinstance(st, int): + logging.info(lf('Ignoring rmdir. Directory present in ' + 'master', gfid=gfid, pgfid_bname=en)) + continue + + if not gconf.get("ignore-deletes"): + if not ignore_entry_ops: + entries.append(edct(ty, gfid=gfid, entry=en)) elif ty in ['CREATE', 'MKDIR', 'MKNOD']: - entry_update() - # stat information present in the changelog itself + # Special case: record mknod as link + if ty in ['MKNOD']: + mode = int(ec[2]) + if mode & 0o1000: + # Avoid stat'ing the file as it + # may be deleted in the interim + st = FreeObject(st_mode=int(ec[2]), + st_uid=int(ec[3]), + st_gid=int(ec[4]), + st_atime=0, + st_mtime=0) + + # So, it may be deleted, but still we are + # append LINK? Because, the file will be + # CREATED if source not exists. + entries.append(edct('LINK', stat=st, entry=en, + gfid=gfid)) + + # Here, we have the assumption that only + # tier-gfid.linkto causes this mknod. Add data + datas.add(os.path.join(pfx, ec[0])) + continue + + # stat info. present in the changelog itself entries.append(edct(ty, gfid=gfid, entry=en, - mode=int(ec[2]), - uid=int(ec[3]), gid=int(ec[4]))) + mode=int(ec[2]), + uid=int(ec[3]), gid=int(ec[4]))) elif ty == "RENAME": go = os.path.join(pfx, gfid) st = lstat(go) if isinstance(st, int): st = {} - entry_update() - e1 = unescape(os.path.join(pfx, ec[self.POS_ENTRY1 - 1])) + rl = None + if st and stat.S_ISLNK(st.st_mode): + rl = errno_wrap(os.readlink, [en], [ENOENT], + [ESTALE, EINTR]) + if isinstance(rl, int): + rl = None + + e1 = unescape_space_newline( + os.path.join(pfx, ec[self.POS_ENTRY1 - 1])) entries.append(edct(ty, gfid=gfid, entry=e1, entry1=en, - stat=st)) + stat=st, link=rl)) + # If src doesn't exist while doing rename, destination + # is created. If data is not followed by rename, this + # remains zero byte file on slave. Hence add data entry + # for renames + datas.add(os.path.join(pfx, gfid)) else: # stat() to get mode and other information + if not matching_disk_gfid(gfid, en): + logging.debug(lf('Ignoring entry, purged in the ' + 'interim', file=en, gfid=gfid)) + continue + go = os.path.join(pfx, gfid) st = lstat(go) if isinstance(st, int): - logging.debug('file %s got purged in the interim' % go) + logging.debug(lf('Ignoring entry, purged in the ' + 'interim', file=en, gfid=gfid)) continue if ty == 'LINK': - entry_update() - entries.append(edct(ty, stat=st, entry=en, gfid=gfid)) + rl = None + if st and stat.S_ISLNK(st.st_mode): + rl = errno_wrap(os.readlink, [en], [ENOENT], + [ESTALE, EINTR]) + if isinstance(rl, int): + rl = None + entries.append(edct(ty, stat=st, entry=en, gfid=gfid, + link=rl)) + # If src doesn't exist while doing link, destination + # is created based on file type. If data is not + # followed by link, this remains zero byte file on + # slave. Hence add data entry for links + if rl is None: + datas.add(os.path.join(pfx, gfid)) elif ty == 'SYMLINK': - rl = errno_wrap(os.readlink, [en], [ENOENT]) + rl = errno_wrap(os.readlink, [en], [ENOENT], + [ESTALE, EINTR]) if isinstance(rl, int): continue - entry_update() + entries.append( edct(ty, stat=st, entry=en, gfid=gfid, link=rl)) else: - logging.warn('ignoring %s [op %s]' % (gfid, ty)) + logging.warn(lf('ignoring op', + gfid=gfid, + type=ty)) elif et == self.TYPE_GFID: # If self.unlinked_gfids is available, then that means it is # retrying the changelog second time. Do not add the GFID's @@ -900,6 +1182,7 @@ class GMasterChangelogMixin(GMasterCommon): else: datas.add(os.path.join(pfx, ec[0])) elif et == self.TYPE_META: + self.update_fop_batch_stats(ec[self.POS_TYPE]) if ec[1] == 'SETATTR': # only setattr's for now... if len(ec) == 5: # In xsync crawl, we already have stat data @@ -907,17 +1190,74 @@ class GMasterChangelogMixin(GMasterCommon): meta_gfid.add((os.path.join(pfx, ec[0]), XCrawlMetadata(st_uid=ec[2], st_gid=ec[3], - st_mode=ec[4]))) + st_mode=ec[4], + st_atime=ec[5], + st_mtime=ec[6]))) else: meta_gfid.add((os.path.join(pfx, ec[0]), )) + elif ec[1] in ['SETXATTR', 'XATTROP', 'FXATTROP']: + # To sync xattr/acls use rsync/tar, --xattrs and --acls + # switch to rsync and tar + if not gconf.get("sync-method") == "tarssh" and \ + (gconf.get("sync-xattrs") or gconf.get("sync-acls")): + datas.add(os.path.join(pfx, ec[0])) else: - logging.warn('got invalid changelog type: %s' % (et)) + logging.warn(lf('got invalid fop type', + type=et)) logging.debug('entries: %s' % repr(entries)) - if not retry: - self.update_worker_cumilitive_status(files_pending) + + # Increment counters for Status + self.files_in_batch += len(datas) + self.status.inc_value("data", len(datas)) + + self.batch_stats["DATA"] += self.files_in_batch - \ + self.batch_stats["SETXATTR"] - \ + self.batch_stats["XATTROP"] + + entry_start_time = time.time() # sync namespace - if entries: - self.slave.server.entry_ops(entries) + if entries and not ignore_entry_ops: + # Increment counters for Status + self.status.inc_value("entry", len(entries)) + + failures = self.slave.server.entry_ops(entries) + + if gconf.get("gfid-conflict-resolution"): + count = 0 + if failures: + logging.info(lf('Entry ops failed with gfid mismatch', + count=len(failures))) + while failures and count < self.MAX_OE_RETRIES: + count += 1 + self.handle_entry_failures(failures, entries) + logging.info(lf('Retry original entries', count=count)) + failures = self.slave.server.entry_ops(entries) + if not failures: + logging.info("Successfully fixed all entry ops with " + "gfid mismatch") + break + + self.log_failures(failures, 'gfid', gauxpfx(), 'ENTRY') + self.status.dec_value("entry", len(entries)) + + # Update Entry stime in Brick Root only in case of Changelog mode + if self.name in ["live_changelog", "history_changelog"]: + entry_stime_to_update = (int(change_ts) - 1, 0) + self.upd_entry_stime(entry_stime_to_update) + self.status.set_field("last_synced_entry", + entry_stime_to_update[0]) + + self.batch_stats["ENTRY_SYNC_TIME"] += time.time() - entry_start_time + + if ignore_entry_ops: + # Book keeping, to show in logs the range of Changelogs skipped + self.num_skipped_entry_changelogs += 1 + if self.skipped_entry_changelogs_first is None: + self.skipped_entry_changelogs_first = change_ts + + self.skipped_entry_changelogs_last = change_ts + + meta_start_time = time.time() # sync metadata if meta_gfid: meta_entries = [] @@ -927,23 +1267,41 @@ class GMasterChangelogMixin(GMasterCommon): else: st = lstat(go[0]) if isinstance(st, int): - logging.debug('file %s got purged in the interim' % go[0]) + logging.debug(lf('file got purged in the interim', + file=go[0])) continue meta_entries.append(edct('META', go=go[0], stat=st)) if meta_entries: - self.slave.server.meta_ops(meta_entries) + self.status.inc_value("meta", len(meta_entries)) + failures = self.slave.server.meta_ops(meta_entries) + self.log_failures(failures, 'go', '', 'META') + self.status.dec_value("meta", len(meta_entries)) + + self.batch_stats["META_SYNC_TIME"] += time.time() - meta_start_time + + if self.batch_stats["DATA_START_TIME"] == 0: + self.batch_stats["DATA_START_TIME"] = time.time() + # sync data if datas: self.a_syncdata(datas) + self.datas_in_batch.update(datas) def process(self, changes, done=1): tries = 0 retry = False + self.unlinked_gfids = set() + self.files_in_batch = 0 + self.datas_in_batch = set() + # Error log disabled till the last round + self.syncer.disable_errorlog() + self.skipped_entry_changelogs_first = None + self.skipped_entry_changelogs_last = None + self.num_skipped_entry_changelogs = 0 + self.batch_start_time = time.time() + self.init_fop_batch_stats() while True: - self.skipped_gfid_list = [] - self.current_files_skipped_count = 0 - # first, fire all changelog transfers in parallel. entry and # metadata are performed synchronously, therefore in serial. # However at the end of each changelog, data is synchronized @@ -951,12 +1309,27 @@ class GMasterChangelogMixin(GMasterCommon): # entries/metadata of that changelog but happens in parallel # with data of other changelogs. - for change in changes: - logging.debug('processing change %s' % change) - self.process_change(change, done, retry) - if not retry: - # number of changelogs processed in the batch - self.turns += 1 + if retry: + if tries == (gconf.get("max-rsync-retries") - 1): + # Enable Error logging if it is last retry + self.syncer.enable_errorlog() + + # Remove Unlinked GFIDs from Queue + for unlinked_gfid in self.unlinked_gfids: + if unlinked_gfid in self.datas_in_batch: + self.datas_in_batch.remove(unlinked_gfid) + + # Retry only Sync. Do not retry entry ops + if self.datas_in_batch: + self.a_syncdata(self.datas_in_batch) + else: + for change in changes: + logging.debug(lf('processing change', + changelog=change)) + self.process_change(change, done, retry) + if not retry: + # number of changelogs processed in the batch + self.turns += 1 # Now we wait for all the data transfers fired off in the above # step to complete. Note that this is not ideal either. Ideally @@ -973,36 +1346,43 @@ class GMasterChangelogMixin(GMasterCommon): # update the slave's time with the timestamp of the _last_ # changelog file time suffix. Since, the changelog prefix time # is the time when the changelog was rolled over, introduce a - # tolerence of 1 second to counter the small delta b/w the + # tolerance of 1 second to counter the small delta b/w the # marker update and gettimeofday(). # NOTE: this is only for changelog mode, not xsync. # @change is the last changelog (therefore max time for this batch) if self.syncdata_wait(): - self.unlinked_gfids = [] + self.unlinked_gfids = set() if done: xtl = (int(change.split('.')[-1]) - 1, 0) self.upd_stime(xtl) - map(self.changelog_done_func, changes) - self.update_worker_files_syncd() + list(map(self.changelog_done_func, changes)) + self.archive_and_purge_changelogs(changes) + + # Reset Data counter after sync + self.status.dec_value("data", self.files_in_batch) + self.files_in_batch = 0 + self.datas_in_batch = set() break # We do not know which changelog transfer failed, retry everything. retry = True tries += 1 - if tries == self.MAX_RETRIES: - logging.warn('changelogs %s could not be processed - ' - 'moving on...' % - ' '.join(map(os.path.basename, changes))) - self.update_worker_total_files_skipped( - self.current_files_skipped_count) - logging.warn('SKIPPED GFID = %s' % - ','.join(self.skipped_gfid_list)) - self.update_worker_files_syncd() + if tries == gconf.get("max-rsync-retries"): + logging.error(lf('changelogs could not be processed ' + 'completely - moving on...', + files=list(map(os.path.basename, changes)))) + + # Reset data counter on failure + self.status.dec_value("data", self.files_in_batch) + self.files_in_batch = 0 + self.datas_in_batch = set() + if done: xtl = (int(change.split('.')[-1]) - 1, 0) self.upd_stime(xtl) - map(self.changelog_done_func, changes) + list(map(self.changelog_done_func, changes)) + self.archive_and_purge_changelogs(changes) break # it's either entry_ops() or Rsync that failed to do it's # job. Mostly it's entry_ops() [which currently has a problem @@ -1011,234 +1391,187 @@ class GMasterChangelogMixin(GMasterCommon): # entry_ops() that failed... so we retry the _whole_ changelog # again. # TODO: remove entry retries when it's gets fixed. - logging.warn('incomplete sync, retrying changelogs: %s' % - ' '.join(map(os.path.basename, changes))) + logging.warn(lf('incomplete sync, retrying changelogs', + files=list(map(os.path.basename, changes)))) + + # Reset the Data counter before Retry + self.status.dec_value("data", self.files_in_batch) + self.files_in_batch = 0 + self.init_fop_batch_stats() time.sleep(0.5) + # Log the Skipped Entry ops range if any + if self.skipped_entry_changelogs_first is not None and \ + self.skipped_entry_changelogs_last is not None: + logging.info(lf("Skipping already processed entry ops", + from_changelog=self.skipped_entry_changelogs_first, + to_changelog=self.skipped_entry_changelogs_last, + num_changelogs=self.num_skipped_entry_changelogs)) + + # Log Current batch details + if changes: + logging.info( + lf("Entry Time Taken", + UNL=self.batch_stats["UNLINK"], + RMD=self.batch_stats["RMDIR"], + CRE=self.batch_stats["CREATE"], + MKN=self.batch_stats["MKNOD"], + MKD=self.batch_stats["MKDIR"], + REN=self.batch_stats["RENAME"], + LIN=self.batch_stats["LINK"], + SYM=self.batch_stats["SYMLINK"], + duration="%.4f" % self.batch_stats["ENTRY_SYNC_TIME"])) + + logging.info( + lf("Data/Metadata Time Taken", + SETA=self.batch_stats["SETATTR"], + meta_duration="%.4f" % self.batch_stats["META_SYNC_TIME"], + SETX=self.batch_stats["SETXATTR"], + XATT=self.batch_stats["XATTROP"], + DATA=self.batch_stats["DATA"], + data_duration="%.4f" % ( + time.time() - self.batch_stats["DATA_START_TIME"]))) + + logging.info( + lf("Batch Completed", + mode=self.name, + duration="%.4f" % (time.time() - self.batch_start_time), + changelog_start=changes[0].split(".")[-1], + changelog_end=changes[-1].split(".")[-1], + num_changelogs=len(changes), + stime=self.get_data_stime(), + entry_stime=self.get_entry_stime())) + + def upd_entry_stime(self, stime): + self.slave.server.set_entry_stime(self.FLAT_DIR_HIERARCHY, + self.uuid, + stime) + def upd_stime(self, stime, path=None): if not path: path = self.FLAT_DIR_HIERARCHY if not stime == URXTIME: self.sendmark(path, stime) - def get_worker_status_file(self): - file_name = gconf.local_path + '.status' - file_name = file_name.replace("/", "_") - worker_status_file = gconf.georep_session_working_dir + file_name - return worker_status_file - - def update_worker_status(self, key, value): - default_data = {"remote_node": "N/A", - "worker status": "Not Started", - "crawl status": "N/A", - "files_syncd": 0, - "files_remaining": 0, - "bytes_remaining": 0, - "purges_remaining": 0, - "total_files_skipped": 0} - worker_status_file = self.get_worker_status_file() - try: - with open(worker_status_file, 'r+') as f: - loaded_data = json.load(f) - loaded_data[key] = value - os.ftruncate(f.fileno(), 0) - os.lseek(f.fileno(), 0, os.SEEK_SET) - json.dump(loaded_data, f) - f.flush() - os.fsync(f.fileno()) - except (IOError, OSError, ValueError): - logging.info('Creating new %s' % worker_status_file) - try: - with open(worker_status_file, 'wb') as f: - default_data[key] = value - json.dump(default_data, f) - f.flush() - os.fsync(f.fileno()) - except: - raise + # Update last_synced_time in status file based on stime + # only update stime if stime xattr set to Brick root + if path == self.FLAT_DIR_HIERARCHY: + chkpt_time = gconf.getr("checkpoint") + checkpoint_time = 0 + if chkpt_time is not None: + checkpoint_time = int(chkpt_time) - def update_worker_cumilitive_status(self, files_pending): - default_data = {"remote_node": "N/A", - "worker status": "Not Started", - "crawl status": "N/A", - "files_syncd": 0, - "files_remaining": 0, - "bytes_remaining": 0, - "purges_remaining": 0, - "total_files_skipped": 0} - worker_status_file = self.get_worker_status_file() - try: - with open(worker_status_file, 'r+') as f: - loaded_data = json.load(f) - loaded_data['files_remaining'] = files_pending['count'] - loaded_data['bytes_remaining'] = files_pending['bytes'] - loaded_data['purges_remaining'] = files_pending['purge'] - os.ftruncate(f.fileno(), 0) - os.lseek(f.fileno(), 0, os.SEEK_SET) - json.dump(loaded_data, f) - f.flush() - os.fsync(f.fileno()) - except (IOError, OSError, ValueError): - logging.info('Creating new %s' % worker_status_file) - try: - with open(worker_status_file, 'wb') as f: - default_data['files_remaining'] = files_pending['count'] - default_data['bytes_remaining'] = files_pending['bytes'] - default_data['purges_remaining'] = files_pending['purge'] - json.dump(default_data, f) - f.flush() - os.fsync(f.fileno()) - except: - raise + self.status.set_last_synced(stime, checkpoint_time) def update_worker_remote_node(self): - node = sys.argv[-1] - node = node.split("@")[-1] - remote_node_ip = node.split(":")[0] - remote_node_vol = node.split(":")[3] - remote_node = remote_node_ip + '::' + remote_node_vol - self.update_worker_status('remote_node', remote_node) - - def update_worker_health(self, state): - self.update_worker_status('worker status', state) - - def update_worker_crawl_status(self, state): - self.update_worker_status('crawl status', state) - - def update_worker_files_syncd(self): - default_data = {"remote_node": "N/A", - "worker status": "Not Started", - "crawl status": "N/A", - "files_syncd": 0, - "files_remaining": 0, - "bytes_remaining": 0, - "purges_remaining": 0, - "total_files_skipped": 0} - worker_status_file = self.get_worker_status_file() - try: - with open(worker_status_file, 'r+') as f: - loaded_data = json.load(f) - loaded_data['files_syncd'] += loaded_data['files_remaining'] - loaded_data['files_remaining'] = 0 - loaded_data['bytes_remaining'] = 0 - loaded_data['purges_remaining'] = 0 - os.ftruncate(f.fileno(), 0) - os.lseek(f.fileno(), 0, os.SEEK_SET) - json.dump(loaded_data, f) - f.flush() - os.fsync(f.fileno()) - except (IOError, OSError, ValueError): - logging.info('Creating new %s' % worker_status_file) - try: - with open(worker_status_file, 'wb') as f: - json.dump(default_data, f) - f.flush() - os.fsync(f.fileno()) - except: - raise + node = rconf.args.resource_remote + node_data = node.split("@") + node = node_data[-1] + remote_node_ip, _ = host_brick_split(node) + self.status.set_slave_node(remote_node_ip) + + def changelogs_batch_process(self, changes): + changelogs_batches = [] + current_size = 0 + for c in changes: + si = os.lstat(c).st_size + if (si + current_size) > gconf.get("changelog-batch-size"): + # Create new batch if single Changelog file greater than + # Max Size! or current batch size exceeds Max size + changelogs_batches.append([c]) + current_size = si + else: + # Append to last batch, if No batches available Create one + current_size += si + if not changelogs_batches: + changelogs_batches.append([c]) + else: + changelogs_batches[-1].append(c) - def update_worker_files_remaining(self, state): - self.update_worker_status('files_remaining', state) - - def update_worker_bytes_remaining(self, state): - self.update_worker_status('bytes_remaining', state) - - def update_worker_purges_remaining(self, state): - self.update_worker_status('purges_remaining', state) - - def update_worker_total_files_skipped(self, value): - default_data = {"remote_node": "N/A", - "worker status": "Not Started", - "crawl status": "N/A", - "files_syncd": 0, - "files_remaining": 0, - "bytes_remaining": 0, - "purges_remaining": 0, - "total_files_skipped": 0} - worker_status_file = self.get_worker_status_file() - try: - with open(worker_status_file, 'r+') as f: - loaded_data = json.load(f) - loaded_data['total_files_skipped'] = value - loaded_data['files_remaining'] -= value - os.ftruncate(f.fileno(), 0) - os.lseek(f.fileno(), 0, os.SEEK_SET) - json.dump(loaded_data, f) - f.flush() - os.fsync(f.fileno()) - except (IOError, OSError, ValueError): - logging.info('Creating new %s' % worker_status_file) - try: - with open(worker_status_file, 'wb') as f: - default_data['total_files_skipped'] = value - json.dump(default_data, f) - f.flush() - os.fsync(f.fileno()) - except: - raise + for batch in changelogs_batches: + logging.debug(lf('processing changes', + batch=batch)) + self.process(batch) - def crawl(self, no_stime_update=False): - self.update_worker_crawl_status("Changelog Crawl") + def crawl(self): + self.status.set_worker_crawl_status("Changelog Crawl") changes = [] # get stime (from the brick) and purge changelogs # that are _historical_ to that time. - purge_time = self.get_purge_time() + data_stime = self.get_data_stime() - self.changelog_agent.scan() + libgfchangelog.scan() self.crawls += 1 - changes = self.changelog_agent.getchanges() + changes = libgfchangelog.getchanges() if changes: - if purge_time: - logging.info("slave's time: %s" % repr(purge_time)) + if data_stime: + logging.info(lf("slave's time", + stime=data_stime)) processed = [x for x in changes - if int(x.split('.')[-1]) < purge_time[0]] + if int(x.split('.')[-1]) < data_stime[0]] for pr in processed: - logging.info( - 'skipping already processed change: %s...' % - os.path.basename(pr)) + logging.debug( + lf('skipping already processed change', + changelog=os.path.basename(pr))) self.changelog_done_func(pr) changes.remove(pr) + self.archive_and_purge_changelogs(processed) - if changes: - logging.debug('processing changes %s' % repr(changes)) - self.process(changes) + self.changelogs_batch_process(changes) - def register(self, register_time, changelog_agent): - self.changelog_agent = changelog_agent - self.sleep_interval = int(gconf.change_interval) - self.changelog_done_func = self.changelog_agent.done + def register(self, register_time, status): + self.sleep_interval = gconf.get("change-interval") + self.changelog_done_func = libgfchangelog.done + self.tempdir = self.setup_working_dir() + self.processed_changelogs_dir = os.path.join(self.tempdir, + ".processed") + self.name = "live_changelog" + self.status = status class GMasterChangeloghistoryMixin(GMasterChangelogMixin): - def register(self, register_time, changelog_agent): - self.changelog_agent = changelog_agent + def register(self, register_time, status): self.changelog_register_time = register_time self.history_crawl_start_time = register_time - self.changelog_done_func = self.changelog_agent.history_done + self.changelog_done_func = libgfchangelog.history_done self.history_turns = 0 + self.tempdir = self.setup_working_dir() + self.processed_changelogs_dir = os.path.join(self.tempdir, + ".history/.processed") + self.name = "history_changelog" + self.status = status - def crawl(self, no_stime_update=False): + def crawl(self): self.history_turns += 1 - self.update_worker_crawl_status("History Crawl") - purge_time = self.get_purge_time() + self.status.set_worker_crawl_status("History Crawl") + data_stime = self.get_data_stime() + + end_time = int(time.time()) + + #as start of historical crawl marks Geo-rep worker restart + if gconf.get("ignore-deletes"): + logging.info(lf('ignore-deletes config option is set', + stime=data_stime)) - logging.info('starting history crawl... turns: %s, stime: %s' - % (self.history_turns, repr(purge_time))) + logging.info(lf('starting history crawl', + turns=self.history_turns, + stime=data_stime, + etime=end_time, + entry_stime=self.get_entry_stime())) - if not purge_time or purge_time == URXTIME: - logging.info("stime not available, abandoning history crawl") - raise NoPurgeTimeAvailable() + if not data_stime or data_stime == URXTIME: + raise NoStimeAvailable() # Changelogs backend path is hardcoded as # <BRICK_PATH>/.glusterfs/changelogs, if user configured to different # location then consuming history will not work(Known issue as of now) - changelog_path = os.path.join(gconf.local_path, + changelog_path = os.path.join(rconf.args.local_path, ".glusterfs/changelogs") - ret, actual_end = self.changelog_agent.history( + ret, actual_end = libgfchangelog.history_changelog( changelog_path, - purge_time[0], - self.changelog_register_time, - int(gconf.sync_jobs)) + data_stime[0], + end_time, + gconf.get("sync-jobs")) # scan followed by getchanges till scan returns zero. # history_scan() is blocking call, till it gets the number @@ -1246,29 +1579,30 @@ class GMasterChangeloghistoryMixin(GMasterChangelogMixin): # to be processed. returns positive value as number of changelogs # to be processed, which will be fetched using # history_getchanges() - while self.changelog_agent.history_scan() > 0: + while libgfchangelog.history_scan() > 0: self.crawls += 1 - changes = self.changelog_agent.history_getchanges() + changes = libgfchangelog.history_getchanges() if changes: - if purge_time: - logging.info("slave's time: %s" % repr(purge_time)) + if data_stime: + logging.info(lf("slave's time", + stime=data_stime)) processed = [x for x in changes - if int(x.split('.')[-1]) < purge_time[0]] + if int(x.split('.')[-1]) < data_stime[0]] for pr in processed: - logging.info('skipping already processed change: ' - '%s...' % os.path.basename(pr)) + logging.debug(lf('skipping already processed change', + changelog=os.path.basename(pr))) self.changelog_done_func(pr) changes.remove(pr) - if changes: - logging.debug('processing changes %s' % repr(changes)) - self.process(changes) + self.changelogs_batch_process(changes) history_turn_time = int(time.time()) - self.history_crawl_start_time - logging.info('finished history crawl syncing, endtime: %s, stime: %s' - % (actual_end, repr(self.get_purge_time()))) + logging.info(lf('finished history crawl', + endtime=actual_end, + stime=self.get_data_stime(), + entry_stime=self.get_entry_stime())) # If TS returned from history_changelog is < register_time # then FS crawl may be required, since history is only available @@ -1282,7 +1616,7 @@ class GMasterChangeloghistoryMixin(GMasterChangelogMixin): self.history_crawl_start_time = int(time.time()) self.crawl() else: - # This exeption will be catched in resource.py and + # This exception will be caught in resource.py and # fallback to xsync for the small gap. raise PartialHistoryAvailable(str(actual_end)) @@ -1291,7 +1625,7 @@ class GMasterXsyncMixin(GMasterChangelogMixin): """ This crawl needs to be xtime based (as of now - it's not. this is beacuse we generate CHANGELOG + it's not. this is because we generate CHANGELOG file during each crawl which is then processed by process_change()). For now it's used as a one-shot initial sync @@ -1301,14 +1635,18 @@ class GMasterXsyncMixin(GMasterChangelogMixin): XSYNC_MAX_ENTRIES = 1 << 13 - def register(self, register_time=None, changelog_agent=None): + def register(self, register_time=None, status=None): + self.status = status self.counter = 0 self.comlist = [] self.stimes = [] self.sleep_interval = 60 self.tempdir = self.setup_working_dir() + logging.info(lf('Working dir', + path=self.tempdir)) self.tempdir = os.path.join(self.tempdir, 'xsync') - logging.info('xsync temp directory: %s' % self.tempdir) + self.processed_changelogs_dir = self.tempdir + self.name = "xsync" try: os.makedirs(self.tempdir) except OSError: @@ -1317,50 +1655,52 @@ class GMasterXsyncMixin(GMasterChangelogMixin): pass else: raise + # Purge stale unprocessed xsync changelogs + for f in os.listdir(self.tempdir): + if f.startswith("XSYNC-CHANGELOG"): + os.remove(os.path.join(self.tempdir, f)) + - def crawl(self, no_stime_update=False): + def crawl(self): """ event dispatcher thread this thread dispatches either changelog or synchronizes stime. - additionally terminates itself on recieving a 'finale' event + additionally terminates itself on receiving a 'finale' event """ def Xsyncer(): self.Xcrawl() t = Thread(target=Xsyncer) t.start() - logging.info('starting hybrid crawl..., stime: %s' - % repr(self.get_purge_time())) - self.update_worker_crawl_status("Hybrid Crawl") + logging.info(lf('starting hybrid crawl', + stime=self.get_data_stime())) + self.status.set_worker_crawl_status("Hybrid Crawl") while True: try: item = self.comlist.pop(0) if item[0] == 'finale': - logging.info('finished hybrid crawl syncing, stime: %s' - % repr(self.get_purge_time())) + logging.info(lf('finished hybrid crawl', + stime=self.get_data_stime())) break elif item[0] == 'xsync': - logging.info('processing xsync changelog %s' % (item[1])) + logging.info(lf('processing xsync changelog', + path=item[1])) self.process([item[1]], 0) + self.archive_and_purge_changelogs([item[1]]) elif item[0] == 'stime': - if not no_stime_update: - # xsync is started after running history but if - # history actual end time is less than register time - # then if we update stime, live changelog processing - # will skip the changelogs for which TS is less than - # stime. During this deletes and renames are not - # propogated. By not setting stime live changelog will - # start processing from the register time. Since we - # have xsync_upper_limit their will not be much - # overlap/redo of changelogs. - logging.debug('setting slave time: %s' % repr(item[1])) - self.upd_stime(item[1][1], item[1][0]) + logging.debug(lf('setting slave time', + time=item[1])) + self.upd_stime(item[1][1], item[1][0]) else: - logging.warn('unknown tuple in comlist (%s)' % repr(item)) + logging.warn(lf('unknown tuple in comlist', + entry=item)) except IndexError: time.sleep(1) def write_entry_change(self, prefix, data=[]): + if not getattr(self, "fh", None): + self.open() + self.fh.write("%s %s\n" % (prefix, ' '.join(data))) def open(self): @@ -1372,7 +1712,11 @@ class GMasterXsyncMixin(GMasterChangelogMixin): raise def close(self): - self.fh.close() + if getattr(self, "fh", None): + self.fh.flush() + os.fsync(self.fh.fileno()) + self.fh.close() + self.fh = None def fname(self): return self.xsync_change @@ -1383,11 +1727,11 @@ class GMasterXsyncMixin(GMasterChangelogMixin): def sync_xsync(self, last): """schedule a processing of changelog""" self.close() - self.put('xsync', self.fname()) + if self.counter > 0: + self.put('xsync', self.fname()) self.counter = 0 if not last: time.sleep(1) # make sure changelogs are 1 second apart - self.open() def sync_stime(self, stime=None, last=False): """schedule a stime synchronization""" @@ -1409,7 +1753,7 @@ class GMasterXsyncMixin(GMasterChangelogMixin): def is_sticky(self, path, mo): """check for DHTs linkto sticky bit file""" sticky = False - if mo & 01000: + if mo & 0o1000: sticky = self.master.server.linkto_check(path) return sticky @@ -1421,15 +1765,15 @@ class GMasterXsyncMixin(GMasterChangelogMixin): the filesystem tree, but set after directory synchronization. """ if path == '.': - self.open() self.crawls += 1 if not xtr_root: # get the root stime and use it for all comparisons xtr_root = self.xtime('.', self.slave) if isinstance(xtr_root, int): if xtr_root != ENOENT: - logging.warn("slave cluster not returning the " - "correct xtime for root (%d)" % xtr_root) + logging.warn(lf("slave cluster not returning the " + "xtime for root", + error=xtr_root)) xtr_root = self.minus_infinity xtl = self.xtime(path) if isinstance(xtl, int): @@ -1437,10 +1781,15 @@ class GMasterXsyncMixin(GMasterChangelogMixin): xtr = self.xtime(path, self.slave) if isinstance(xtr, int): if xtr != ENOENT: - logging.warn("slave cluster not returning the " - "correct xtime for %s (%d)" % (path, xtr)) + logging.warn(lf("slave cluster not returning the " + "xtime for dir", + path=path, + error=xtr)) xtr = self.minus_infinity xtr = max(xtr, xtr_root) + zero_zero = (0, 0) + if xtr_root == zero_zero: + xtr = self.minus_infinity if not self.need_sync(path, xtl, xtr): if path == '.': self.sync_done([(path, xtl)], True) @@ -1450,48 +1799,65 @@ class GMasterXsyncMixin(GMasterChangelogMixin): dem = self.master.server.entries(path) pargfid = self.master.server.gfid(path) if isinstance(pargfid, int): - logging.warn('skipping directory %s' % (path)) + logging.warn(lf('skipping directory', + path=path)) for e in dem: bname = e e = os.path.join(path, e) xte = self.xtime(e) if isinstance(xte, int): - logging.warn("irregular xtime for %s: %s" % - (e, errno.errorcode[xte])) + logging.warn(lf("irregular xtime", + path=e, + error=errno.errorcode[xte])) continue if not self.need_sync(e, xte, xtr): continue st = self.master.server.lstat(e) if isinstance(st, int): - logging.warn('%s got purged in the interim ...' % e) + logging.warn(lf('got purged in the interim', + path=e)) continue if self.is_sticky(e, st.st_mode): - logging.debug('ignoring sticky bit file %s' % e) + logging.debug(lf('ignoring sticky bit file', + path=e)) continue gfid = self.master.server.gfid(e) if isinstance(gfid, int): - logging.warn('skipping entry %s..' % e) + logging.warn(lf('skipping entry', + path=e)) continue mo = st.st_mode - self.counter += 1 + self.counter += 1 if ((stat.S_ISDIR(mo) or + stat.S_ISLNK(mo) or + stat.S_ISREG(mo))) else 0 if self.counter == self.XSYNC_MAX_ENTRIES: self.sync_done(self.stimes, False) self.stimes = [] if stat.S_ISDIR(mo): - self.write_entry_change("E", [gfid, 'MKDIR', str(mo), str( - st.st_uid), str(st.st_gid), escape(os.path.join(pargfid, - bname))]) + self.write_entry_change("E", + [gfid, 'MKDIR', str(mo), + str(0), str(0), escape_space_newline( + os.path.join(pargfid, bname))]) self.write_entry_change("M", [gfid, "SETATTR", str(st.st_uid), - str(st.st_gid), str(st.st_mode)]) + str(st.st_gid), str(st.st_mode), + str(st.st_atime), + str(st.st_mtime)]) self.Xcrawl(e, xtr_root) stime_to_update = xte - if self.xsync_upper_limit: - stime_to_update = min(self.xsync_upper_limit, xte) + # Live Changelog Start time indicates that from that time + # onwards Live changelogs are available. If we update stime + # greater than live_changelog_start time then Geo-rep will + # skip those changelogs as already processed. But Xsync + # actually failed to sync the deletes and Renames. Update + # stime as min(Live_changelogs_time, Actual_stime) When it + # switches to Changelog mode, it syncs Deletes and Renames. + if self.live_changelog_start_time: + stime_to_update = min(self.live_changelog_start_time, xte) self.stimes.append((e, stime_to_update)) elif stat.S_ISLNK(mo): self.write_entry_change( - "E", [gfid, 'SYMLINK', escape(os.path.join(pargfid, - bname))]) + "E", [gfid, 'SYMLINK', escape_space_newline( + os.path.join(pargfid, bname))]) elif stat.S_ISREG(mo): nlink = st.st_nlink nlink -= 1 # fixup backend stat link count @@ -1501,19 +1867,19 @@ class GMasterXsyncMixin(GMasterChangelogMixin): if nlink == 1: self.write_entry_change("E", [gfid, 'MKNOD', str(mo), - str(st.st_uid), - str(st.st_gid), - escape(os.path.join( - pargfid, bname))]) + str(0), str(0), + escape_space_newline( + os.path.join( + pargfid, bname))]) else: self.write_entry_change( - "E", [gfid, 'LINK', escape(os.path.join(pargfid, - bname))]) + "E", [gfid, 'LINK', escape_space_newline( + os.path.join(pargfid, bname))]) self.write_entry_change("D", [gfid]) if path == '.': stime_to_update = xtl - if self.xsync_upper_limit: - stime_to_update = min(self.xsync_upper_limit, xtl) + if self.live_changelog_start_time: + stime_to_update = min(self.live_changelog_start_time, xtl) self.stimes.append((path, stime_to_update)) self.sync_done(self.stimes, True) @@ -1599,16 +1965,17 @@ class Syncer(object): def __init__(self, slave, sync_engine, resilient_errnos=[]): """spawn worker threads""" + self.log_err = False self.slave = slave self.lock = Lock() self.pb = PostBox() self.sync_engine = sync_engine self.errnos_ok = resilient_errnos - for i in range(int(gconf.sync_jobs)): - t = Thread(target=self.syncjob) + for i in range(gconf.get("sync-jobs")): + t = Thread(target=self.syncjob, args=(i + 1, )) t.start() - def syncjob(self): + def syncjob(self, job_id): """the life of a worker""" while True: pb = None @@ -1621,7 +1988,14 @@ class Syncer(object): break time.sleep(0.5) pb.close() - po = self.sync_engine(pb) + start = time.time() + po = self.sync_engine(pb, self.log_err) + logging.info(lf("Sync Time Taken", + job=job_id, + num_files=len(pb), + return_code=po.returncode, + duration="%.4f" % (time.time() - start))) + if po.returncode == 0: ret = (True, 0) elif po.returncode in self.errnos_ok: @@ -1638,3 +2012,9 @@ class Syncer(object): return pb except BoxClosedErr: pass + + def enable_errorlog(self): + self.log_err = True + + def disable_errorlog(self): + self.log_err = False diff --git a/geo-replication/syncdaemon/monitor.py b/geo-replication/syncdaemon/monitor.py index 3e0360332bd..6aa7b9dfc99 100644 --- a/geo-replication/syncdaemon/monitor.py +++ b/geo-replication/syncdaemon/monitor.py @@ -13,91 +13,49 @@ import sys import time import signal import logging -import uuid import xml.etree.ElementTree as XET -from subprocess import PIPE -from resource import Popen, FILE, GLUSTER, SSH from threading import Lock -import re +from errno import ECHILD, ESRCH import random -from gconf import gconf -from syncdutils import update_file, select, waitpid -from syncdutils import set_term_handler, is_host_local, GsyncdError -from syncdutils import escape, Thread, finalize, memoize +from resource import SSH +import gsyncdconfig as gconf +import libgfchangelog +from rconf import rconf +from syncdutils import (select, waitpid, errno_wrap, lf, grabpidfile, + set_term_handler, GsyncdError, + Thread, finalize, Volinfo, VolinfoFromGconf, + gf_event, EVENT_GEOREP_FAULTY, get_up_nodes, + unshare_propagation_supported) +from gsyncdstatus import GeorepStatus, set_monitor_status +import py2py3 +from py2py3 import pipe ParseError = XET.ParseError if hasattr(XET, 'ParseError') else SyntaxError -def get_slave_bricks_status(host, vol): - po = Popen(['gluster', '--xml', '--remote-host=' + host, - 'volume', 'status', vol, "detail"], - stdout=PIPE, stderr=PIPE) - vix = po.stdout.read() - po.wait() - po.terminate_geterr() - vi = XET.fromstring(vix) - if vi.find('opRet').text != '0': - logging.info("Unable to get list of up nodes of %s, " - "returning empty list: %s" % - (vol, vi.find('opErrstr').text)) - return [] - - up_hosts = set() - - try: - for el in vi.findall('volStatus/volumes/volume/node'): - if el.find('status').text == '1': - up_hosts.add(el.find('hostname').text) - except (ParseError, AttributeError, ValueError) as e: - logging.info("Parsing failed to get list of up nodes of %s, " - "returning empty list: %s" % (vol, e)) - - return list(up_hosts) - - -class Volinfo(object): - - def __init__(self, vol, host='localhost', prelude=[]): - po = Popen(prelude + ['gluster', '--xml', '--remote-host=' + host, - 'volume', 'info', vol], - stdout=PIPE, stderr=PIPE) - vix = po.stdout.read() - po.wait() - po.terminate_geterr() - vi = XET.fromstring(vix) - if vi.find('opRet').text != '0': - if prelude: - via = '(via %s) ' % prelude.join(' ') - else: - via = ' ' - raise GsyncdError('getting volume info of %s%s ' - 'failed with errorcode %s', - (vol, via, vi.find('opErrno').text)) - self.tree = vi - self.volume = vol - self.host = host - - def get(self, elem): - return self.tree.findall('.//' + elem) - - @property - @memoize - def bricks(self): - def bparse(b): - host, dirp = b.text.split(':', 2) - return {'host': host, 'dir': dirp} - return [bparse(b) for b in self.get('brick')] - - @property - @memoize - def uuid(self): - ids = self.get('id') - if len(ids) != 1: - raise GsyncdError("volume info of %s obtained from %s: " - "ambiguous uuid", - self.volume, self.host) - return ids[0].text +def get_subvol_num(brick_idx, vol, hot): + tier = vol.is_tier() + disperse_count = vol.disperse_count(tier, hot) + replica_count = vol.replica_count(tier, hot) + distribute_count = vol.distribution_count(tier, hot) + gconf.setconfig("master-distribution-count", distribute_count) + + if (tier and not hot): + brick_idx = brick_idx - vol.get_hot_bricks_count(tier) + + subvol_size = disperse_count if disperse_count > 0 else replica_count + cnt = int((brick_idx + 1) / subvol_size) + rem = (brick_idx + 1) % subvol_size + if rem > 0: + cnt = cnt + 1 + + if (tier and hot): + return "hot_" + str(cnt) + elif (tier and not hot): + return "cold_" + str(cnt) + else: + return str(cnt) class Monitor(object): @@ -105,41 +63,16 @@ class Monitor(object): """class which spawns and manages gsyncd workers""" ST_INIT = 'Initializing...' - ST_STABLE = 'Stable' - ST_FAULTY = 'faulty' + ST_STARTED = 'Started' + ST_STABLE = 'Active' + ST_FAULTY = 'Faulty' ST_INCON = 'inconsistent' _ST_ORD = [ST_STABLE, ST_INIT, ST_FAULTY, ST_INCON] def __init__(self): self.lock = Lock() self.state = {} - - def set_state(self, state, w=None): - """set the state that can be used by external agents - like glusterd for status reporting""" - computestate = lambda: self.state and self._ST_ORD[ - max(self._ST_ORD.index(s) for s in self.state.values())] - if w: - self.lock.acquire() - old_state = computestate() - self.state[w] = state - state = computestate() - self.lock.release() - if state != old_state: - self.set_state(state) - else: - if getattr(gconf, 'state_file', None): - # If previous state is paused, suffix the - # new state with '(Paused)' - try: - with open(gconf.state_file, "r") as f: - content = f.read() - if "paused" in content.lower(): - state = state + '(Paused)' - except IOError: - pass - logging.info('new state: %s' % state) - update_file(gconf.state_file, lambda f: f.write(state + '\n')) + self.status = {} @staticmethod def terminate(): @@ -147,9 +80,10 @@ class Monitor(object): # standard handler set_term_handler(lambda *a: set_term_handler()) # give a chance to graceful exit - os.kill(-os.getpid(), signal.SIGTERM) + errno_wrap(os.kill, [-os.getpid(), signal.SIGTERM], [ESRCH]) - def monitor(self, w, argv, cpids, agents, slave_vol, slave_host): + def monitor(self, w, argv, cpids, slave_vol, slave_host, master, + suuid, slavenodes): """the monitor loop Basic logic is a blantantly simple blunt heuristics: @@ -168,19 +102,32 @@ class Monitor(object): blown worker blows up on EPIPE if the net goes down, due to the keep-alive thread) """ - - self.set_state(self.ST_INIT, w) - + if not self.status.get(w[0]['dir'], None): + self.status[w[0]['dir']] = GeorepStatus(gconf.get("state-file"), + w[0]['host'], + w[0]['dir'], + w[0]['uuid'], + master, + "%s::%s" % (slave_host, + slave_vol)) ret = 0 def nwait(p, o=0): - p2, r = waitpid(p, o) - if not p2: - return - return r + try: + p2, r = waitpid(p, o) + if not p2: + return + return r + except OSError as e: + # no child process, this happens if the child process + # already died and has been cleaned up + if e.errno == ECHILD: + return -1 + else: + raise def exit_signalled(s): - """ child teminated due to receipt of SIGUSR1 """ + """ child terminated due to receipt of SIGUSR1 """ return (os.WIFSIGNALED(s) and (os.WTERMSIG(s) == signal.SIGUSR1)) def exit_status(s): @@ -188,72 +135,76 @@ class Monitor(object): return os.WEXITSTATUS(s) return 1 - conn_timeout = int(gconf.connection_timeout) + conn_timeout = gconf.get("connection-timeout") while ret in (0, 1): - remote_host = w[1] + remote_user, remote_host = w[1][0].split("@") + remote_id = w[1][1] # Check the status of the connected slave node # If the connected slave node is down then try to connect to # different up node. - m = re.match("(ssh|gluster|file):\/\/(.+)@([^:]+):(.+)", - remote_host) - if m: - current_slave_host = m.group(3) - slave_up_hosts = get_slave_bricks_status( - slave_host, slave_vol) - - if current_slave_host not in slave_up_hosts: - if len(slave_up_hosts) > 0: - remote_host = "%s://%s@%s:%s" % (m.group(1), - m.group(2), - random.choice( - slave_up_hosts), - m.group(4)) - - # Spawn the worker and agent in lock to avoid fd leak + current_slave_host = remote_host + slave_up_hosts = get_up_nodes(slavenodes, gconf.get("ssh-port")) + + if (current_slave_host, remote_id) not in slave_up_hosts: + if len(slave_up_hosts) > 0: + remote_new = random.choice(slave_up_hosts) + remote_host = "%s@%s" % (remote_user, remote_new[0]) + remote_id = remote_new[1] + + # Spawn the worker in lock to avoid fd leak self.lock.acquire() - logging.info('-' * conn_timeout) - logging.info('starting gsyncd worker') - - # Couple of pipe pairs for RPC communication b/w - # worker and changelog agent. - - # read/write end for agent - (ra, ww) = os.pipe() - # read/write end for worker - (rw, wa) = os.pipe() - - # spawn the agent process - apid = os.fork() - if apid == 0: - os.execv(sys.executable, argv + ['--local-path', w[0], - '--agent', - '--rpc-fd', - ','.join([str(ra), str(wa), - str(rw), str(ww)])]) - pr, pw = os.pipe() + self.status[w[0]['dir']].set_worker_status(self.ST_INIT) + logging.info(lf('starting gsyncd worker', + brick=w[0]['dir'], + slave_node=remote_host)) + + pr, pw = pipe() cpid = os.fork() if cpid == 0: os.close(pr) - os.execv(sys.executable, argv + ['--feedback-fd', str(pw), - '--local-path', w[0], - '--local-id', - '.' + escape(w[0]), - '--rpc-fd', - ','.join([str(rw), str(ww), - str(ra), str(wa)]), - '--resource-remote', - remote_host]) + + args_to_worker = argv + [ + 'worker', + rconf.args.master, + rconf.args.slave, + '--feedback-fd', str(pw), + '--local-path', w[0]['dir'], + '--local-node', w[0]['host'], + '--local-node-id', w[0]['uuid'], + '--slave-id', suuid, + '--subvol-num', str(w[2]), + '--resource-remote', remote_host, + '--resource-remote-id', remote_id + ] + + if rconf.args.config_file is not None: + args_to_worker += ['-c', rconf.args.config_file] + + if w[3]: + args_to_worker.append("--is-hottier") + + if rconf.args.debug: + args_to_worker.append("--debug") + + access_mount = gconf.get("access-mount") + if access_mount: + os.execv(sys.executable, args_to_worker) + else: + if unshare_propagation_supported(): + logging.debug("Worker would mount volume privately") + unshare_cmd = ['unshare', '-m', '--propagation', + 'private'] + cmd = unshare_cmd + args_to_worker + os.execvp("unshare", cmd) + else: + logging.debug("Mount is not private. It would be lazy" + " umounted") + os.execv(sys.executable, args_to_worker) cpids.add(cpid) - agents.add(apid) os.close(pw) - # close all RPC pipes in monitor - os.close(ra) - os.close(wa) - os.close(rw) - os.close(ww) self.lock.release() t0 = time.time() @@ -262,126 +213,183 @@ class Monitor(object): if so: ret = nwait(cpid, os.WNOHANG) + if ret is not None: - logging.info("worker(%s) died before establishing " - "connection" % w[0]) - nwait(apid) #wait for agent + logging.info(lf("worker died before establishing " + "connection", + brick=w[0]['dir'])) else: - logging.debug("worker(%s) connected" % w[0]) + logging.debug("worker(%s) connected" % w[0]['dir']) while time.time() < t0 + conn_timeout: ret = nwait(cpid, os.WNOHANG) + if ret is not None: - logging.info("worker(%s) died in startup " - "phase" % w[0]) - nwait(apid) #wait for agent + logging.info(lf("worker died in startup phase", + brick=w[0]['dir'])) break + time.sleep(1) else: - logging.info("worker(%s) not confirmed in %d sec, " - "aborting it" % (w[0], conn_timeout)) - os.kill(cpid, signal.SIGKILL) - nwait(apid) #wait for agent + logging.info( + lf("Worker not confirmed after wait, aborting it. " + "Gsyncd invocation on remote slave via SSH or " + "gluster master mount might have hung. Please " + "check the above logs for exact issue and check " + "master or slave volume for errors. Restarting " + "master/slave volume accordingly might help.", + brick=w[0]['dir'], + timeout=conn_timeout)) + errno_wrap(os.kill, [cpid, signal.SIGKILL], [ESRCH]) ret = nwait(cpid) if ret is None: - self.set_state(self.ST_STABLE, w) - #If worker dies, agent terminates on EOF. - #So lets wait for agent first. - nwait(apid) ret = nwait(cpid) if exit_signalled(ret): ret = 0 else: ret = exit_status(ret) if ret in (0, 1): - self.set_state(self.ST_FAULTY, w) + self.status[w[0]['dir']].set_worker_status(self.ST_FAULTY) + gf_event(EVENT_GEOREP_FAULTY, + master_volume=master.volume, + master_node=w[0]['host'], + master_node_id=w[0]['uuid'], + slave_host=slave_host, + slave_volume=slave_vol, + current_slave_host=current_slave_host, + brick_path=w[0]['dir']) time.sleep(10) - self.set_state(self.ST_INCON, w) + self.status[w[0]['dir']].set_worker_status(self.ST_INCON) return ret - def multiplex(self, wspx, suuid, slave_vol, slave_host): - argv = sys.argv[:] - for o in ('-N', '--no-daemon', '--monitor'): - while o in argv: - argv.remove(o) - argv.extend(('-N', '-p', '', '--slave-id', suuid)) - argv.insert(0, os.path.basename(sys.executable)) + def multiplex(self, wspx, suuid, slave_vol, slave_host, master, slavenodes): + argv = [os.path.basename(sys.executable), sys.argv[0]] cpids = set() - agents = set() ta = [] for wx in wspx: def wmon(w): - cpid, _ = self.monitor(w, argv, cpids, agents, slave_vol, - slave_host) + cpid, _ = self.monitor(w, argv, cpids, slave_vol, + slave_host, master, suuid, slavenodes) time.sleep(1) self.lock.acquire() for cpid in cpids: - os.kill(cpid, signal.SIGKILL) - for apid in agents: - os.kill(apid, signal.SIGKILL) + errno_wrap(os.kill, [cpid, signal.SIGKILL], [ESRCH]) self.lock.release() finalize(exval=1) t = Thread(target=wmon, args=[wx]) t.start() ta.append(t) + + # monitor status was being updated in each monitor thread. It + # should not be done as it can cause deadlock for a worker start. + # set_monitor_status uses flock to synchronize multple instances + # updating the file. Since each monitor thread forks worker, + # these processes can hold the reference to fd of status + # file causing deadlock to workers which starts later as flock + # will not be release until all references to same fd is closed. + # It will also cause fd leaks. + + self.lock.acquire() + set_monitor_status(gconf.get("state-file"), self.ST_STARTED) + self.lock.release() for t in ta: t.join() -def distribute(*resources): - master, slave = resources - mvol = Volinfo(master.volume, master.host) +def distribute(master, slave): + if rconf.args.use_gconf_volinfo: + mvol = VolinfoFromGconf(master.volume, master=True) + else: + mvol = Volinfo(master.volume, master.host, master=True) logging.debug('master bricks: ' + repr(mvol.bricks)) prelude = [] - si = slave slave_host = None slave_vol = None - if isinstance(slave, SSH): - prelude = gconf.ssh_command.split() + [slave.remote_addr] - si = slave.inner_rsc - logging.debug('slave SSH gateway: ' + slave.remote_addr) - if isinstance(si, FILE): - sbricks = {'host': 'localhost', 'dir': si.path} - suuid = uuid.uuid5(uuid.NAMESPACE_URL, slave.get_url(canonical=True)) - elif isinstance(si, GLUSTER): - svol = Volinfo(si.volume, slave.remote_addr.split('@')[-1]) - sbricks = svol.bricks - suuid = svol.uuid - slave_host = slave.remote_addr.split('@')[-1] - slave_vol = si.volume - else: - raise GsyncdError("unkown slave type " + slave.url) - logging.info('slave bricks: ' + repr(sbricks)) - if isinstance(si, FILE): - slaves = [slave.url] - else: - slavenodes = set(b['host'] for b in sbricks) - if isinstance(slave, SSH) and not gconf.isolated_slave: - rap = SSH.parse_ssh_address(slave) - slaves = ['ssh://' + rap['user'] + '@' + h + ':' + si.url - for h in slavenodes] - else: - slavevols = [h + ':' + si.volume for h in slavenodes] - if isinstance(slave, SSH): - slaves = ['ssh://' + rap.remote_addr + ':' + v - for v in slavevols] - else: - slaves = slavevols - - workerspex = [(brick['dir'], slaves[idx % len(slaves)]) - for idx, brick in enumerate(mvol.bricks) - if is_host_local(brick['host'])] - logging.info('worker specs: ' + repr(workerspex)) - return workerspex, suuid, slave_vol, slave_host + prelude = [gconf.get("ssh-command")] + \ + gconf.get("ssh-options").split() + \ + ["-p", str(gconf.get("ssh-port"))] + \ + [slave.remote_addr] + logging.debug('slave SSH gateway: ' + slave.remote_addr) -def monitor(*resources): + if rconf.args.use_gconf_volinfo: + svol = VolinfoFromGconf(slave.volume, master=False) + else: + svol = Volinfo(slave.volume, "localhost", prelude, master=False) + + sbricks = svol.bricks + suuid = svol.uuid + slave_host = slave.remote_addr.split('@')[-1] + slave_vol = slave.volume + + # save this xattr for the session delete command + old_stime_xattr_prefix = gconf.get("stime-xattr-prefix", None) + new_stime_xattr_prefix = "trusted.glusterfs." + mvol.uuid + "." + \ + svol.uuid + if not old_stime_xattr_prefix or \ + old_stime_xattr_prefix != new_stime_xattr_prefix: + gconf.setconfig("stime-xattr-prefix", new_stime_xattr_prefix) + + logging.debug('slave bricks: ' + repr(sbricks)) + + slavenodes = set((b['host'], b["uuid"]) for b in sbricks) + rap = SSH.parse_ssh_address(slave) + slaves = [(rap['user'] + '@' + h[0], h[1]) for h in slavenodes] + + workerspex = [] + for idx, brick in enumerate(mvol.bricks): + if rconf.args.local_node_id == brick['uuid']: + is_hot = mvol.is_hot(":".join([brick['host'], brick['dir']])) + workerspex.append((brick, + slaves[idx % len(slaves)], + get_subvol_num(idx, mvol, is_hot), + is_hot)) + logging.debug('worker specs: ' + repr(workerspex)) + return workerspex, suuid, slave_vol, slave_host, master, slavenodes + + +def monitor(local, remote): # Check if gsyncd restarted in pause state. If # yes, send SIGSTOP to negative of monitor pid # to go back to pause state. - if gconf.pause_on_start: - os.kill(-os.getpid(), signal.SIGSTOP) + if rconf.args.pause_on_start: + errno_wrap(os.kill, [-os.getpid(), signal.SIGSTOP], [ESRCH]) """oh yeah, actually Monitor is used as singleton, too""" - return Monitor().multiplex(*distribute(*resources)) + return Monitor().multiplex(*distribute(local, remote)) + + +def startup(go_daemon=True): + """set up logging, pidfile grabbing, daemonization""" + pid_file = gconf.get("pid-file") + if not grabpidfile(): + sys.stderr.write("pidfile is taken, exiting.\n") + sys.exit(2) + rconf.pid_file_owned = True + + if not go_daemon: + return + + x, y = pipe() + cpid = os.fork() + if cpid: + os.close(x) + sys.exit() + os.close(y) + os.setsid() + dn = os.open(os.devnull, os.O_RDWR) + for f in (sys.stdin, sys.stdout, sys.stderr): + os.dup2(dn, f.fileno()) + + if not grabpidfile(pid_file + '.tmp'): + raise GsyncdError("cannot grab temporary pidfile") + + os.rename(pid_file + '.tmp', pid_file) + + # wait for parent to terminate + # so we can start up with + # no messing from the dirty + # ol' bustard + select((x,), (), ()) + os.close(x) diff --git a/geo-replication/syncdaemon/py2py3.py b/geo-replication/syncdaemon/py2py3.py new file mode 100644 index 00000000000..f9c76e1b50a --- /dev/null +++ b/geo-replication/syncdaemon/py2py3.py @@ -0,0 +1,184 @@ +# +# Copyright (c) 2018 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. + +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +# All python2/python3 compatibility routines + +import sys +import os +import stat +import struct +from syncdutils import umask +from ctypes import create_string_buffer + +if sys.version_info >= (3,): + def pipe(): + (r, w) = os.pipe() + os.set_inheritable(r, True) + os.set_inheritable(w, True) + return (r, w) + + # Raw conversion of bytearray to string. Used in the cases where + # buffer is created by create_string_buffer which is a 8-bit char + # array and passed to syscalls to fetch results. Using encode/decode + # doesn't work as it converts to string altering the size. + def bytearray_to_str(byte_arr): + return ''.join([chr(b) for b in byte_arr]) + + # Raw conversion of string to bytes. This is required to convert + # back the string into bytearray(c char array) to use in struc + # pack/unpacking. Again encode/decode can't be used as it + # converts it alters size. + def str_to_bytearray(string): + return bytes([ord(c) for c in string]) + + def gr_create_string_buffer(size): + return create_string_buffer(b'\0', size) + + def gr_query_xattr(cls, path, size, syscall, attr=None): + if attr: + return cls._query_xattr(path.encode(), size, syscall, + attr.encode()) + else: + return cls._query_xattr(path.encode(), size, syscall) + + def gr_lsetxattr(cls, path, attr, val): + return cls.libc.lsetxattr(path.encode(), attr.encode(), val, + len(val), 0) + + def gr_lremovexattr(cls, path, attr): + return cls.libc.lremovexattr(path.encode(), attr.encode()) + + def gr_cl_register(libgfapi, brick, path, log_file, log_level, retries): + return libgfapi.gf_changelog_register(brick.encode(), + path.encode(), + log_file.encode(), + log_level, retries) + + def gr_cl_done(libgfapi, clfile): + return libgfapi.gf_changelog_done(clfile.encode()) + + def gr_cl_history_changelog(libgfapi, changelog_path, start, end, num_parallel, + actual_end): + return libgfapi.gf_history_changelog(changelog_path.encode(), + start, end, num_parallel, + actual_end) + + def gr_cl_history_done(libgfapi, clfile): + return libgfapi.gf_history_changelog_done(clfile.encode()) + + # regular file + + def entry_pack_reg(cls, gf, bn, mo, uid, gid): + bn_encoded = bn.encode() + blen = len(bn_encoded) + return struct.pack(cls._fmt_mknod(blen), + uid, gid, gf.encode(), mo, bn_encoded, + stat.S_IMODE(mo), 0, umask()) + + def entry_pack_reg_stat(cls, gf, bn, st): + bn_encoded = bn.encode() + blen = len(bn_encoded) + mo = st['mode'] + return struct.pack(cls._fmt_mknod(blen), + st['uid'], st['gid'], + gf.encode(), mo, bn_encoded, + stat.S_IMODE(mo), 0, umask()) + # mkdir + + def entry_pack_mkdir(cls, gf, bn, mo, uid, gid): + bn_encoded = bn.encode() + blen = len(bn_encoded) + return struct.pack(cls._fmt_mkdir(blen), + uid, gid, gf.encode(), mo, bn_encoded, + stat.S_IMODE(mo), umask()) + # symlink + + def entry_pack_symlink(cls, gf, bn, lnk, st): + bn_encoded = bn.encode() + blen = len(bn_encoded) + lnk_encoded = lnk.encode() + llen = len(lnk_encoded) + return struct.pack(cls._fmt_symlink(blen, llen), + st['uid'], st['gid'], + gf.encode(), st['mode'], bn_encoded, + lnk_encoded) +else: + def pipe(): + (r, w) = os.pipe() + return (r, w) + + # Raw conversion of bytearray to string + def bytearray_to_str(byte_arr): + return byte_arr + + # Raw conversion of string to bytearray + def str_to_bytearray(string): + return string + + def gr_create_string_buffer(size): + return create_string_buffer('\0', size) + + def gr_query_xattr(cls, path, size, syscall, attr=None): + if attr: + return cls._query_xattr(path, size, syscall, attr) + else: + return cls._query_xattr(path, size, syscall) + + def gr_lsetxattr(cls, path, attr, val): + return cls.libc.lsetxattr(path, attr, val, len(val), 0) + + def gr_lremovexattr(cls, path, attr): + return cls.libc.lremovexattr(path, attr) + + def gr_cl_register(libgfapi, brick, path, log_file, log_level, retries): + return libgfapi.gf_changelog_register(brick, path, log_file, + log_level, retries) + + def gr_cl_done(libgfapi, clfile): + return libgfapi.gf_changelog_done(clfile) + + def gr_cl_history_changelog(libgfapi, changelog_path, start, end, num_parallel, + actual_end): + return libgfapi.gf_history_changelog(changelog_path, start, end, + num_parallel, actual_end) + + def gr_cl_history_done(libgfapi, clfile): + return libgfapi.gf_history_changelog_done(clfile) + + # regular file + + def entry_pack_reg(cls, gf, bn, mo, uid, gid): + blen = len(bn) + return struct.pack(cls._fmt_mknod(blen), + uid, gid, gf, mo, bn, + stat.S_IMODE(mo), 0, umask()) + + def entry_pack_reg_stat(cls, gf, bn, st): + blen = len(bn) + mo = st['mode'] + return struct.pack(cls._fmt_mknod(blen), + st['uid'], st['gid'], + gf, mo, bn, + stat.S_IMODE(mo), 0, umask()) + # mkdir + + def entry_pack_mkdir(cls, gf, bn, mo, uid, gid): + blen = len(bn) + return struct.pack(cls._fmt_mkdir(blen), + uid, gid, gf, mo, bn, + stat.S_IMODE(mo), umask()) + # symlink + + def entry_pack_symlink(cls, gf, bn, lnk, st): + blen = len(bn) + llen = len(lnk) + return struct.pack(cls._fmt_symlink(blen, llen), + st['uid'], st['gid'], + gf, st['mode'], bn, lnk) diff --git a/geo-replication/syncdaemon/gconf.py b/geo-replication/syncdaemon/rconf.py index 1fc7c381bc4..ff716ee4d6d 100644 --- a/geo-replication/syncdaemon/gconf.py +++ b/geo-replication/syncdaemon/rconf.py @@ -9,9 +9,9 @@ # -class GConf(object): +class RConf(object): - """singleton class to store globals + """singleton class to store runtime globals shared between gsyncd modules""" ssh_ctl_dir = None @@ -21,5 +21,11 @@ class GConf(object): log_exit = False permanent_handles = [] log_metadata = {} + mgmt_lock_fd = None + args = None + turns = 0 + mountbroker = False + mount_point = None + mbr_umount_cmd = [] -gconf = GConf() +rconf = RConf() diff --git a/geo-replication/syncdaemon/repce.py b/geo-replication/syncdaemon/repce.py index d7b17dda796..c622afa6373 100644 --- a/geo-replication/syncdaemon/repce.py +++ b/geo-replication/syncdaemon/repce.py @@ -14,30 +14,27 @@ import time import logging from threading import Condition try: - import thread -except ImportError: - # py 3 import _thread as thread -try: - from Queue import Queue except ImportError: - # py 3 + import thread +try: from queue import Queue +except ImportError: + from Queue import Queue try: import cPickle as pickle except ImportError: - # py 3 import pickle -from syncdutils import Thread, select +from syncdutils import Thread, select, lf -pickle_proto = -1 +pickle_proto = 2 repce_version = 1.0 def ioparse(i, o): if isinstance(i, int): - i = os.fdopen(i) + i = os.fdopen(i, 'rb') # rely on duck typing for recognizing # streams as that works uniformly # in py2 and py3 @@ -57,8 +54,15 @@ def send(out, *args): def recv(inf): - """load an object from input stream""" - return pickle.load(inf) + """load an object from input stream + python2 and python3 compatibility, inf is sys.stdin + and is opened as text stream by default. Hence using the + buffer attribute in python3 + """ + if hasattr(inf, "buffer"): + return pickle.load(inf.buffer) + else: + return pickle.load(inf) class RepceServer(object): @@ -196,15 +200,17 @@ class RepceClient(object): """RePCe client is callabe, calling it implements a synchronous remote call. - We do a .push with a cbk which does a wakeup upon receiving anwser, + We do a .push with a cbk which does a wakeup upon receiving answer, then wait on the RepceJob. """ rjob = self.push( meth, *args, **{'cbk': lambda rj, res: rj.wakeup(res)}) exc, res = rjob.wait() if exc: - logging.error('call %s (%s) failed on peer with %s' % - (repr(rjob), meth, str(type(res).__name__))) + logging.error(lf('call failed', + call=repr(rjob), + method=meth, + error=str(type(res).__name__))) raise res logging.debug("call %s %s -> %s" % (repr(rjob), meth, repr(res))) return res diff --git a/geo-replication/syncdaemon/resource.py b/geo-replication/syncdaemon/resource.py index 1e11b980ad3..f12c7ceaa36 100644 --- a/geo-replication/syncdaemon/resource.py +++ b/geo-replication/syncdaemon/resource.py @@ -13,259 +13,43 @@ import os import sys import stat import time -import signal import fcntl -import errno import types import struct -import socket import logging import tempfile -import threading import subprocess -from errno import EEXIST, ENOENT, ENODATA, ENOTDIR, ELOOP -from errno import EISDIR, ENOTEMPTY, ESTALE, EINVAL -from select import error as SelectError -import shutil +from errno import (EEXIST, ENOENT, ENODATA, ENOTDIR, ELOOP, EACCES, + EISDIR, ENOTEMPTY, ESTALE, EINVAL, EBUSY, EPERM) +import errno + +from rconf import rconf +import gsyncdconfig as gconf +import libgfchangelog -from gconf import gconf import repce from repce import RepceServer, RepceClient from master import gmaster_builder import syncdutils -from syncdutils import GsyncdError, select, privileged, boolify, funcode -from syncdutils import umask, entry2pb, gauxpfx, errno_wrap, lstat -from syncdutils import NoPurgeTimeAvailable, PartialHistoryAvailable -from syncdutils import ChangelogException -from syncdutils import CHANGELOG_AGENT_CLIENT_VERSION +from syncdutils import (GsyncdError, select, privileged, funcode, + entry2pb, gauxpfx, errno_wrap, lstat, + NoStimeAvailable, PartialHistoryAvailable, + ChangelogException, ChangelogHistoryNotAvailable, + get_changelog_log_level, get_rsync_version, + GX_GFID_CANONICAL_LEN, + gf_mount_ready, lf, Popen, sup, + Xattr, matching_disk_gfid, get_gfid_from_mnt, + unshare_propagation_supported, get_slv_dir_path) +from gsyncdstatus import GeorepStatus +from py2py3 import (pipe, str_to_bytearray, entry_pack_reg, + entry_pack_reg_stat, entry_pack_mkdir, + entry_pack_symlink) -UrlRX = re.compile('\A(\w+)://([^ *?[]*)\Z') -HostRX = re.compile('[a-z\d](?:[a-z\d.-]*[a-z\d])?', re.I) -UserRX = re.compile("[\w!\#$%&'*+-\/=?^_`{|}~]+") +ENOTSUP = getattr(errno, 'ENOTSUP', 'EOPNOTSUPP') - -def sup(x, *a, **kw): - """a rubyesque "super" for python ;) - - invoke caller method in parent class with given args. - """ - return getattr(super(type(x), x), - sys._getframe(1).f_code.co_name)(*a, **kw) - - -def desugar(ustr): - """transform sugared url strings to standard <scheme>://<urlbody> form - - parsing logic enforces the constraint that sugared forms should contatin - a ':' or a '/', which ensures that sugared urls do not conflict with - gluster volume names. - """ - m = re.match('([^:]*):(.*)', ustr) - if m: - if not m.groups()[0]: - return "gluster://localhost" + ustr - elif '@' in m.groups()[0] or re.search('[:/]', m.groups()[1]): - return "ssh://" + ustr - else: - return "gluster://" + ustr - else: - if ustr[0] != '/': - raise GsyncdError("cannot resolve sugared url '%s'" % ustr) - ap = os.path.normpath(ustr) - if ap.startswith('//'): - ap = ap[1:] - return "file://" + ap - - -def gethostbyname(hnam): - """gethostbyname wrapper""" - try: - return socket.gethostbyname(hnam) - except socket.gaierror: - ex = sys.exc_info()[1] - raise GsyncdError("failed to resolve %s: %s" % - (hnam, ex.strerror)) - - -def parse_url(ustr): - """instantiate an url object by scheme-to-class dispatch - - The url classes taken into consideration are the ones in - this module whose names are full-caps. - """ - m = UrlRX.match(ustr) - if not m: - ustr = desugar(ustr) - m = UrlRX.match(ustr) - if not m: - raise GsyncdError("malformed url") - sch, path = m.groups() - this = sys.modules[__name__] - if not hasattr(this, sch.upper()): - raise GsyncdError("unknown url scheme " + sch) - return getattr(this, sch.upper())(path) - - -class _MetaXattr(object): - - """singleton class, a lazy wrapper around the - libcxattr module - - libcxattr (a heavy import due to ctypes) is - loaded only when when the single - instance is tried to be used. - - This reduces runtime for those invocations - which do not need filesystem manipulation - (eg. for config, url parsing) - """ - - def __getattr__(self, meth): - from libcxattr import Xattr as LXattr - xmeth = [m for m in dir(LXattr) if m[0] != '_'] - if not meth in xmeth: - return - for m in xmeth: - setattr(self, m, getattr(LXattr, m)) - return getattr(self, meth) - - -Xattr = _MetaXattr() - - -class Popen(subprocess.Popen): - - """customized subclass of subprocess.Popen with a ring - buffer for children error output""" - - @classmethod - def init_errhandler(cls): - """start the thread which handles children's error output""" - cls.errstore = {} - - def tailer(): - while True: - errstore = cls.errstore.copy() - try: - poe, _, _ = select( - [po.stderr for po in errstore], [], [], 1) - except (ValueError, SelectError): - continue - for po in errstore: - if po.stderr not in poe: - continue - po.lock.acquire() - try: - if po.on_death_row: - continue - la = errstore[po] - try: - fd = po.stderr.fileno() - except ValueError: # file is already closed - continue - l = os.read(fd, 1024) - if not l: - continue - tots = len(l) - for lx in la: - tots += len(lx) - while tots > 1 << 20 and la: - tots -= len(la.pop(0)) - la.append(l) - finally: - po.lock.release() - t = syncdutils.Thread(target=tailer) - t.start() - cls.errhandler = t - - @classmethod - def fork(cls): - """fork wrapper that restarts errhandler thread in child""" - pid = os.fork() - if not pid: - cls.init_errhandler() - return pid - - def __init__(self, args, *a, **kw): - """customizations for subprocess.Popen instantiation - - - 'close_fds' is taken to be the default - - if child's stderr is chosen to be managed, - register it with the error handler thread - """ - self.args = args - if 'close_fds' not in kw: - kw['close_fds'] = True - self.lock = threading.Lock() - self.on_death_row = False - try: - sup(self, args, *a, **kw) - except: - ex = sys.exc_info()[1] - if not isinstance(ex, OSError): - raise - raise GsyncdError("""execution of "%s" failed with %s (%s)""" % - (args[0], errno.errorcode[ex.errno], - os.strerror(ex.errno))) - if kw.get('stderr') == subprocess.PIPE: - assert(getattr(self, 'errhandler', None)) - self.errstore[self] = [] - - def errlog(self): - """make a log about child's failure event""" - filling = "" - if self.elines: - filling = ", saying:" - logging.error("""command "%s" returned with %s%s""" % - (" ".join(self.args), repr(self.returncode), filling)) - lp = '' - - def logerr(l): - logging.error(self.args[0] + "> " + l) - for l in self.elines: - ls = l.split('\n') - ls[0] = lp + ls[0] - lp = ls.pop() - for ll in ls: - logerr(ll) - if lp: - logerr(lp) - - def errfail(self): - """fail nicely if child did not terminate with success""" - self.errlog() - syncdutils.finalize(exval=1) - - def terminate_geterr(self, fail_on_err=True): - """kill child, finalize stderr harvesting (unregister - from errhandler, set up .elines), fail on error if - asked for - """ - self.lock.acquire() - try: - self.on_death_row = True - finally: - self.lock.release() - elines = self.errstore.pop(self) - if self.poll() is None: - self.terminate() - if self.poll() is None: - time.sleep(0.1) - self.kill() - self.wait() - while True: - if not select([self.stderr], [], [], 0.1)[0]: - break - b = os.read(self.stderr.fileno(), 1024) - if b: - elines.append(b) - else: - break - self.stderr.close() - self.elines = elines - if fail_on_err and self.returncode != 0: - self.errfail() +slv_volume = None +slv_host = None class Server(object): @@ -282,7 +66,6 @@ class Server(object): NTV_FMTSTR = "!" + "B" * 19 + "II" FRGN_XTRA_FMT = "I" FRGN_FMTSTR = NTV_FMTSTR + FRGN_XTRA_FMT - GX_GFID_CANONICAL_LEN = 37 # canonical gfid len + '\0' # for backend gfid fetch, do not use GX_NSPACE_PFX GFID_XATTR = 'trusted.gfid' @@ -292,15 +75,15 @@ class Server(object): @classmethod def _fmt_mknod(cls, l): - return "!II%dsI%dsIII" % (cls.GX_GFID_CANONICAL_LEN, l + 1) + return "!II%dsI%dsIII" % (GX_GFID_CANONICAL_LEN, l + 1) @classmethod def _fmt_mkdir(cls, l): - return "!II%dsI%dsII" % (cls.GX_GFID_CANONICAL_LEN, l + 1) + return "!II%dsI%dsII" % (GX_GFID_CANONICAL_LEN, l + 1) @classmethod def _fmt_symlink(cls, l1, l2): - return "!II%dsI%ds%ds" % (cls.GX_GFID_CANONICAL_LEN, l1 + 1, l2 + 1) + return "!II%dsI%ds%ds" % (GX_GFID_CANONICAL_LEN, l1 + 1, l2 + 1) def _pathguard(f): """decorator method that checks @@ -312,14 +95,14 @@ class Server(object): fc = funcode(f) pi = list(fc.co_varnames).index('path') - def ff(*a): - path = a[pi] + def ff(*args): + path = args[pi] ps = path.split('/') if path[0] == '/' or '..' in ps: raise ValueError('unsafe path') - a = list(a) - a[pi] = os.path.join(a[0].local_path, path) - return f(*a) + args = list(args) + args[pi] = os.path.join(args[0].local_path, path) + return f(*args) return ff @classmethod @@ -360,23 +143,15 @@ class Server(object): @classmethod @_pathguard def gfid(cls, path): - try: - buf = Xattr.lgetxattr(path, cls.GFID_XATTR, 16) + buf = errno_wrap(Xattr.lgetxattr, [path, cls.GFID_XATTR, 16], + [ENOENT], [ESTALE, ENODATA]) + if buf == ENOENT: + return buf + else: + buf = str_to_bytearray(buf) m = re.match('(.{8})(.{4})(.{4})(.{4})(.{12})', "".join( ['%02x' % x for x in struct.unpack(cls.GFID_FMTSTR, buf)])) return '-'.join(m.groups()) - except (IOError, OSError): - ex = sys.exc_info()[1] - if ex.errno == ENOENT: - return ex.errno - else: - raise - - @classmethod - def gfid_mnt(cls, gfidpath): - return errno_wrap(Xattr.lgetxattr, - [gfidpath, 'glusterfs.gfid.string', - cls.GX_GFID_CANONICAL_LEN], [ENOENT]) @classmethod @_pathguard @@ -464,6 +239,7 @@ class Server(object): val = Xattr.lgetxattr(path, '.'.join([cls.GX_NSPACE, uuid, 'xtime']), 8) + val = str_to_bytearray(val) return struct.unpack('!II', val) except OSError: ex = sys.exc_info()[1] @@ -486,6 +262,7 @@ class Server(object): val = Xattr.lgetxattr(path, '.'.join([cls.GX_NSPACE, uuid, 'stime']), 8) + val = str_to_bytearray(val) return struct.unpack('!II', val) except OSError: ex = sys.exc_info()[1] @@ -508,6 +285,31 @@ class Server(object): val = Xattr.lgetxattr(path, '.'.join([cls.GX_NSPACE, uuid, 'stime']), 8) + val = str_to_bytearray(val) + return struct.unpack('!II', val) + except OSError: + ex = sys.exc_info()[1] + if ex.errno in (ENOENT, ENODATA, ENOTDIR): + return ex.errno + else: + raise + + @classmethod + @_pathguard + def entry_stime(cls, path, uuid): + """ + entry_stime xattr to reduce the number of retry of Entry changes when + Geo-rep worker crashes and restarts. entry_stime is updated after + processing every changelog file. On failure and restart, worker only + have to reprocess the last changelog for Entry ops. + Xattr Key: <PFX>.<MASTERVOL_UUID>.<SLAVEVOL_UUID>.entry_stime + """ + try: + val = Xattr.lgetxattr(path, + '.'.join([cls.GX_NSPACE, uuid, + 'entry_stime']), + 8) + val = str_to_bytearray(val) return struct.unpack('!II', val) except OSError: ex = sys.exc_info()[1] @@ -529,17 +331,34 @@ class Server(object): @_pathguard def set_stime(cls, path, uuid, mark): """set @mark as stime for @uuid on @path""" - Xattr.lsetxattr( - path, '.'.join([cls.GX_NSPACE, uuid, 'stime']), - struct.pack('!II', *mark)) + errno_wrap(Xattr.lsetxattr, + [path, + '.'.join([cls.GX_NSPACE, uuid, 'stime']), + struct.pack('!II', *mark)], + [ENOENT], + [ESTALE, EINVAL]) + + @classmethod + @_pathguard + def set_entry_stime(cls, path, uuid, mark): + """set @mark as stime for @uuid on @path""" + errno_wrap(Xattr.lsetxattr, + [path, + '.'.join([cls.GX_NSPACE, uuid, 'entry_stime']), + struct.pack('!II', *mark)], + [ENOENT], + [ESTALE, EINVAL]) @classmethod @_pathguard def set_xtime(cls, path, uuid, mark): """set @mark as xtime for @uuid on @path""" - Xattr.lsetxattr( - path, '.'.join([cls.GX_NSPACE, uuid, 'xtime']), - struct.pack('!II', *mark)) + errno_wrap(Xattr.lsetxattr, + [path, + '.'.join([cls.GX_NSPACE, uuid, 'xtime']), + struct.pack('!II', *mark)], + [ENOENT], + [ESTALE, EINVAL]) @classmethod @_pathguard @@ -559,119 +378,390 @@ class Server(object): def entry_ops(cls, entries): pfx = gauxpfx() logging.debug('entries: %s' % repr(entries)) - # regular file - - def entry_pack_reg(gf, bn, mo, uid, gid): - blen = len(bn) - return struct.pack(cls._fmt_mknod(blen), - uid, gid, gf, mo, bn, - stat.S_IMODE(mo), 0, umask()) - - def entry_pack_reg_stat(gf, bn, st): - blen = len(bn) - mo = st['mode'] - return struct.pack(cls._fmt_mknod(blen), - st['uid'], st['gid'], - gf, mo, bn, - stat.S_IMODE(mo), 0, umask()) - # mkdir - - def entry_pack_mkdir(gf, bn, mo, uid, gid): - blen = len(bn) - return struct.pack(cls._fmt_mkdir(blen), - uid, gid, gf, mo, bn, - stat.S_IMODE(mo), umask()) - # symlink - - def entry_pack_symlink(gf, bn, lnk, st): - blen = len(bn) - llen = len(lnk) - return struct.pack(cls._fmt_symlink(blen, llen), - st['uid'], st['gid'], - gf, st['mode'], bn, lnk) - - def entry_purge(entry, gfid): + dist_count = rconf.args.master_dist_count + + def entry_purge(op, entry, gfid, e, uid, gid): # This is an extremely racy code and needs to be fixed ASAP. # The GFID check here is to be sure that the pargfid/bname # to be purged is the GFID gotten from the changelog. # (a stat(changelog_gfid) would also be valid here) # The race here is between the GFID check and the purge. - disk_gfid = cls.gfid_mnt(entry) - if isinstance(disk_gfid, int): + + # If the entry or the gfid of the file to be deleted is not present + # on slave, we can ignore the unlink/rmdir + if isinstance(lstat(entry), int) or \ + isinstance(lstat(os.path.join(pfx, gfid)), int): return - if not gfid == disk_gfid: + + if not matching_disk_gfid(gfid, entry): + collect_failure(e, EEXIST, uid, gid) return - er = errno_wrap(os.unlink, [entry], [ENOENT, EISDIR]) - if isinstance(er, int): + + if op == 'UNLINK': + er = errno_wrap(os.unlink, [entry], [ENOENT, ESTALE], [EBUSY]) + # EISDIR is safe error, ignore. This can only happen when + # unlink is sent from master while fixing gfid conflicts. + if er != EISDIR: + return er + + elif op == 'RMDIR': + er = errno_wrap(os.rmdir, [entry], [ENOENT, ESTALE, + ENOTEMPTY], [EBUSY]) + if er == ENOTEMPTY: + return er + + def collect_failure(e, cmd_ret, uid, gid, dst=False): + slv_entry_info = {} + slv_entry_info['gfid_mismatch'] = False + slv_entry_info['name_mismatch'] = False + slv_entry_info['dst'] = dst + slv_entry_info['slave_isdir'] = False + slv_entry_info['slave_name'] = None + slv_entry_info['slave_gfid'] = None + # We do this for failing fops on Slave + # Master should be logging this + if cmd_ret is None: + return False + + if e.get("stat", {}): + # Copy actual UID/GID value back to entry stat + e['stat']['uid'] = uid + e['stat']['gid'] = gid + + if cmd_ret in [EEXIST, ESTALE]: + if dst: + en = e['entry1'] + else: + en = e['entry'] + disk_gfid = get_gfid_from_mnt(en) + if isinstance(disk_gfid, str) and \ + e['gfid'] != disk_gfid: + slv_entry_info['gfid_mismatch'] = True + st = lstat(en) + if not isinstance(st, int): + if st and stat.S_ISDIR(st.st_mode): + slv_entry_info['slave_isdir'] = True + dir_name = get_slv_dir_path(slv_host, slv_volume, + disk_gfid) + slv_entry_info['slave_name'] = dir_name + else: + slv_entry_info['slave_isdir'] = False + slv_entry_info['slave_gfid'] = disk_gfid + failures.append((e, cmd_ret, slv_entry_info)) + else: + return False + else: + failures.append((e, cmd_ret, slv_entry_info)) + + return True + + failures = [] + + def recursive_rmdir(gfid, entry, path): + """disk_gfid check added for original path for which + recursive_delete is called. This disk gfid check executed + before every Unlink/Rmdir. If disk gfid is not matching + with GFID from Changelog, that means other worker + deleted the directory. Even if the subdir/file present, + it belongs to different parent. Exit without performing + further deletes. + """ + if not matching_disk_gfid(gfid, entry): + return + + names = [] + names = errno_wrap(os.listdir, [path], [ENOENT], [ESTALE, ENOTSUP]) + if isinstance(names, int): + return + + for name in names: + fullname = os.path.join(path, name) + if not matching_disk_gfid(gfid, entry): + return + er = errno_wrap(os.remove, [fullname], [ENOENT, ESTALE, + EISDIR], [EBUSY]) + if er == EISDIR: - er = errno_wrap(os.rmdir, [entry], [ENOENT, ENOTEMPTY]) - if er == ENOTEMPTY: - return er + recursive_rmdir(gfid, entry, fullname) + + if not matching_disk_gfid(gfid, entry): + return + + errno_wrap(os.rmdir, [path], [ENOENT, ESTALE], [EBUSY]) + + def rename_with_disk_gfid_confirmation(gfid, entry, en, uid, gid): + if not matching_disk_gfid(gfid, entry): + logging.error(lf("RENAME ignored: source entry does not match " + "with on-disk gfid", + source=entry, + gfid=gfid, + disk_gfid=get_gfid_from_mnt(entry), + target=en)) + collect_failure(e, EEXIST, uid, gid) + return + + cmd_ret = errno_wrap(os.rename, + [entry, en], + [ENOENT, EEXIST], [ESTALE, EBUSY]) + collect_failure(e, cmd_ret, uid, gid) + for e in entries: blob = None op = e['op'] gfid = e['gfid'] entry = e['entry'] + uid = 0 + gid = 0 + + # Skip entry processing if it's marked true during gfid + # conflict resolution + if e['skip_entry']: + continue + + if e.get("stat", {}): + # Copy UID/GID value and then reset to zero. Copied UID/GID + # will be used to run chown once entry is created. + uid = e['stat']['uid'] + gid = e['stat']['gid'] + e['stat']['uid'] = 0 + e['stat']['gid'] = 0 + (pg, bname) = entry2pb(entry) if op in ['RMDIR', 'UNLINK']: - while True: - er = entry_purge(entry, gfid) - if isinstance(er, int): - if er == ENOTEMPTY and op == 'RMDIR': - er1 = errno_wrap(shutil.rmtree, - [os.path.join(pg, bname)], - [ENOENT]) - if not isinstance(er1, int): - logging.info("Removed %s/%s recursively" % - (pg, bname)) - break - - logging.warn("Failed to remove %s => %s/%s. %s" % - (gfid, pg, bname, os.strerror(er))) - time.sleep(1) + # Try once, if rmdir failed with ENOTEMPTY + # then delete recursively. + er = entry_purge(op, entry, gfid, e, uid, gid) + if isinstance(er, int): + if er == ENOTEMPTY and op == 'RMDIR': + # Retry if ENOTEMPTY, ESTALE + er1 = errno_wrap(recursive_rmdir, + [gfid, entry, + os.path.join(pg, bname)], + [], [ENOTEMPTY, ESTALE, ENODATA]) + if not isinstance(er1, int): + logging.debug("Removed %s => %s/%s recursively" % + (gfid, pg, bname)) + else: + logging.warn(lf("Recursive remove failed", + gfid=gfid, + pgfid=pg, + bname=bname, + error=os.strerror(er1))) else: - break + logging.warn(lf("Failed to remove", + gfid=gfid, + pgfid=pg, + bname=bname, + error=os.strerror(er))) elif op in ['CREATE', 'MKNOD']: - blob = entry_pack_reg( - gfid, bname, e['mode'], e['uid'], e['gid']) + slink = os.path.join(pfx, gfid) + st = lstat(slink) + # don't create multiple entries with same gfid + if isinstance(st, int): + blob = entry_pack_reg(cls, gfid, bname, + e['mode'], e['uid'], e['gid']) + # Self healed hardlinks are recorded as MKNOD. + # So if the gfid already exists, it should be + # processed as hard link not mknod. + elif op in ['MKNOD']: + cmd_ret = errno_wrap(os.link, + [slink, entry], + [ENOENT, EEXIST], [ESTALE]) + collect_failure(e, cmd_ret, uid, gid) elif op == 'MKDIR': - blob = entry_pack_mkdir( - gfid, bname, e['mode'], e['uid'], e['gid']) + en = e['entry'] + slink = os.path.join(pfx, gfid) + st = lstat(slink) + # don't create multiple entries with same gfid + if isinstance(st, int): + blob = entry_pack_mkdir(cls, gfid, bname, + e['mode'], e['uid'], e['gid']) + elif (isinstance(lstat(en), int) or + not matching_disk_gfid(gfid, en)): + # If gfid of a directory exists on slave but path based + # create is getting EEXIST. This means the directory is + # renamed in master but recorded as MKDIR during hybrid + # crawl. Get the directory path by reading the backend + # symlink and trying to rename to new name as said by + # master. + logging.info(lf("Special case: rename on mkdir", + gfid=gfid, entry=repr(entry))) + src_entry = get_slv_dir_path(slv_host, slv_volume, gfid) + if src_entry is None: + collect_failure(e, ENOENT, uid, gid) + if src_entry is not None and src_entry != entry: + slv_entry_info = {} + slv_entry_info['gfid_mismatch'] = False + slv_entry_info['name_mismatch'] = True + slv_entry_info['dst'] = False + slv_entry_info['slave_isdir'] = True + slv_entry_info['slave_gfid'] = gfid + slv_entry_info['slave_entry'] = src_entry + + failures.append((e, EEXIST, slv_entry_info)) elif op == 'LINK': slink = os.path.join(pfx, gfid) st = lstat(slink) if isinstance(st, int): (pg, bname) = entry2pb(entry) - blob = entry_pack_reg_stat(gfid, bname, e['stat']) + if stat.S_ISREG(e['stat']['mode']): + blob = entry_pack_reg_stat(cls, gfid, bname, e['stat']) + elif stat.S_ISLNK(e['stat']['mode']): + blob = entry_pack_symlink(cls, gfid, bname, e['link'], + e['stat']) else: - errno_wrap(os.link, [slink, entry], [ENOENT, EEXIST]) + cmd_ret = errno_wrap(os.link, + [slink, entry], + [ENOENT, EEXIST], [ESTALE]) + collect_failure(e, cmd_ret, uid, gid) elif op == 'SYMLINK': - blob = entry_pack_symlink(gfid, bname, e['link'], e['stat']) - elif op == 'RENAME': - en = e['entry1'] + en = e['entry'] st = lstat(entry) if isinstance(st, int): + blob = entry_pack_symlink(cls, gfid, bname, e['link'], + e['stat']) + elif not matching_disk_gfid(gfid, en): + collect_failure(e, EEXIST, uid, gid) + elif op == 'RENAME': + en = e['entry1'] + # The matching disk gfid check validates two things + # 1. Validates name is present, return false otherwise + # 2. Validates gfid is same, returns false otherwise + # So both validations are necessary to decide src doesn't + # exist. We can't rely on only gfid stat as hardlink could + # be present and we can't rely only on name as name could + # exist with different gfid. + if not matching_disk_gfid(gfid, entry): if e['stat'] and not stat.S_ISDIR(e['stat']['mode']): - (pg, bname) = entry2pb(en) - blob = entry_pack_reg_stat(gfid, bname, e['stat']) + if stat.S_ISLNK(e['stat']['mode']): + # src is not present, so don't sync symlink as + # we don't know target. It's ok to ignore. If + # it's unliked, it's fine. If it's renamed to + # something else, it will be synced then. + if e['link'] is not None: + st1 = lstat(en) + if isinstance(st1, int): + (pg, bname) = entry2pb(en) + blob = entry_pack_symlink(cls, gfid, bname, + e['link'], + e['stat']) + elif not matching_disk_gfid(gfid, en): + collect_failure(e, EEXIST, uid, gid, True) + else: + slink = os.path.join(pfx, gfid) + st = lstat(slink) + # don't create multiple entries with same gfid + if isinstance(st, int): + (pg, bname) = entry2pb(en) + blob = entry_pack_reg_stat(cls, gfid, bname, + e['stat']) + else: + cmd_ret = errno_wrap(os.link, [slink, en], + [ENOENT, EEXIST], [ESTALE]) + collect_failure(e, cmd_ret, uid, gid) else: - errno_wrap(os.rename, [entry, en], [ENOENT, EEXIST]) + st = lstat(entry) + st1 = lstat(en) + if isinstance(st1, int): + rename_with_disk_gfid_confirmation(gfid, entry, en, + uid, gid) + else: + if st.st_ino == st1.st_ino: + # we have a hard link, we can now unlink source + try: + errno_wrap(os.unlink, [entry], + [ENOENT, ESTALE], [EBUSY]) + except OSError as e: + if e.errno == EISDIR: + try: + errno_wrap(os.rmdir, [entry], + [ENOENT, ESTALE], [EBUSY]) + except OSError as e: + if e.errno == ENOTEMPTY: + logging.error( + lf("Directory Rename failed. " + "Both Old and New" + " directories exists", + old=entry, + new=en)) + else: + raise + else: + raise + elif not matching_disk_gfid(gfid, en) and dist_count > 1: + collect_failure(e, EEXIST, uid, gid, True) + else: + # We are here which means matching_disk_gfid for + # both source and destination has returned false + # and distribution count for master vol is greater + # then one. Which basically says both the source and + # destination exist and not hardlinks. + # So we are safe to go ahead with rename here. + rename_with_disk_gfid_confirmation(gfid, entry, en, + uid, gid) if blob: - errno_wrap(Xattr.lsetxattr, - [pg, 'glusterfs.gfid.newfile', blob], - [EEXIST], - [ENOENT, ESTALE, EINVAL]) + cmd_ret = errno_wrap(Xattr.lsetxattr, + [pg, 'glusterfs.gfid.newfile', blob], + [EEXIST, ENOENT, ESTALE], + [ESTALE, EINVAL, EBUSY]) + collect_failure(e, cmd_ret, uid, gid) + + # If UID/GID is different than zero that means we are trying + # create Entry with different UID/GID. Create Entry with + # UID:0 and GID:0, and then call chown to set UID/GID + if uid != 0 or gid != 0: + path = os.path.join(pfx, gfid) + cmd_ret = errno_wrap(os.lchown, [path, uid, gid], [ENOENT], + [ESTALE, EINVAL]) + collect_failure(e, cmd_ret, uid, gid) + + return failures @classmethod def meta_ops(cls, meta_entries): logging.debug('Meta-entries: %s' % repr(meta_entries)) + failures = [] for e in meta_entries: mode = e['stat']['mode'] uid = e['stat']['uid'] gid = e['stat']['gid'] + atime = e['stat']['atime'] + mtime = e['stat']['mtime'] go = e['go'] - errno_wrap(os.chmod, [go, mode], [ENOENT], [ESTALE, EINVAL]) - errno_wrap(os.chown, [go, uid, gid], [ENOENT], [ESTALE, EINVAL]) + # Linux doesn't support chmod on symlink itself. + # It is always applied to the target file. So + # changelog would record target file's gfid + # and we are good. But 'chown' is supported on + # symlink file. So changelog would record symlink + # gfid in such cases. Since we do 'chown' 'chmod' + # 'utime' for each gfid recorded for metadata, and + # we know from changelog the metadata is on symlink's + # gfid or target file's gfid, we should be doing + # 'lchown' 'lchmod' 'utime with no-deference' blindly. + # But since 'lchmod' and 'utime with no de-reference' is + # not supported in python3, we have to rely on 'chmod' + # and 'utime with de-reference'. Hence avoiding 'chmod' + # and 'utime' if it's symlink file. + + is_symlink = False + cmd_ret = errno_wrap(os.lchown, [go, uid, gid], [ENOENT], + [ESTALE, EINVAL]) + if isinstance(cmd_ret, int): + continue + + is_symlink = os.path.islink(go) + + if not is_symlink: + cmd_ret = errno_wrap(os.chmod, [go, mode], + [ENOENT, EACCES, EPERM], [ESTALE, EINVAL]) + if isinstance(cmd_ret, int): + failures.append((e, cmd_ret, "chmod")) + + cmd_ret = errno_wrap(os.utime, [go, (atime, mtime)], + [ENOENT, EACCES, EPERM], [ESTALE, EINVAL]) + if isinstance(cmd_ret, int): + failures.append((e, cmd_ret, "utime")) + return failures @classmethod @_pathguard @@ -727,211 +817,273 @@ class Server(object): return 1.0 -class SlaveLocal(object): - - """mix-in class to implement some factes of a slave server - - ("mix-in" is sort of like "abstract class", ie. it's not - instantiated just included in the ancesty DAG. I use "mix-in" - to indicate that it's not used as an abstract base class, - rather just taken in to implement additional functionality - on the basis of the assumed availability of certain interfaces.) - """ - - def can_connect_to(self, remote): - """determine our position in the connectibility matrix""" - return not remote - - def service_loop(self): - """start a RePCe server serving self's server +class Mounter(object): - stop servicing if a timeout is configured and got no - keep-alime in that inteval - """ - - if boolify(gconf.use_rsync_xattrs) and not privileged(): - raise GsyncdError( - "using rsync for extended attributes is not supported") + """Abstract base class for mounter backends""" - repce = RepceServer( - self.server, sys.stdin, sys.stdout, int(gconf.sync_jobs)) - t = syncdutils.Thread(target=lambda: (repce.service_loop(), - syncdutils.finalize())) - t.start() - logging.info("slave listening") - if gconf.timeout and int(gconf.timeout) > 0: - while True: - lp = self.server.last_keep_alive - time.sleep(int(gconf.timeout)) - if lp == self.server.last_keep_alive: - logging.info( - "connection inactive for %d seconds, stopping" % - int(gconf.timeout)) - break - else: - select((), (), ()) + def __init__(self, params): + self.params = params + self.mntpt = None + self.umount_cmd = [] + @classmethod + def get_glusterprog(cls): + gluster_cmd_dir = gconf.get("gluster-command-dir") + if rconf.args.subcmd == "slave": + gluster_cmd_dir = gconf.get("slave-gluster-command-dir") + return os.path.join(gluster_cmd_dir, cls.glusterprog) + + def umount_l(self, d): + """perform lazy umount""" + po = Popen(self.make_umount_argv(d), stderr=subprocess.PIPE, + universal_newlines=True) + po.wait() + return po -class SlaveRemote(object): + @classmethod + def make_umount_argv(cls, d): + raise NotImplementedError - """mix-in class to implement an interface to a remote slave""" + def make_mount_argv(self, label=None): + raise NotImplementedError - def connect_remote(self, rargs=[], **opts): - """connects to a remote slave + def cleanup_mntpt(self, *a): + pass - Invoke an auxiliary utility (slave gsyncd, possibly wrapped) - which sets up the connection and set up a RePCe client to - communicate throuh its stdio. - """ - slave = opts.get('slave', self.url) - extra_opts = [] - so = getattr(gconf, 'session_owner', None) - if so: - extra_opts += ['--session-owner', so] - if boolify(gconf.use_rsync_xattrs): - extra_opts.append('--use-rsync-xattrs') - po = Popen(rargs + gconf.remote_gsyncd.split() + extra_opts + - ['-N', '--listen', '--timeout', str(gconf.timeout), - slave], - stdin=subprocess.PIPE, stdout=subprocess.PIPE, - stderr=subprocess.PIPE) - gconf.transport = po - return self.start_fd_client(po.stdout, po.stdin, **opts) + def handle_mounter(self, po): + po.wait() - def start_fd_client(self, i, o, **opts): - """set up RePCe client, handshake with server + def inhibit(self, label): + """inhibit a gluster filesystem - It's cut out as a separate method to let - subclasses hook into client startup + Mount glusterfs over a temporary mountpoint, + change into the mount, and lazy unmount the + filesystem. """ - self.server = RepceClient(i, o) - rv = self.server.__version__() - exrv = {'proto': repce.repce_version, 'object': Server.version()} - da0 = (rv, exrv) - da1 = ({}, {}) - for i in range(2): - for k, v in da0[i].iteritems(): - da1[i][k] = int(v) - if da1[0] != da1[1]: - raise GsyncdError( - "RePCe major version mismatch: local %s, remote %s" % - (exrv, rv)) - - def rsync(self, files, *args): - """invoke rsync""" - if not files: - raise GsyncdError("no files to sync") - logging.debug("files: " + ", ".join(files)) - argv = gconf.rsync_command.split() + \ - ['-avR0', '--inplace', '--files-from=-', '--super', - '--stats', '--numeric-ids', '--no-implied-dirs'] + \ - gconf.rsync_options.split() + \ - (boolify(gconf.use_rsync_xattrs) and ['--xattrs'] or []) + \ - ['.'] + list(args) - po = Popen(argv, stdin=subprocess.PIPE, stderr=subprocess.PIPE) - for f in files: - po.stdin.write(f) - po.stdin.write('\0') - - po.stdin.close() - po.wait() - po.terminate_geterr(fail_on_err=False) + mpi, mpo = pipe() + mh = Popen.fork() + if mh: + # Parent + os.close(mpi) + fcntl.fcntl(mpo, fcntl.F_SETFD, fcntl.FD_CLOEXEC) + d = None + margv = self.make_mount_argv(label) + if self.mntpt: + # mntpt is determined pre-mount + d = self.mntpt + mnt_msg = d + '\0' + encoded_msg = mnt_msg.encode() + os.write(mpo, encoded_msg) + po = Popen(margv, **self.mountkw) + self.handle_mounter(po) + po.terminate_geterr() + logging.debug('auxiliary glusterfs mount in place') + if not d: + # mntpt is determined during mount + d = self.mntpt + mnt_msg = d + '\0' + encoded_msg = mnt_msg.encode() + os.write(mpo, encoded_msg) + encoded_msg = 'M'.encode() + os.write(mpo, encoded_msg) + t = syncdutils.Thread(target=lambda: os.chdir(d)) + t.start() + tlim = rconf.starttime + gconf.get("connection-timeout") + while True: + if not t.isAlive(): + break - return po + if time.time() >= tlim: + syncdutils.finalize(exval=1) + time.sleep(1) + os.close(mpo) + _, rv = syncdutils.waitpid(mh, 0) + if rv: + rv = (os.WIFEXITED(rv) and os.WEXITSTATUS(rv) or 0) - \ + (os.WIFSIGNALED(rv) and os.WTERMSIG(rv) or 0) + logging.warn(lf('stale mount possibly left behind', + path=d)) + raise GsyncdError("cleaning up temp mountpoint %s " + "failed with status %d" % + (d, rv)) + else: + rv = 0 + try: + os.setsid() + os.close(mpo) + mntdata = '' + while True: + c = os.read(mpi, 1) + c = c.decode() + if not c: + break + mntdata += c + if mntdata: + mounted = False + if mntdata[-1] == 'M': + mntdata = mntdata[:-1] + assert(mntdata) + mounted = True + assert(mntdata[-1] == '\0') + mntpt = mntdata[:-1] + assert(mntpt) + + umount_master = False + umount_slave = False + if rconf.args.subcmd == "worker" \ + and not unshare_propagation_supported() \ + and not gconf.get("access-mount"): + umount_master = True + if rconf.args.subcmd == "slave" \ + and not gconf.get("slave-access-mount"): + umount_slave = True + + if mounted and (umount_master or umount_slave): + po = self.umount_l(mntpt) + po.terminate_geterr(fail_on_err=False) + if po.returncode != 0: + po.errlog() + rv = po.returncode + logging.debug("Lazy umount done: %s" % mntpt) + if umount_master or umount_slave: + self.cleanup_mntpt(mntpt) + except: + logging.exception('mount cleanup failure:') + rv = 200 + os._exit(rv) + + #Polling the dht.subvol.status value. + RETRIES = 10 + while not gf_mount_ready(): + if RETRIES < 0: + logging.error('Subvols are not up') + break + RETRIES -= 1 + time.sleep(0.2) - def tarssh(self, files, slaveurl): - """invoke tar+ssh - -z (compress) can be use if needed, but ommitting it now - as it results in wierd error (tar+ssh errors out (errcode: 2) - """ - if not files: - raise GsyncdError("no files to sync") - logging.debug("files: " + ", ".join(files)) - (host, rdir) = slaveurl.split(':') - tar_cmd = ["tar", "-cf", "-", "--files-from", "-"] - ssh_cmd = gconf.ssh_command_tar.split() + \ - [host, "tar", "--overwrite", "-xf", "-", "-C", rdir] - p0 = Popen(tar_cmd, stdout=subprocess.PIPE, - stdin=subprocess.PIPE, stderr=subprocess.PIPE) - p1 = Popen(ssh_cmd, stdin=p0.stdout, stderr=subprocess.PIPE) - for f in files: - p0.stdin.write(f) - p0.stdin.write('\n') - p0.stdin.close() + logging.debug('auxiliary glusterfs mount prepared') - # wait() for tar to terminate, collecting any errors, further - # waiting for transfer to complete - p0.wait() - p0.terminate_geterr(fail_on_err=False) - p1.wait() - p1.terminate_geterr(fail_on_err=False) +class DirectMounter(Mounter): - return p1 + """mounter backend which calls mount(8), umount(8) directly""" + mountkw = {'stderr': subprocess.PIPE, 'universal_newlines': True} + glusterprog = 'glusterfs' -class AbstractUrl(object): + @staticmethod + def make_umount_argv(d): + return ['umount', '-l', d] - """abstract base class for url scheme classes""" + def make_mount_argv(self, label=None): + self.mntpt = tempfile.mkdtemp(prefix='gsyncd-aux-mount-') + rconf.mount_point = self.mntpt + return [self.get_glusterprog()] + \ + ['--' + p for p in self.params] + [self.mntpt] - def __init__(self, path, pattern): - m = re.search(pattern, path) - if not m: - raise GsyncdError("malformed path") - self.path = path - return m.groups() + def cleanup_mntpt(self, mntpt=None): + if not mntpt: + mntpt = self.mntpt + errno_wrap(os.rmdir, [mntpt], [ENOENT, EBUSY]) - @property - def scheme(self): - return type(self).__name__.lower() - def canonical_path(self): - return self.path +class MountbrokerMounter(Mounter): - def get_url(self, canonical=False, escaped=False): - """format self's url in various styles""" - if canonical: - pa = self.canonical_path() - else: - pa = self.path - u = "://".join((self.scheme, pa)) - if escaped: - u = syncdutils.escape(u) - return u + """mounter backend using the mountbroker gluster service""" - @property - def url(self): - return self.get_url() + mountkw = {'stderr': subprocess.PIPE, 'stdout': subprocess.PIPE, + 'universal_newlines': True} + glusterprog = 'gluster' + @classmethod + def make_cli_argv(cls): + return [cls.get_glusterprog()] + ['--remote-host=localhost'] + \ + gconf.get("gluster-cli-options").split() + ['system::'] -class FILE(AbstractUrl, SlaveLocal, SlaveRemote): + @classmethod + def make_umount_argv(cls, d): + return cls.make_cli_argv() + ['umount', d, 'lazy'] - """scheme class for file:// urls + def make_mount_argv(self, label): + return self.make_cli_argv() + \ + ['mount', label, 'user-map-root=' + + syncdutils.getusername()] + self.params - can be used to represent a file slave server - on slave side, or interface to a remote file - file server on master side - """ + def handle_mounter(self, po): + self.mntpt = po.stdout.readline()[:-1] + rconf.mount_point = self.mntpt + rconf.mountbroker = True + self.umount_cmd = self.make_cli_argv() + ['umount'] + rconf.mbr_umount_cmd = self.umount_cmd + po.stdout.close() + sup(self, po) + if po.returncode != 0: + # if cli terminated with error due to being + # refused by glusterd, what it put + # out on stdout is a diagnostic message + logging.error(lf('glusterd answered', mnt=self.mntpt)) - class FILEServer(Server): - """included server flavor""" - pass +class GLUSTERServer(Server): - server = FILEServer + "server enhancements for a glusterfs backend""" - def __init__(self, path): - sup(self, path, '^/') + @classmethod + def _attr_unpack_dict(cls, xattr, extra_fields=''): + """generic volume mark fetching/parsing backed""" + fmt_string = cls.NTV_FMTSTR + extra_fields + buf = Xattr.lgetxattr('.', xattr, struct.calcsize(fmt_string)) + buf = str_to_bytearray(buf) + vm = struct.unpack(fmt_string, buf) + m = re.match( + '(.{8})(.{4})(.{4})(.{4})(.{12})', + "".join(['%02x' % x for x in vm[2:18]])) + uuid = '-'.join(m.groups()) + volinfo = {'version': vm[0:2], + 'uuid': uuid, + 'retval': vm[18], + 'volume_mark': vm[19:21], + } + if extra_fields: + return volinfo, vm[-len(extra_fields):] + else: + return volinfo - def connect(self): - """inhibit the resource beyond""" - os.chdir(self.path) + @classmethod + def foreign_volume_infos(cls): + """return list of valid (not expired) foreign volume marks""" + dict_list = [] + xattr_list = Xattr.llistxattr_buf('.') + for ele in xattr_list: + if ele.find('.'.join([cls.GX_NSPACE, 'volume-mark', ''])) == 0: + d, x = cls._attr_unpack_dict(ele, cls.FRGN_XTRA_FMT) + now = int(time.time()) + if x[0] > now: + logging.debug("volinfo[%s] expires: %d " + "(%d sec later)" % + (d['uuid'], x[0], x[0] - now)) + d['timeout'] = x[0] + dict_list.append(d) + else: + try: + Xattr.lremovexattr('.', ele) + except OSError: + pass + return dict_list - def rsync(self, files): - return sup(self, files, self.path) + @classmethod + def native_volume_info(cls): + """get the native volume mark of the underlying gluster volume""" + try: + return cls._attr_unpack_dict('.'.join([cls.GX_NSPACE, + 'volume-mark'])) + except OSError: + ex = sys.exc_info()[1] + if ex.errno != ENODATA: + raise -class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote): +class GLUSTER(object): """scheme class for gluster:// urls @@ -941,238 +1093,17 @@ class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote): (slave-ish features come from the mixins, master functionality is outsourced to GMaster from master) """ - - class GLUSTERServer(Server): - - "server enhancements for a glusterfs backend""" - - @classmethod - def _attr_unpack_dict(cls, xattr, extra_fields=''): - """generic volume mark fetching/parsing backed""" - fmt_string = cls.NTV_FMTSTR + extra_fields - buf = Xattr.lgetxattr('.', xattr, struct.calcsize(fmt_string)) - vm = struct.unpack(fmt_string, buf) - m = re.match( - '(.{8})(.{4})(.{4})(.{4})(.{12})', - "".join(['%02x' % x for x in vm[2:18]])) - uuid = '-'.join(m.groups()) - volinfo = {'version': vm[0:2], - 'uuid': uuid, - 'retval': vm[18], - 'volume_mark': vm[19:21], - } - if extra_fields: - return volinfo, vm[-len(extra_fields):] - else: - return volinfo - - @classmethod - def foreign_volume_infos(cls): - """return list of valid (not expired) foreign volume marks""" - dict_list = [] - xattr_list = Xattr.llistxattr_buf('.') - for ele in xattr_list: - if ele.find('.'.join([cls.GX_NSPACE, 'volume-mark', ''])) == 0: - d, x = cls._attr_unpack_dict(ele, cls.FRGN_XTRA_FMT) - now = int(time.time()) - if x[0] > now: - logging.debug("volinfo[%s] expires: %d " - "(%d sec later)" % - (d['uuid'], x[0], x[0] - now)) - d['timeout'] = x[0] - dict_list.append(d) - else: - try: - Xattr.lremovexattr('.', ele) - except OSError: - pass - return dict_list - - @classmethod - def native_volume_info(cls): - """get the native volume mark of the underlying gluster volume""" - try: - return cls._attr_unpack_dict('.'.join([cls.GX_NSPACE, - 'volume-mark'])) - except OSError: - ex = sys.exc_info()[1] - if ex.errno != ENODATA: - raise - server = GLUSTERServer - def __init__(self, path): - self.host, self.volume = sup(self, path, '^(%s):(.+)' % HostRX.pattern) - - def canonical_path(self): - return ':'.join([gethostbyname(self.host), self.volume]) - - def can_connect_to(self, remote): - """determine our position in the connectibility matrix""" - return not remote or \ - (isinstance(remote, SSH) and isinstance(remote.inner_rsc, GLUSTER)) - - class Mounter(object): - - """Abstract base class for mounter backends""" - - def __init__(self, params): - self.params = params - self.mntpt = None - - @classmethod - def get_glusterprog(cls): - return os.path.join(gconf.gluster_command_dir, cls.glusterprog) - - def umount_l(self, d): - """perform lazy umount""" - po = Popen(self.make_umount_argv(d), stderr=subprocess.PIPE) - po.wait() - return po - - @classmethod - def make_umount_argv(cls, d): - raise NotImplementedError - - def make_mount_argv(self, *a): - raise NotImplementedError - - def cleanup_mntpt(self, *a): - pass - - def handle_mounter(self, po): - po.wait() - - def inhibit(self, *a): - """inhibit a gluster filesystem - - Mount glusterfs over a temporary mountpoint, - change into the mount, and lazy unmount the - filesystem. - """ - - mpi, mpo = os.pipe() - mh = Popen.fork() - if mh: - os.close(mpi) - fcntl.fcntl(mpo, fcntl.F_SETFD, fcntl.FD_CLOEXEC) - d = None - margv = self.make_mount_argv(*a) - if self.mntpt: - # mntpt is determined pre-mount - d = self.mntpt - os.write(mpo, d + '\0') - po = Popen(margv, **self.mountkw) - self.handle_mounter(po) - po.terminate_geterr() - logging.debug('auxiliary glusterfs mount in place') - if not d: - # mntpt is determined during mount - d = self.mntpt - os.write(mpo, d + '\0') - os.write(mpo, 'M') - t = syncdutils.Thread(target=lambda: os.chdir(d)) - t.start() - tlim = gconf.starttime + int(gconf.connection_timeout) - while True: - if not t.isAlive(): - break - if time.time() >= tlim: - syncdutils.finalize(exval=1) - time.sleep(1) - os.close(mpo) - _, rv = syncdutils.waitpid(mh, 0) - if rv: - rv = (os.WIFEXITED(rv) and os.WEXITSTATUS(rv) or 0) - \ - (os.WIFSIGNALED(rv) and os.WTERMSIG(rv) or 0) - logging.warn('stale mount possibly left behind on ' + d) - raise GsyncdError("cleaning up temp mountpoint %s " - "failed with status %d" % - (d, rv)) - else: - rv = 0 - try: - os.setsid() - os.close(mpo) - mntdata = '' - while True: - c = os.read(mpi, 1) - if not c: - break - mntdata += c - if mntdata: - mounted = False - if mntdata[-1] == 'M': - mntdata = mntdata[:-1] - assert(mntdata) - mounted = True - assert(mntdata[-1] == '\0') - mntpt = mntdata[:-1] - assert(mntpt) - if mounted: - po = self.umount_l(mntpt) - po.terminate_geterr(fail_on_err=False) - if po.returncode != 0: - po.errlog() - rv = po.returncode - self.cleanup_mntpt(mntpt) - except: - logging.exception('mount cleanup failure:') - rv = 200 - os._exit(rv) - logging.debug('auxiliary glusterfs mount prepared') - - class DirectMounter(Mounter): - - """mounter backend which calls mount(8), umount(8) directly""" - - mountkw = {'stderr': subprocess.PIPE} - glusterprog = 'glusterfs' - - @staticmethod - def make_umount_argv(d): - return ['umount', '-l', d] - - def make_mount_argv(self): - self.mntpt = tempfile.mkdtemp(prefix='gsyncd-aux-mount-') - return [self.get_glusterprog()] + \ - ['--' + p for p in self.params] + [self.mntpt] - - def cleanup_mntpt(self, mntpt=None): - if not mntpt: - mntpt = self.mntpt - os.rmdir(mntpt) - - class MountbrokerMounter(Mounter): - - """mounter backend using the mountbroker gluster service""" - - mountkw = {'stderr': subprocess.PIPE, 'stdout': subprocess.PIPE} - glusterprog = 'gluster' - - @classmethod - def make_cli_argv(cls): - return [cls.get_glusterprog()] + ['--remote-host=localhost'] + \ - gconf.gluster_cli_options.split() + ['system::'] - - @classmethod - def make_umount_argv(cls, d): - return cls.make_cli_argv() + ['umount', d, 'lazy'] - - def make_mount_argv(self, label): - return self.make_cli_argv() + \ - ['mount', label, 'user-map-root=' + - syncdutils.getusername()] + self.params + def __init__(self, host, volume): + self.path = "%s:%s" % (host, volume) + self.host = host + self.volume = volume - def handle_mounter(self, po): - self.mntpt = po.stdout.readline()[:-1] - po.stdout.close() - sup(self, po) - if po.returncode != 0: - # if cli terminated with error due to being - # refused by glusterd, what it put - # out on stdout is a diagnostic message - logging.error('glusterd answered: %s' % self.mntpt) + global slv_volume + global slv_host + slv_volume = self.volume + slv_host = self.host def connect(self): """inhibit the resource beyond @@ -1182,20 +1113,30 @@ class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote): with given backend """ - label = getattr(gconf, 'mountbroker', None) + logging.info("Mounting gluster volume locally...") + t0 = time.time() + label = gconf.get('mountbroker', None) if not label and not privileged(): label = syncdutils.getusername() - mounter = label and self.MountbrokerMounter or self.DirectMounter - params = gconf.gluster_params.split() + \ - (gconf.gluster_log_level and ['log-level=' + - gconf.gluster_log_level] or []) + \ - ['log-file=' + gconf.gluster_log_file, 'volfile-server=' + - self.host, 'volfile-id=' + self.volume, 'client-pid=-1'] - mounter(params).inhibit(*[l for l in [label] if l]) - - def connect_remote(self, *a, **kw): - sup(self, *a, **kw) - self.slavedir = "/proc/%d/cwd" % self.server.pid() + mounter = label and MountbrokerMounter or DirectMounter + + log_file = gconf.get("gluster-log-file") + if rconf.args.subcmd == "slave": + log_file = gconf.get("slave-gluster-log-file") + + log_level = gconf.get("gluster-log-level") + if rconf.args.subcmd == "slave": + log_level = gconf.get("slave-gluster-log-level") + + params = gconf.get("gluster-params").split() + \ + ['log-level=' + log_level] + \ + ['log-file=' + log_file, 'volfile-server=' + self.host] + \ + ['volfile-id=' + self.volume, 'client-pid=-1'] + + self.mounter = mounter(params) + self.mounter.inhibit(label) + logging.info(lf("Mounted gluster volume", + duration="%.4f" % (time.time() - t0))) def gmaster_instantiate_tuple(self, slave): """return a tuple of the 'one shot' and the 'main crawl' @@ -1204,7 +1145,7 @@ class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote): gmaster_builder()(self, slave), gmaster_builder('changeloghistory')(self, slave)) - def service_loop(self, *args): + def service_loop(self, slave=None): """enter service loop - if slave given, instantiate GMaster and @@ -1212,152 +1153,173 @@ class GLUSTER(AbstractUrl, SlaveLocal, SlaveRemote): master behavior - else do that's what's inherited """ - if args: - slave = args[0] - if gconf.local_path: - class brickserver(FILE.FILEServer): - local_path = gconf.local_path - aggregated = self.server - - @classmethod - def entries(cls, path): - e = super(brickserver, cls).entries(path) - # on the brick don't mess with /.glusterfs - if path == '.': - try: - e.remove('.glusterfs') - except ValueError: - pass - return e - - @classmethod - def lstat(cls, e): - """ path based backend stat """ - return super(brickserver, cls).lstat(e) - - @classmethod - def gfid(cls, e): - """ path based backend gfid fetch """ - return super(brickserver, cls).gfid(e) - - @classmethod - def linkto_check(cls, e): - return super(brickserver, cls).linkto_check(e) - if gconf.slave_id: - # define {,set_}xtime in slave, thus preempting - # the call to remote, so that it takes data from - # the local brick - slave.server.xtime = types.MethodType( - lambda _self, path, uuid: ( - brickserver.xtime(path, - uuid + '.' + gconf.slave_id) - ), - slave.server) - slave.server.stime = types.MethodType( - lambda _self, path, uuid: ( - brickserver.stime(path, - uuid + '.' + gconf.slave_id) - ), - slave.server) - slave.server.set_stime = types.MethodType( - lambda _self, path, uuid, mark: ( - brickserver.set_stime(path, - uuid + '.' + gconf.slave_id, - mark) - ), - slave.server) - (g1, g2, g3) = self.gmaster_instantiate_tuple(slave) - g1.master.server = brickserver - g2.master.server = brickserver - g3.master.server = brickserver - else: - (g1, g2, g3) = self.gmaster_instantiate_tuple(slave) - g1.master.server.aggregated = gmaster.master.server - g2.master.server.aggregated = gmaster.master.server - g3.master.server.aggregated = gmaster.master.server - # bad bad bad: bad way to do things like this - # need to make this elegant - # register the crawlers and start crawling - # g1 ==> Xsync, g2 ==> config.change_detector(changelog by default) - # g3 ==> changelog History - changelog_register_failed = False - (inf, ouf, ra, wa) = gconf.rpc_fd.split(',') - os.close(int(ra)) - os.close(int(wa)) - changelog_agent = RepceClient(int(inf), int(ouf)) - rv = changelog_agent.version() - if int(rv) != CHANGELOG_AGENT_CLIENT_VERSION: + if rconf.args.subcmd == "slave": + if gconf.get("use-rsync-xattrs") and not privileged(): raise GsyncdError( - "RePCe major version mismatch(changelog agent): " - "local %s, remote %s" % - (CHANGELOG_AGENT_CLIENT_VERSION, rv)) + "using rsync for extended attributes is not supported") + + repce = RepceServer( + self.server, sys.stdin, sys.stdout, gconf.get("sync-jobs")) + t = syncdutils.Thread(target=lambda: (repce.service_loop(), + syncdutils.finalize())) + t.start() + logging.info("slave listening") + if gconf.get("slave-timeout") and gconf.get("slave-timeout") > 0: + while True: + lp = self.server.last_keep_alive + time.sleep(gconf.get("slave-timeout")) + if lp == self.server.last_keep_alive: + logging.info( + lf("connection inactive, stopping", + timeout=gconf.get("slave-timeout"))) + break + else: + select((), (), ()) - try: - workdir = g2.setup_working_dir() - # Register only when change_detector is not set to - # xsync, else agent will generate changelog files - # in .processing directory of working dir - if gconf.change_detector != 'xsync': - # register with the changelog library - # 9 == log level (DEBUG) - # 5 == connection retries - changelog_agent.register(gconf.local_path, - workdir, gconf.changelog_log_file, - g2.CHANGELOG_LOG_LEVEL, - g2.CHANGELOG_CONN_RETRIES) - - register_time = int(time.time()) - g2.register(register_time, changelog_agent) - g3.register(register_time, changelog_agent) - except ChangelogException: - changelog_register_failed = True - register_time = None - logging.info("Changelog register failed, fallback to xsync") - - g1.register() - logging.info("Register time: %s" % register_time) - # oneshot: Try to use changelog history api, if not - # available switch to FS crawl - # Note: if config.change_detector is xsync then - # it will not use changelog history api - try: - if not changelog_register_failed: - g3.crawlwrap(oneshot=True) - else: - g1.crawlwrap(oneshot=True) - except (ChangelogException, NoPurgeTimeAvailable, - PartialHistoryAvailable) as e: - if isinstance(e, ChangelogException): - logging.info('Changelog history crawl failed, fallback ' - 'to xsync: %s - %s' % (e.errno, e.strerror)) - elif isinstance(e, PartialHistoryAvailable): - logging.info('Partial history available, using xsync crawl' - ' after consuming history ' - 'till %s' % str(e)) - g1.crawlwrap(oneshot=True, no_stime_update=True, - register_time=register_time) - - # crawl loop: Try changelog crawl, if failed - # switch to FS crawl - try: - if not changelog_register_failed: - g2.crawlwrap() - else: - g1.crawlwrap() - except ChangelogException as e: - logging.info('Changelog crawl failed, fallback to xsync') - g1.crawlwrap() - else: - sup(self, *args) + return + + class brickserver(Server): + local_path = rconf.args.local_path + aggregated = self.server - def rsync(self, files): - return sup(self, files, self.slavedir) + @classmethod + def entries(cls, path): + e = super(brickserver, cls).entries(path) + # on the brick don't mess with /.glusterfs + if path == '.': + try: + e.remove('.glusterfs') + e.remove('.trashcan') + except ValueError: + pass + return e + + @classmethod + def lstat(cls, e): + """ path based backend stat """ + return super(brickserver, cls).lstat(e) + + @classmethod + def gfid(cls, e): + """ path based backend gfid fetch """ + return super(brickserver, cls).gfid(e) + + @classmethod + def linkto_check(cls, e): + return super(brickserver, cls).linkto_check(e) + + # define {,set_}xtime in slave, thus preempting + # the call to remote, so that it takes data from + # the local brick + slave.server.xtime = types.MethodType( + lambda _self, path, uuid: ( + brickserver.xtime(path, + uuid + '.' + rconf.args.slave_id) + ), + slave.server) + slave.server.stime = types.MethodType( + lambda _self, path, uuid: ( + brickserver.stime(path, + uuid + '.' + rconf.args.slave_id) + ), + slave.server) + slave.server.entry_stime = types.MethodType( + lambda _self, path, uuid: ( + brickserver.entry_stime( + path, + uuid + '.' + rconf.args.slave_id) + ), + slave.server) + slave.server.set_stime = types.MethodType( + lambda _self, path, uuid, mark: ( + brickserver.set_stime(path, + uuid + '.' + rconf.args.slave_id, + mark) + ), + slave.server) + slave.server.set_entry_stime = types.MethodType( + lambda _self, path, uuid, mark: ( + brickserver.set_entry_stime( + path, + uuid + '.' + rconf.args.slave_id, + mark) + ), + slave.server) + + (g1, g2, g3) = self.gmaster_instantiate_tuple(slave) + g1.master.server = brickserver + g2.master.server = brickserver + g3.master.server = brickserver + + # bad bad bad: bad way to do things like this + # need to make this elegant + # register the crawlers and start crawling + # g1 ==> Xsync, g2 ==> config.change_detector(changelog by default) + # g3 ==> changelog History + status = GeorepStatus(gconf.get("state-file"), + rconf.args.local_node, + rconf.args.local_path, + rconf.args.local_node_id, + rconf.args.master, + rconf.args.slave) + status.reset_on_worker_start() + + try: + workdir = g2.setup_working_dir() + # Register only when change_detector is not set to + # xsync, else agent will generate changelog files + # in .processing directory of working dir + if gconf.get("change-detector") != 'xsync': + # register with the changelog library + # 9 == log level (DEBUG) + # 5 == connection retries + libgfchangelog.register(rconf.args.local_path, + workdir, + gconf.get("changelog-log-file"), + get_changelog_log_level( + gconf.get("changelog-log-level")), + g2.CHANGELOG_CONN_RETRIES) + + register_time = int(time.time()) + g2.register(register_time, status) + g3.register(register_time, status) + except ChangelogException as e: + logging.error(lf("Changelog register failed", error=e)) + sys.exit(1) + + g1.register(status=status) + logging.info(lf("Register time", + time=register_time)) + # oneshot: Try to use changelog history api, if not + # available switch to FS crawl + # Note: if config.change_detector is xsync then + # it will not use changelog history api + try: + g3.crawlwrap(oneshot=True) + except PartialHistoryAvailable as e: + logging.info(lf('Partial history available, using xsync crawl' + ' after consuming history', + till=e)) + g1.crawlwrap(oneshot=True, register_time=register_time) + except ChangelogHistoryNotAvailable: + logging.info('Changelog history not available, using xsync') + g1.crawlwrap(oneshot=True, register_time=register_time) + except NoStimeAvailable: + logging.info('No stime available, using xsync crawl') + g1.crawlwrap(oneshot=True, register_time=register_time) + except ChangelogException as e: + logging.error(lf("Changelog History Crawl failed", + error=e)) + sys.exit(1) - def tarssh(self, files): - return sup(self, files, self.slavedir) + try: + g2.crawlwrap() + except ChangelogException as e: + logging.error(lf("Changelog crawl failed", error=e)) + sys.exit(1) -class SSH(AbstractUrl, SlaveRemote): +class SSH(object): """scheme class for ssh:// urls @@ -1365,13 +1327,9 @@ class SSH(AbstractUrl, SlaveRemote): implementing an ssh based proxy """ - def __init__(self, path): - self.remote_addr, inner_url = sup(self, path, - '^((?:%s@)?%s):(.+)' % - tuple([r.pattern - for r in (UserRX, HostRX)])) - self.inner_rsc = parse_url(inner_url) - self.volume = inner_url[1:] + def __init__(self, host, volume): + self.remote_addr = host + self.volume = volume @staticmethod def parse_ssh_address(self): @@ -1383,35 +1341,28 @@ class SSH(AbstractUrl, SlaveRemote): self.remotehost = h return {'user': u, 'host': h} - def canonical_path(self): - rap = self.parse_ssh_address(self) - remote_addr = '@'.join([rap['user'], gethostbyname(rap['host'])]) - return ':'.join([remote_addr, self.inner_rsc.get_url(canonical=True)]) - - def can_connect_to(self, remote): - """determine our position in the connectibility matrix""" - return False - - def start_fd_client(self, *a, **opts): - """customizations for client startup + def start_fd_client(self, i, o): + """set up RePCe client, handshake with server - - be a no-op if we are to daemonize (client startup is deferred - to post-daemon stage) - - determine target url for rsync after consulting server + It's cut out as a separate method to let + subclasses hook into client startup """ - if opts.get('deferred'): - return a - sup(self, *a) - ityp = type(self.inner_rsc) - if ityp == FILE: - slavepath = self.inner_rsc.path - elif ityp == GLUSTER: - slavepath = "/proc/%d/cwd" % self.server.pid() - else: - raise NotImplementedError + self.server = RepceClient(i, o) + rv = self.server.__version__() + exrv = {'proto': repce.repce_version, 'object': Server.version()} + da0 = (rv, exrv) + da1 = ({}, {}) + for i in range(2): + for k, v in da0[i].items(): + da1[i][k] = int(v) + if da1[0] != da1[1]: + raise GsyncdError( + "RePCe major version mismatch: local %s, remote %s" % + (exrv, rv)) + slavepath = "/proc/%d/cwd" % self.server.pid() self.slaveurl = ':'.join([self.remote_addr, slavepath]) - def connect_remote(self, go_daemon=None): + def connect_remote(self): """connect to inner slave url through outer ssh url Wrap the connecting utility in ssh. @@ -1429,41 +1380,204 @@ class SSH(AbstractUrl, SlaveRemote): [NB. ATM gluster product does not makes use of interactive authentication.] """ - if go_daemon == 'done': - return self.start_fd_client(*self.fd_pair) - syncdutils.setup_ssh_ctl(tempfile.mkdtemp(prefix='gsyncd-aux-ssh-'), self.remote_addr, - self.inner_rsc.url) - - deferred = go_daemon == 'postconn' - ret = sup(self, gconf.ssh_command.split() + gconf.ssh_ctl_args + - [self.remote_addr], - slave=self.inner_rsc.url, deferred=deferred) - - if deferred: - # send a message to peer so that we can wait for - # the answer from which we know connection is - # established and we can proceed with daemonization - # (doing that too early robs the ssh passwd prompt...) - # However, we'd better not start the RepceClient - # before daemonization (that's not preserved properly - # in daemon), we just do a an ad-hoc linear put/get. - i, o = ret - inf = os.fdopen(i) - repce.send(o, None, '__repce_version__') - select((inf,), (), ()) - repce.recv(inf) - # hack hack hack: store a global reference to the file - # to save it from getting GC'd which implies closing it - gconf.permanent_handles.append(inf) - self.fd_pair = (i, o) - return 'should' - - def rsync(self, files): - return sup(self, files, '-e', - " ".join(gconf.ssh_command.split() + gconf.ssh_ctl_args), - *(gconf.rsync_ssh_options.split() + [self.slaveurl])) - - def tarssh(self, files): - return sup(self, files, self.slaveurl) + self.volume) + + logging.info("Initializing SSH connection between master and slave...") + t0 = time.time() + + extra_opts = [] + remote_gsyncd = gconf.get("remote-gsyncd") + if remote_gsyncd == "": + remote_gsyncd = "/nonexistent/gsyncd" + + if gconf.get("use-rsync-xattrs"): + extra_opts.append('--use-rsync-xattrs') + + args_to_slave = [gconf.get("ssh-command")] + \ + gconf.get("ssh-options").split() + \ + ["-p", str(gconf.get("ssh-port"))] + \ + rconf.ssh_ctl_args + [self.remote_addr] + \ + [remote_gsyncd, "slave"] + \ + extra_opts + \ + [rconf.args.master, rconf.args.slave] + \ + [ + '--master-node', rconf.args.local_node, + '--master-node-id', rconf.args.local_node_id, + '--master-brick', rconf.args.local_path, + '--local-node', rconf.args.resource_remote, + '--local-node-id', rconf.args.resource_remote_id] + \ + [ + # Add all config arguments here, slave gsyncd will not use + # config file in slave side, so all overriding options should + # be sent as arguments + '--slave-timeout', str(gconf.get("slave-timeout")), + '--slave-log-level', gconf.get("slave-log-level"), + '--slave-gluster-log-level', + gconf.get("slave-gluster-log-level"), + '--slave-gluster-command-dir', + gconf.get("slave-gluster-command-dir"), + '--master-dist-count', + str(gconf.get("master-distribution-count"))] + + if gconf.get("slave-access-mount"): + args_to_slave.append('--slave-access-mount') + + if rconf.args.debug: + args_to_slave.append('--debug') + + po = Popen(args_to_slave, + stdin=subprocess.PIPE, stdout=subprocess.PIPE, + stderr=subprocess.PIPE) + rconf.transport = po + self.start_fd_client(po.stdout, po.stdin) + logging.info(lf("SSH connection between master and slave established.", + duration="%.4f" % (time.time() - t0))) + + def rsync(self, files, *args, **kw): + """invoke rsync""" + if not files: + raise GsyncdError("no files to sync") + logging.debug("files: " + ", ".join(files)) + + extra_rsync_flags = [] + # Performance flag, --ignore-missing-args, if rsync version is + # greater than 3.1.0 then include this flag. + if gconf.get("rsync-opt-ignore-missing-args") and \ + get_rsync_version(gconf.get("rsync-command")) >= "3.1.0": + extra_rsync_flags = ["--ignore-missing-args"] + + rsync_ssh_opts = [gconf.get("ssh-command")] + \ + gconf.get("ssh-options").split() + \ + ["-p", str(gconf.get("ssh-port"))] + \ + rconf.ssh_ctl_args + \ + gconf.get("rsync-ssh-options").split() + + argv = [ + gconf.get("rsync-command"), + '-aR0', + '--inplace', + '--files-from=-', + '--super', + '--stats', + '--numeric-ids', + '--no-implied-dirs' + ] + + if gconf.get("rsync-opt-existing"): + argv += ["--existing"] + + if gconf.get("sync-xattrs"): + argv += ['--xattrs'] + + if gconf.get("sync-acls"): + argv += ['--acls'] + + argv = argv + \ + gconf.get("rsync-options").split() + \ + extra_rsync_flags + ['.'] + \ + ["-e", " ".join(rsync_ssh_opts)] + \ + [self.slaveurl] + + log_rsync_performance = gconf.getr("log-rsync-performance", False) + + if log_rsync_performance: + # use stdout=PIPE only when log_rsync_performance enabled + # Else rsync will write to stdout and nobody is there + # to consume. If PIPE is full rsync hangs. + po = Popen(argv, stdin=subprocess.PIPE, stdout=subprocess.PIPE, + stderr=subprocess.PIPE, universal_newlines=True) + else: + po = Popen(argv, stdin=subprocess.PIPE, stderr=subprocess.PIPE, + universal_newlines=True) + + for f in files: + po.stdin.write(f) + po.stdin.write('\0') + + stdout, stderr = po.communicate() + + if kw.get("log_err", False): + for errline in stderr.strip().split("\n")[:-1]: + logging.error(lf("SYNC Error", + sync_engine="Rsync", + error=errline)) + + if log_rsync_performance: + rsync_msg = [] + for line in stdout.split("\n"): + if line.startswith("Number of files:") or \ + line.startswith("Number of regular files transferred:") or \ + line.startswith("Total file size:") or \ + line.startswith("Total transferred file size:") or \ + line.startswith("Literal data:") or \ + line.startswith("Matched data:") or \ + line.startswith("Total bytes sent:") or \ + line.startswith("Total bytes received:") or \ + line.startswith("sent "): + rsync_msg.append(line) + logging.info(lf("rsync performance", + data=", ".join(rsync_msg))) + + return po + + def tarssh(self, files, log_err=False): + """invoke tar+ssh + -z (compress) can be use if needed, but omitting it now + as it results in weird error (tar+ssh errors out (errcode: 2) + """ + if not files: + raise GsyncdError("no files to sync") + logging.debug("files: " + ", ".join(files)) + (host, rdir) = self.slaveurl.split(':') + + tar_cmd = ["tar"] + \ + ["--sparse", "-cf", "-", "--files-from", "-"] + ssh_cmd = gconf.get("ssh-command").split() + \ + gconf.get("ssh-options-tar").split() + \ + ["-p", str(gconf.get("ssh-port"))] + \ + [host, "tar"] + \ + ["--overwrite", "-xf", "-", "-C", rdir] + p0 = Popen(tar_cmd, stdout=subprocess.PIPE, + stdin=subprocess.PIPE, stderr=subprocess.PIPE, + universal_newlines=True) + p1 = Popen(ssh_cmd, stdin=p0.stdout, stderr=subprocess.PIPE, + universal_newlines=True) + for f in files: + p0.stdin.write(f) + p0.stdin.write('\n') + + p0.stdin.close() + p0.stdout.close() # Allow p0 to receive a SIGPIPE if p1 exits. + + # stdin and stdout of p0 is already closed, Reset to None and + # wait for child process to complete + p0.stdin = None + p0.stdout = None + + def wait_for_tar(p0): + _, stderr = p0.communicate() + if log_err: + for errline in stderr.strip().split("\n")[:-1]: + if "No such file or directory" not in errline: + logging.error(lf("SYNC Error", + sync_engine="Tarssh", + error=errline)) + + t = syncdutils.Thread(target=wait_for_tar, args=(p0, )) + # wait for tar to terminate, collecting any errors, further + # waiting for transfer to complete + t.start() + + # wait for ssh process + _, stderr1 = p1.communicate() + t.join() + + if log_err: + for errline in stderr1.strip().split("\n")[:-1]: + logging.error(lf("SYNC Error", + sync_engine="Tarssh", + error=errline)) + + return p1 diff --git a/geo-replication/syncdaemon/subcmds.py b/geo-replication/syncdaemon/subcmds.py new file mode 100644 index 00000000000..b8508532e30 --- /dev/null +++ b/geo-replication/syncdaemon/subcmds.py @@ -0,0 +1,335 @@ +# -*- coding: utf-8 -*- +# +# Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com> +# This file is part of GlusterFS. +# +# This file is licensed to you under your choice of the GNU Lesser +# General Public License, version 3 or any later version (LGPLv3 or +# later), or the GNU General Public License, version 2 (GPLv2), in all +# cases as published by the Free Software Foundation. +# + +from __future__ import print_function +from syncdutils import lf +import logging +import gsyncdconfig as gconf + + +ERROR_CONFIG_INVALID = 2 +ERROR_CONFIG_INVALID_VALUE = 3 +ERROR_CONFIG_NOT_CONFIGURABLE = 4 + + +def subcmd_monitor_status(args): + from gsyncdstatus import set_monitor_status + from rconf import rconf + + set_monitor_status(gconf.get("state-file"), args.status) + rconf.log_exit = False + logging.info(lf("Monitor Status Change", status=args.status)) + + +def subcmd_status(args): + from gsyncdstatus import GeorepStatus + + master_name = args.master.replace(":", "") + slave_data = args.slave.replace("ssh://", "") + + brick_status = GeorepStatus(gconf.get("state-file"), + "", + args.local_path, + "", + master_name, + slave_data, + gconf.get("pid-file")) + checkpoint_time = gconf.get("checkpoint", 0) + brick_status.print_status(checkpoint_time=checkpoint_time, + json_output=args.json) + + +def subcmd_monitor(args): + import monitor + from resource import GLUSTER, SSH, Popen + go_daemon = False if args.debug else True + + monitor.startup(go_daemon) + Popen.init_errhandler() + local = GLUSTER("localhost", args.master) + slavehost, slavevol = args.slave.split("::") + remote = SSH(slavehost, slavevol) + return monitor.monitor(local, remote) + + +def subcmd_verify_spawning(args): + logging.info("Able to spawn gsyncd.py") + + +def subcmd_worker(args): + import os + import fcntl + + from resource import GLUSTER, SSH, Popen + + Popen.init_errhandler() + fcntl.fcntl(args.feedback_fd, fcntl.F_SETFD, fcntl.FD_CLOEXEC) + local = GLUSTER("localhost", args.master) + slave_url, slavevol = args.slave.split("::") + if "@" not in slave_url: + slavehost = args.resource_remote + else: + slavehost = "%s@%s" % (slave_url.split("@")[0], args.resource_remote) + remote = SSH(slavehost, slavevol) + remote.connect_remote() + local.connect() + logging.info("Worker spawn successful. Acknowledging back to monitor") + os.close(args.feedback_fd) + local.service_loop(remote) + + +def subcmd_slave(args): + from resource import GLUSTER, Popen + + Popen.init_errhandler() + slavevol = args.slave.split("::")[-1] + local = GLUSTER("localhost", slavevol) + + local.connect() + local.service_loop() + + +def subcmd_voluuidget(args): + from subprocess import Popen, PIPE + import xml.etree.ElementTree as XET + + ParseError = XET.ParseError if hasattr(XET, 'ParseError') else SyntaxError + + cmd = ['gluster', '--xml', '--remote-host=' + args.host, + 'volume', 'info', args.volname] + + if args.inet6: + cmd.append("--inet6") + + po = Popen(cmd, bufsize=0, + stdin=None, stdout=PIPE, stderr=PIPE, + universal_newlines=True) + + vix, err = po.communicate() + if po.returncode != 0: + logging.info(lf("Volume info failed, unable to get " + "volume uuid of slavevol, " + "returning empty string", + slavevol=args.volname, + slavehost=args.host, + error=po.returncode)) + return "" + vi = XET.fromstring(vix) + if vi.find('opRet').text != '0': + logging.info(lf("Unable to get volume uuid of slavevol, " + "returning empty string", + slavevol=args.volname, + slavehost=args.host, + error=vi.find('opErrstr').text)) + return "" + + try: + voluuid = vi.find("volInfo/volumes/volume/id").text + except (ParseError, AttributeError, ValueError) as e: + logging.info(lf("Parsing failed to volume uuid of slavevol, " + "returning empty string", + slavevol=args.volname, + slavehost=args.host, + error=e)) + voluuid = "" + + print(voluuid) + + +def _unlink(path): + import os + from errno import ENOENT + from syncdutils import GsyncdError + import sys + + try: + os.unlink(path) + except (OSError, IOError): + if sys.exc_info()[1].errno == ENOENT: + pass + else: + raise GsyncdError('Unlink error: %s' % path) + + +def subcmd_delete(args): + import logging + import shutil + import glob + import sys + from errno import ENOENT, ENODATA + import struct + + from syncdutils import GsyncdError, Xattr, errno_wrap + import gsyncdconfig as gconf + + logging.info('geo-replication delete') + # remove the stime xattr from all the brick paths so that + # a re-create of a session will start sync all over again + stime_xattr_prefix = gconf.get('stime-xattr-prefix', None) + + # Delete pid file, status file, socket file + cleanup_paths = [] + cleanup_paths.append(gconf.get("pid-file")) + + # Cleanup Session dir + try: + shutil.rmtree(gconf.get("georep-session-working-dir")) + except (IOError, OSError): + if sys.exc_info()[1].errno == ENOENT: + pass + else: + raise GsyncdError( + 'Error while removing working dir: %s' % + gconf.get("georep-session-working-dir")) + + # Cleanup changelog working dirs + try: + shutil.rmtree(gconf.get("working-dir")) + except (IOError, OSError): + if sys.exc_info()[1].errno == ENOENT: + pass + else: + raise GsyncdError( + 'Error while removing working dir: %s' % + gconf.get("working-dir")) + + for path in cleanup_paths: + # To delete temp files + for f in glob.glob(path + "*"): + _unlink(f) + + if args.reset_sync_time and stime_xattr_prefix: + for p in args.paths: + if p != "": + # set stime to (0,0) to trigger full volume content resync + # to slave on session recreation + # look at master.py::Xcrawl hint: zero_zero + errno_wrap(Xattr.lsetxattr, + (p, stime_xattr_prefix + ".stime", + struct.pack("!II", 0, 0)), + [ENOENT, ENODATA]) + errno_wrap(Xattr.lremovexattr, + (p, stime_xattr_prefix + ".entry_stime"), + [ENOENT, ENODATA]) + + return + + +def print_config(name, value, only_value=False, use_underscore=False): + val = value + if isinstance(value, bool): + val = str(value).lower() + + if only_value: + print(val) + else: + if use_underscore: + name = name.replace("-", "_") + + print(("%s:%s" % (name, val))) + + +def config_name_format(val): + return val.replace("_", "-") + + +def subcmd_config_get(args): + import sys + import json + + all_config = gconf.getall(show_defaults=args.show_defaults, + show_non_configurable=True) + if args.name is not None: + val = all_config.get(config_name_format(args.name), None) + if val is None: + sys.stderr.write("Invalid config name \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_INVALID) + + print_config(args.name, val["value"], only_value=args.only_value, + use_underscore=args.use_underscore) + return + + if args.json: + out = [] + # Convert all values as string + for k in sorted(all_config): + v = all_config[k] + out.append({ + "name": k, + "value": str(v["value"]), + "default": str(v["default"]), + "configurable": v["configurable"], + "modified": v["modified"] + }) + + print((json.dumps(out))) + return + + for k in sorted(all_config): + print_config(k, all_config[k]["value"], + use_underscore=args.use_underscore) + + +def subcmd_config_check(args): + import sys + + try: + gconf.check(config_name_format(args.name), value=args.value, + with_conffile=False) + except gconf.GconfNotConfigurable: + cnf_val = gconf.get(config_name_format(args.name), None) + if cnf_val is None: + sys.stderr.write("Invalid config name \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_INVALID) + + # Not configurable + sys.stderr.write("Not configurable \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_NOT_CONFIGURABLE) + except gconf.GconfInvalidValue: + sys.stderr.write("Invalid config value \"%s=%s\"\n" % (args.name, + args.value)) + sys.exit(ERROR_CONFIG_INVALID_VALUE) + + +def subcmd_config_set(args): + import sys + + try: + gconf.setconfig(config_name_format(args.name), args.value) + except gconf.GconfNotConfigurable: + cnf_val = gconf.get(config_name_format(args.name), None) + if cnf_val is None: + sys.stderr.write("Invalid config name \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_INVALID) + + # Not configurable + sys.stderr.write("Not configurable \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_NOT_CONFIGURABLE) + except gconf.GconfInvalidValue: + sys.stderr.write("Invalid config value \"%s=%s\"\n" % (args.name, + args.value)) + sys.exit(ERROR_CONFIG_INVALID_VALUE) + + +def subcmd_config_reset(args): + import sys + + try: + gconf.resetconfig(config_name_format(args.name)) + except gconf.GconfNotConfigurable: + cnf_val = gconf.get(config_name_format(args.name), None) + if cnf_val is None: + sys.stderr.write("Invalid config name \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_INVALID) + + # Not configurable + sys.stderr.write("Not configurable \"%s\"\n" % args.name) + sys.exit(ERROR_CONFIG_NOT_CONFIGURABLE) diff --git a/geo-replication/syncdaemon/syncdutils.py b/geo-replication/syncdaemon/syncdutils.py index 9eda6044472..a3df103e76c 100644 --- a/geo-replication/syncdaemon/syncdutils.py +++ b/geo-replication/syncdaemon/syncdutils.py @@ -15,53 +15,105 @@ import time import fcntl import shutil import logging +import errno +import threading +import subprocess import socket +from subprocess import PIPE from threading import Lock, Thread as baseThread -from errno import EACCES, EAGAIN, EPIPE, ENOTCONN, ECONNABORTED -from errno import EINTR, ENOENT, EPERM, ESTALE, errorcode +from errno import (EACCES, EAGAIN, EPIPE, ENOTCONN, ENOMEM, ECONNABORTED, + EINTR, ENOENT, ESTALE, EBUSY, ENODATA, errorcode, EIO) from signal import signal, SIGTERM import select as oselect from os import waitpid as owaitpid - +import xml.etree.ElementTree as XET +from select import error as SelectError try: from cPickle import PickleError except ImportError: - # py 3 from pickle import PickleError -from gconf import gconf - +from conf import GLUSTERFS_LIBEXECDIR, UUID_FILE +sys.path.insert(1, GLUSTERFS_LIBEXECDIR) +EVENTS_ENABLED = True try: - # py 3 - from urllib import parse as urllib + from gfevents.eventtypes import GEOREP_FAULTY as EVENT_GEOREP_FAULTY + from gfevents.eventtypes import GEOREP_ACTIVE as EVENT_GEOREP_ACTIVE + from gfevents.eventtypes import GEOREP_PASSIVE as EVENT_GEOREP_PASSIVE + from gfevents.eventtypes import GEOREP_CHECKPOINT_COMPLETED \ + as EVENT_GEOREP_CHECKPOINT_COMPLETED except ImportError: - import urllib + # Events APIs not installed, dummy eventtypes with None + EVENTS_ENABLED = False + EVENT_GEOREP_FAULTY = None + EVENT_GEOREP_ACTIVE = None + EVENT_GEOREP_PASSIVE = None + EVENT_GEOREP_CHECKPOINT_COMPLETED = None -try: - from hashlib import md5 as md5 -except ImportError: - # py 2.4 - from md5 import new as md5 +import gsyncdconfig as gconf +from rconf import rconf -# auxillary gfid based access prefix +from hashlib import sha256 as sha256 + +ENOTSUP = getattr(errno, 'ENOTSUP', 'EOPNOTSUPP') + +# auxiliary gfid based access prefix _CL_AUX_GFID_PFX = ".gfid/" -GF_OP_RETRIES = 20 +ROOT_GFID = "00000000-0000-0000-0000-000000000001" +GF_OP_RETRIES = 10 + +GX_GFID_CANONICAL_LEN = 37 # canonical gfid len + '\0' + +NodeID = None +rsync_version = None +unshare_mnt_propagation = None +slv_bricks = None +SPACE_ESCAPE_CHAR = "%20" +NEWLINE_ESCAPE_CHAR = "%0A" +PERCENTAGE_ESCAPE_CHAR = "%25" -CHANGELOG_AGENT_SERVER_VERSION = 1.0 -CHANGELOG_AGENT_CLIENT_VERSION = 1.0 +final_lock = Lock() + +def sup(x, *a, **kw): + """a rubyesque "super" for python ;) + + invoke caller method in parent class with given args. + """ + return getattr(super(type(x), x), + sys._getframe(1).f_code.co_name)(*a, **kw) def escape(s): """the chosen flavor of string escaping, used all over to turn whatever data to creatable representation""" - return urllib.quote_plus(s) + return s.replace("/", "-").strip("-") -def unescape(s): - """inverse of .escape""" - return urllib.unquote_plus(s) +def escape_space_newline(s): + return s.replace("%", PERCENTAGE_ESCAPE_CHAR)\ + .replace(" ", SPACE_ESCAPE_CHAR)\ + .replace("\n", NEWLINE_ESCAPE_CHAR) +def unescape_space_newline(s): + return s.replace(SPACE_ESCAPE_CHAR, " ")\ + .replace(NEWLINE_ESCAPE_CHAR, "\n")\ + .replace(PERCENTAGE_ESCAPE_CHAR, "%") + +# gf_mount_ready() returns 1 if all subvols are up, else 0 +def gf_mount_ready(): + ret = errno_wrap(Xattr.lgetxattr, + ['.', 'dht.subvol.status', 16], + [ENOENT, ENOTSUP, ENODATA], [ENOMEM]) + + if isinstance(ret, int): + logging.error("failed to get the xattr value") + return 1 + ret = ret.rstrip('\x00') + if ret == "1": + return 1 + return 0 + def norm(s): if s: return s.replace('-', '_') @@ -119,17 +171,26 @@ def setup_ssh_ctl(ctld, remote_addr, resource_url): """ Setup GConf ssh control path parameters """ - gconf.ssh_ctl_dir = ctld + rconf.ssh_ctl_dir = ctld content = "SLAVE_HOST=%s\nSLAVE_RESOURCE_URL=%s" % (remote_addr, resource_url) - content_md5 = md5hex(content) - fname = os.path.join(gconf.ssh_ctl_dir, - "%s.mft" % content_md5) + encoded_content = content.encode() + content_sha256 = sha256hex(encoded_content) + """ + The length of ctl_path for ssh connection should not be > 108. + ssh fails with ctl_path too long if it is so. But when rsync + is piped to ssh, it is not taking > 90. Hence using first 32 + bytes of hash. Hash collision doesn't matter as only one sock + file is created per directory. + """ + content_sha256 = content_sha256[:32] + fname = os.path.join(rconf.ssh_ctl_dir, + "%s.mft" % content_sha256) - create_manifest(fname, content) - ssh_ctl_path = os.path.join(gconf.ssh_ctl_dir, - "%s.sock" % content_md5) - gconf.ssh_ctl_args = ["-oControlMaster=auto", "-S", ssh_ctl_path] + create_manifest(fname, encoded_content) + ssh_ctl_path = os.path.join(rconf.ssh_ctl_dir, + "%s.sock" % content_sha256) + rconf.ssh_ctl_args = ["-oControlMaster=auto", "-S", ssh_ctl_path] def grabfile(fname, content=None): @@ -139,7 +200,7 @@ def grabfile(fname, content=None): """ # damn those messy open() mode codes fd = os.open(fname, os.O_CREAT | os.O_RDWR) - f = os.fdopen(fd, 'r+b', 0) + f = os.fdopen(fd, 'r+') try: fcntl.lockf(f, fcntl.LOCK_EX | fcntl.LOCK_NB) except: @@ -153,34 +214,34 @@ def grabfile(fname, content=None): try: f.truncate() f.write(content) + f.flush() except: f.close() raise - gconf.permanent_handles.append(f) + rconf.permanent_handles.append(f) return f def grabpidfile(fname=None, setpid=True): """.grabfile customization for pid files""" if not fname: - fname = gconf.pid_file + fname = gconf.get("pid-file") content = None if setpid: content = str(os.getpid()) + '\n' return grabfile(fname, content=content) -final_lock = Lock() - -def finalize(*a, **kw): +def finalize(*args, **kwargs): """all those messy final steps we go trough upon termination Do away with pidfile, ssh control dir and logging. """ + final_lock.acquire() - if getattr(gconf, 'pid_file', None): - rm_pidf = gconf.pid_file_owned - if gconf.cpid: + if gconf.get('pid_file'): + rm_pidf = rconf.pid_file_owned + if rconf.cpid: # exit path from parent branch of daemonization rm_pidf = False while True: @@ -188,33 +249,50 @@ def finalize(*a, **kw): if not f: # child has already taken over pidfile break - if os.waitpid(gconf.cpid, os.WNOHANG)[0] == gconf.cpid: + if os.waitpid(rconf.cpid, os.WNOHANG)[0] == rconf.cpid: # child has terminated rm_pidf = True break time.sleep(0.1) if rm_pidf: try: - os.unlink(gconf.pid_file) + os.unlink(rconf.pid_file) except: ex = sys.exc_info()[1] if ex.errno == ENOENT: pass else: raise - if gconf.ssh_ctl_dir and not gconf.cpid: - shutil.rmtree(gconf.ssh_ctl_dir) - if getattr(gconf, 'state_socket', None): - try: - os.unlink(gconf.state_socket) - except: - if sys.exc_info()[0] == OSError: + if rconf.ssh_ctl_dir and not rconf.cpid: + def handle_rm_error(func, path, exc_info): + if exc_info[1].errno == ENOENT: + return + raise exc_info[1] + + shutil.rmtree(rconf.ssh_ctl_dir, onerror=handle_rm_error) + + """ Unmount if not done """ + if rconf.mount_point: + if rconf.mountbroker: + umount_cmd = rconf.mbr_umount_cmd + [rconf.mount_point, 'lazy'] + else: + umount_cmd = ['umount', '-l', rconf.mount_point] + p0 = subprocess.Popen(umount_cmd, stderr=subprocess.PIPE, + universal_newlines=True) + _, errdata = p0.communicate() + if p0.returncode == 0: + try: + os.rmdir(rconf.mount_point) + except OSError: pass - if gconf.log_exit: + else: + pass + + if rconf.log_exit: logging.info("exiting.") sys.stdout.flush() sys.stderr.flush() - os._exit(kw.get('exval', 0)) + os._exit(kwargs.get('exval', 0)) def log_raise_exception(excont): @@ -224,6 +302,7 @@ def log_raise_exception(excont): Translate some weird sounding but well understood exceptions into human-friendly lingo """ + is_filelog = False for h in logging.getLogger().handlers: fno = getattr(getattr(h, 'stream', None), 'fileno', None) @@ -244,31 +323,45 @@ def log_raise_exception(excont): ((isinstance(exc, OSError) or isinstance(exc, IOError)) and exc.errno == EPIPE): logging.error('connection to peer is broken') - if hasattr(gconf, 'transport'): - gconf.transport.wait() - if gconf.transport.returncode == 127: - logging.warn("!!!!!!!!!!!!!") - logging.warn('!!! getting "No such file or directory" ' - "errors is most likely due to " - "MISCONFIGURATION" - ", please consult https://access.redhat.com" - "/site/documentation/en-US/Red_Hat_Storage" - "/2.1/html/Administration_Guide" - "/chap-User_Guide-Geo_Rep-Preparation-" - "Settingup_Environment.html") - logging.warn("!!!!!!!!!!!!!") - gconf.transport.terminate_geterr() + if hasattr(rconf, 'transport'): + rconf.transport.wait() + if rconf.transport.returncode == 127: + logging.error("getting \"No such file or directory\"" + "errors is most likely due to " + "MISCONFIGURATION, please remove all " + "the public keys added by geo-replication " + "from authorized_keys file in slave nodes " + "and run Geo-replication create " + "command again.") + logging.error("If `gsec_create container` was used, then " + "run `gluster volume geo-replication " + "<MASTERVOL> [<SLAVEUSER>@]<SLAVEHOST>::" + "<SLAVEVOL> config remote-gsyncd " + "<GSYNCD_PATH> (Example GSYNCD_PATH: " + "`/usr/libexec/glusterfs/gsyncd`)") + rconf.transport.terminate_geterr() elif isinstance(exc, OSError) and exc.errno in (ENOTCONN, ECONNABORTED): - logging.error('glusterfs session went down [%s]', - errorcode[exc.errno]) + logging.error(lf('Gluster Mount process exited', + error=errorcode[exc.errno])) + elif isinstance(exc, OSError) and exc.errno == EIO: + logging.error("Getting \"Input/Output error\" " + "is most likely due to " + "a. Brick is down or " + "b. Split brain issue.") + logging.error("This is expected as per design to " + "keep the consistency of the file system. " + "Once the above issue is resolved " + "geo-replication would automatically " + "proceed further.") + logtag = "FAIL" else: logtag = "FAIL" if not logtag and logging.getLogger().isEnabledFor(logging.DEBUG): logtag = "FULL EXCEPTION TRACE" if logtag: logging.exception(logtag + ": ") - sys.stderr.write("failed with %s.\n" % type(exc).__name__) + sys.stderr.write("failed with %s: %s.\n" % (type(exc).__name__, exc)) excont.exval = 1 sys.exit(excont.exval) @@ -291,20 +384,20 @@ class Thread(baseThread): function coughs up an exception """ - def __init__(self, *a, **kw): - tf = kw.get('target') + def __init__(self, *args, **kwargs): + tf = kwargs.get('target') if tf: - def twrap(*aa): + def twrap(*aargs): excont = FreeObject(exval=0) try: - tf(*aa) + tf(*aargs) except: try: log_raise_exception(excont) finally: finalize(exval=excont.exval) - kw['target'] = twrap - baseThread.__init__(self, *a, **kw) + kwargs['target'] = twrap + baseThread.__init__(self, *args, **kwargs) self.setDaemon(True) @@ -312,6 +405,33 @@ class GsyncdError(Exception): pass +class _MetaXattr(object): + + """singleton class, a lazy wrapper around the + libcxattr module + + libcxattr (a heavy import due to ctypes) is + loaded only when when the single + instance is tried to be used. + + This reduces runtime for those invocations + which do not need filesystem manipulation + (eg. for config, url parsing) + """ + + def __getattr__(self, meth): + from libcxattr import Xattr as LXattr + xmeth = [m for m in dir(LXattr) if m[0] != '_'] + if meth not in xmeth: + return + for m in xmeth: + setattr(self, m, getattr(LXattr, m)) + return getattr(self, meth) + + +Xattr = _MetaXattr() + + def getusername(uid=None): if uid is None: uid = os.geteuid() @@ -342,81 +462,63 @@ def boolify(s): lstr = s.lower() if lstr in true_list: rv = True - elif not lstr in false_list: - logging.warn("Unknown string (%s) in string to boolean conversion " - "defaulting to False\n" % (s)) + elif lstr not in false_list: + logging.warn(lf("Unknown string in \"string to boolean\" conversion, " + "defaulting to False", + str=s)) return rv -def eintr_wrap(func, exc, *a): +def eintr_wrap(func, exc, *args): """ wrapper around syscalls resilient to interrupt caused by signals """ while True: try: - return func(*a) + return func(*args) except exc: ex = sys.exc_info()[1] if not ex.args[0] == EINTR: raise -def select(*a): - return eintr_wrap(oselect.select, oselect.error, *a) +def select(*args): + return eintr_wrap(oselect.select, oselect.error, *args) + +def waitpid(*args): + return eintr_wrap(owaitpid, OSError, *args) -def waitpid(*a): - return eintr_wrap(owaitpid, OSError, *a) +def term_handler_default_hook(signum, frame): + finalize(signum, frame, exval=1) -def set_term_handler(hook=lambda *a: finalize(*a, **{'exval': 1})): + +def set_term_handler(hook=term_handler_default_hook): signal(SIGTERM, hook) -def is_host_local(host): - locaddr = False - for ai in socket.getaddrinfo(host, None): - # cf. http://github.com/gluster/glusterfs/blob/ce111f47/xlators - # /mgmt/glusterd/src/glusterd-utils.c#L125 - if ai[0] == socket.AF_INET: - if ai[-1][0].split(".")[0] == "127": - locaddr = True - break - elif ai[0] == socket.AF_INET6: - if ai[-1][0] == "::1": - locaddr = True +def get_node_uuid(): + global NodeID + if NodeID is not None: + return NodeID + + NodeID = "" + with open(UUID_FILE) as f: + for line in f: + if line.startswith("UUID="): + NodeID = line.strip().split("=")[-1] break - else: - continue - try: - # use ICMP socket to avoid net.ipv4.ip_nonlocal_bind issue, - # cf. https://bugzilla.redhat.com/show_bug.cgi?id=890587 - s = socket.socket(ai[0], socket.SOCK_RAW, socket.IPPROTO_ICMP) - except socket.error: - ex = sys.exc_info()[1] - if ex.errno != EPERM: - raise - f = None - try: - f = open("/proc/sys/net/ipv4/ip_nonlocal_bind") - if int(f.read()) != 0: - raise GsyncdError( - "non-local bind is set and not allowed to create " - "raw sockets, cannot determine if %s is local" % host) - s = socket.socket(ai[0], socket.SOCK_DGRAM) - finally: - if f: - f.close() - try: - s.bind(ai[-1]) - locaddr = True - break - except: - pass - s.close() - return locaddr + + if NodeID == "": + raise GsyncdError("Failed to get Host UUID from %s" % UUID_FILE) + return NodeID + + +def is_host_local(host_id): + return host_id == get_node_uuid() def funcode(f): @@ -452,17 +554,16 @@ def gauxpfx(): return _CL_AUX_GFID_PFX -def md5hex(s): - return md5(s).hexdigest() +def sha256hex(s): + return sha256(s).hexdigest() def selfkill(sig=SIGTERM): os.kill(os.getpid(), sig) -def errno_wrap(call, arg=[], errnos=[], retry_errnos=[ESTALE]): +def errno_wrap(call, arg=[], errnos=[], retry_errnos=[]): """ wrapper around calls resilient to errnos. - retry in case of ESTALE by default. """ nr_tries = 0 while True: @@ -472,28 +573,39 @@ def errno_wrap(call, arg=[], errnos=[], retry_errnos=[ESTALE]): ex = sys.exc_info()[1] if ex.errno in errnos: return ex.errno - if not ex.errno in retry_errnos: + if ex.errno not in retry_errnos: raise nr_tries += 1 if nr_tries == GF_OP_RETRIES: # probably a screwed state, cannot do much... - logging.warn('reached maximum retries (%s)...' % repr(arg)) - return + logging.warn(lf('reached maximum retries', + args=repr(arg), + error=ex)) + raise time.sleep(0.250) # retry the call def lstat(e): - try: - return os.lstat(e) - except (IOError, OSError): - ex = sys.exc_info()[1] - if ex.errno == ENOENT: - return ex.errno - else: - raise + return errno_wrap(os.lstat, [e], [ENOENT], [ESTALE, EBUSY]) + +def get_gfid_from_mnt(gfidpath): + return errno_wrap(Xattr.lgetxattr, + [gfidpath, 'glusterfs.gfid.string', + GX_GFID_CANONICAL_LEN], [ENOENT], [ESTALE]) + +def matching_disk_gfid(gfid, entry): + disk_gfid = get_gfid_from_mnt(entry) + if isinstance(disk_gfid, int): + return False -class NoPurgeTimeAvailable(Exception): + if not gfid == disk_gfid: + return False + + return True + + +class NoStimeAvailable(Exception): pass @@ -501,5 +613,503 @@ class PartialHistoryAvailable(Exception): pass +class ChangelogHistoryNotAvailable(Exception): + pass + + class ChangelogException(OSError): pass + + +def gf_event(event_type, **kwargs): + if EVENTS_ENABLED: + from gfevents.gf_event import gf_event as gfevent + gfevent(event_type, **kwargs) + + +class GlusterLogLevel(object): + NONE = 0 + EMERG = 1 + ALERT = 2 + CRITICAL = 3 + ERROR = 4 + WARNING = 5 + NOTICE = 6 + INFO = 7 + DEBUG = 8 + TRACE = 9 + + +def get_changelog_log_level(lvl): + return getattr(GlusterLogLevel, lvl, GlusterLogLevel.INFO) + + +def get_master_and_slave_data_from_args(args): + master_name = None + slave_data = None + for arg in args: + if arg.startswith(":"): + master_name = arg.replace(":", "") + if "::" in arg: + slave_data = arg.replace("ssh://", "") + + return (master_name, slave_data) + +def unshare_propagation_supported(): + global unshare_mnt_propagation + if unshare_mnt_propagation is not None: + return unshare_mnt_propagation + + unshare_mnt_propagation = False + p = subprocess.Popen(["unshare", "--help"], + stderr=subprocess.PIPE, + stdout=subprocess.PIPE, + universal_newlines=True) + out, err = p.communicate() + if p.returncode == 0: + if "propagation" in out: + unshare_mnt_propagation = True + + return unshare_mnt_propagation + + +def get_rsync_version(rsync_cmd): + global rsync_version + if rsync_version is not None: + return rsync_version + + rsync_version = "0" + p = subprocess.Popen([rsync_cmd, "--version"], + stderr=subprocess.PIPE, + stdout=subprocess.PIPE, + universal_newlines=True) + out, err = p.communicate() + if p.returncode == 0: + rsync_version = out.split(" ", 4)[3] + + return rsync_version + + +def get_slv_dir_path(slv_host, slv_volume, gfid): + global slv_bricks + + dir_path = ENOENT + pfx = gauxpfx() + + if not slv_bricks: + slv_info = Volinfo(slv_volume, slv_host, master=False) + slv_bricks = slv_info.bricks + # Result of readlink would be of format as below. + # readlink = "../../pgfid[0:2]/pgfid[2:4]/pgfid/basename" + for brick in slv_bricks: + dir_path = errno_wrap(os.path.join, + [brick['dir'], + ".glusterfs", gfid[0:2], + gfid[2:4], + gfid], [ENOENT], [ESTALE]) + if dir_path != ENOENT: + try: + realpath = errno_wrap(os.readlink, [dir_path], + [ENOENT], [ESTALE]) + if not isinstance(realpath, int): + realpath_parts = realpath.split('/') + pargfid = realpath_parts[-2] + basename = realpath_parts[-1] + dir_entry = os.path.join(pfx, pargfid, basename) + return dir_entry + except OSError: + # .gfid/GFID + gfidpath = unescape_space_newline(os.path.join(pfx, gfid)) + realpath = errno_wrap(Xattr.lgetxattr_buf, + [gfidpath, 'glusterfs.gfid2path'], [ENOENT], [ESTALE]) + if not isinstance(realpath, int): + basename = os.path.basename(realpath).rstrip('\x00') + dirpath = os.path.dirname(realpath) + if dirpath == "/": + pargfid = ROOT_GFID + else: + dirpath = dirpath.strip("/") + pargfid = get_gfid_from_mnt(dirpath) + if isinstance(pargfid, int): + return None + dir_entry = os.path.join(pfx, pargfid, basename) + return dir_entry + + return None + + +def lf(event, **kwargs): + """ + Log Format helper function, log messages can be + easily modified to structured log format. + lf("Config Change", sync_jobs=4, brick=/bricks/b1) will be + converted as "Config Change [{brick=/bricks/b1}, {sync_jobs=4}]" + """ + msgparts = [] + for k, v in kwargs.items(): + msgparts.append("{%s=%s}" % (k, v)) + return "%s [%s]" % (event, ", ".join(msgparts)) + + +class Popen(subprocess.Popen): + + """customized subclass of subprocess.Popen with a ring + buffer for children error output""" + + @classmethod + def init_errhandler(cls): + """start the thread which handles children's error output""" + cls.errstore = {} + + def tailer(): + while True: + errstore = cls.errstore.copy() + try: + poe, _, _ = select( + [po.stderr for po in errstore], [], [], 1) + except (ValueError, SelectError): + # stderr is already closed wait for some time before + # checking next error + time.sleep(0.5) + continue + for po in errstore: + if po.stderr not in poe: + continue + po.lock.acquire() + try: + if po.on_death_row: + continue + la = errstore[po] + try: + fd = po.stderr.fileno() + except ValueError: # file is already closed + time.sleep(0.5) + continue + + try: + l = os.read(fd, 1024) + except OSError: + time.sleep(0.5) + continue + + if not l: + continue + tots = len(l) + for lx in la: + tots += len(lx) + while tots > 1 << 20 and la: + tots -= len(la.pop(0)) + la.append(l) + finally: + po.lock.release() + t = Thread(target=tailer) + t.start() + cls.errhandler = t + + @classmethod + def fork(cls): + """fork wrapper that restarts errhandler thread in child""" + pid = os.fork() + if not pid: + cls.init_errhandler() + return pid + + def __init__(self, args, *a, **kw): + """customizations for subprocess.Popen instantiation + + - 'close_fds' is taken to be the default + - if child's stderr is chosen to be managed, + register it with the error handler thread + """ + self.args = args + if 'close_fds' not in kw: + kw['close_fds'] = True + self.lock = threading.Lock() + self.on_death_row = False + self.elines = [] + try: + sup(self, args, *a, **kw) + except: + ex = sys.exc_info()[1] + if not isinstance(ex, OSError): + raise + raise GsyncdError("""execution of "%s" failed with %s (%s)""" % + (args[0], errno.errorcode[ex.errno], + os.strerror(ex.errno))) + if kw.get('stderr') == subprocess.PIPE: + assert(getattr(self, 'errhandler', None)) + self.errstore[self] = [] + + def errlog(self): + """make a log about child's failure event""" + logging.error(lf("command returned error", + cmd=" ".join(self.args), + error=self.returncode)) + lp = '' + + def logerr(l): + logging.error(self.args[0] + "> " + l) + for l in self.elines: + ls = l.split('\n') + ls[0] = lp + ls[0] + lp = ls.pop() + for ll in ls: + logerr(ll) + if lp: + logerr(lp) + + def errfail(self): + """fail nicely if child did not terminate with success""" + self.errlog() + finalize(exval=1) + + def terminate_geterr(self, fail_on_err=True): + """kill child, finalize stderr harvesting (unregister + from errhandler, set up .elines), fail on error if + asked for + """ + self.lock.acquire() + try: + self.on_death_row = True + finally: + self.lock.release() + elines = self.errstore.pop(self) + if self.poll() is None: + self.terminate() + if self.poll() is None: + time.sleep(0.1) + self.kill() + self.wait() + while True: + if not select([self.stderr], [], [], 0.1)[0]: + break + b = os.read(self.stderr.fileno(), 1024) + if b: + elines.append(b.decode()) + else: + break + self.stderr.close() + self.elines = elines + if fail_on_err and self.returncode != 0: + self.errfail() + + +def host_brick_split(value): + """ + IPv6 compatible way to split and get the host + and brick information. Example inputs: + node1.example.com:/exports/bricks/brick1/brick + fe80::af0f:df82:844f:ef66%utun0:/exports/bricks/brick1/brick + """ + parts = value.split(":") + brick = parts[-1] + hostparts = parts[0:-1] + return (":".join(hostparts), brick) + + +class Volinfo(object): + + def __init__(self, vol, host='localhost', prelude=[], master=True): + if master: + gluster_cmd_dir = gconf.get("gluster-command-dir") + else: + gluster_cmd_dir = gconf.get("slave-gluster-command-dir") + + gluster_cmd = os.path.join(gluster_cmd_dir, 'gluster') + po = Popen(prelude + [gluster_cmd, '--xml', '--remote-host=' + host, + 'volume', 'info', vol], + stdout=PIPE, stderr=PIPE, universal_newlines=True) + vix = po.stdout.read() + po.wait() + po.terminate_geterr() + vi = XET.fromstring(vix) + if vi.find('opRet').text != '0': + if prelude: + via = '(via %s) ' % prelude.join(' ') + else: + via = ' ' + raise GsyncdError('getting volume info of %s%s ' + 'failed with errorcode %s' % + (vol, via, vi.find('opErrno').text)) + self.tree = vi + self.volume = vol + self.host = host + + def get(self, elem): + return self.tree.findall('.//' + elem) + + def is_tier(self): + return (self.get('typeStr')[0].text == 'Tier') + + def is_hot(self, brickpath): + logging.debug('brickpath: ' + repr(brickpath)) + return brickpath in self.hot_bricks + + @property + @memoize + def bricks(self): + def bparse(b): + host, dirp = host_brick_split(b.find("name").text) + return {'host': host, 'dir': dirp, 'uuid': b.find("hostUuid").text} + return [bparse(b) for b in self.get('brick')] + + @property + @memoize + def uuid(self): + ids = self.get('id') + if len(ids) != 1: + raise GsyncdError("volume info of %s obtained from %s: " + "ambiguous uuid" % (self.volume, self.host)) + return ids[0].text + + def replica_count(self, tier, hot): + if (tier and hot): + return int(self.get('hotBricks/hotreplicaCount')[0].text) + elif (tier and not hot): + return int(self.get('coldBricks/coldreplicaCount')[0].text) + else: + return int(self.get('replicaCount')[0].text) + + def disperse_count(self, tier, hot): + if (tier and hot): + # Tiering doesn't support disperse volume as hot brick, + # hence no xml output, so returning 0. In case, if it's + # supported later, we should change here. + return 0 + elif (tier and not hot): + return int(self.get('coldBricks/colddisperseCount')[0].text) + else: + return int(self.get('disperseCount')[0].text) + + def distribution_count(self, tier, hot): + if (tier and hot): + return int(self.get('hotBricks/hotdistCount')[0].text) + elif (tier and not hot): + return int(self.get('coldBricks/colddistCount')[0].text) + else: + return int(self.get('distCount')[0].text) + + @property + @memoize + def hot_bricks(self): + return [b.text for b in self.get('hotBricks/brick')] + + def get_hot_bricks_count(self, tier): + if (tier): + return int(self.get('hotBricks/hotbrickCount')[0].text) + else: + return 0 + + +class VolinfoFromGconf(object): + # Glusterd will generate following config items before Geo-rep start + # So that Geo-rep need not run gluster commands from inside + # Volinfo object API/interface kept as is so that caller need not + # change anything except calling this instead of Volinfo() + # + # master-bricks= + # master-bricks=NODEID:HOSTNAME:PATH,.. + # slave-bricks=NODEID:HOSTNAME,.. + # master-volume-id= + # slave-volume-id= + # master-replica-count= + # master-disperse_count= + def __init__(self, vol, host='localhost', master=True): + self.volume = vol + self.host = host + self.master = master + + def is_tier(self): + return False + + def is_hot(self, brickpath): + return False + + def is_uuid(self, value): + try: + uuid.UUID(value) + return True + except ValueError: + return False + + def possible_path(self, value): + return "/" in value + + @property + @memoize + def bricks(self): + pfx = "master-" if self.master else "slave-" + bricks_data = gconf.get(pfx + "bricks") + if bricks_data is None: + return [] + + bricks_data = bricks_data.split(",") + bricks_data = [b.strip() for b in bricks_data] + out = [] + for b in bricks_data: + parts = b.split(":") + b_uuid = None + if self.is_uuid(parts[0]): + b_uuid = parts[0] + # Set all parts except first + parts = parts[1:] + + if self.possible_path(parts[-1]): + bpath = parts[-1] + # Set all parts except last + parts = parts[0:-1] + + out.append({ + "host": ":".join(parts), # if remaining parts are IPv6 name + "dir": bpath, + "uuid": b_uuid + }) + + return out + + @property + @memoize + def uuid(self): + if self.master: + return gconf.get("master-volume-id") + else: + return gconf.get("slave-volume-id") + + def replica_count(self, tier, hot): + return gconf.get("master-replica-count") + + def disperse_count(self, tier, hot): + return gconf.get("master-disperse-count") + + def distribution_count(self, tier, hot): + return gconf.get("master-distribution-count") + + @property + @memoize + def hot_bricks(self): + return [] + + def get_hot_bricks_count(self, tier): + return 0 + + +def can_ssh(host, port=22): + s = socket.socket(socket.AF_INET, socket.SOCK_STREAM) + try: + s.connect((host, port)) + flag = True + except socket.error: + flag = False + + s.close() + return flag + + +def get_up_nodes(hosts, port): + # List of hosts with Hostname/IP and UUID + up_nodes = [] + for h in hosts: + if can_ssh(h[0], port): + up_nodes.append(h) + + return up_nodes |