summaryrefslogtreecommitdiffstats
path: root/geo-replication/syncdaemon/monitor.py
blob: ca7dd055d4b4e0458e1fb88bc1c3c5662aa98414 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
import os
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
from gconf import gconf
from syncdutils import update_file, select, waitpid, set_term_handler, is_host_local, GsyncdError
from syncdutils import escape, Thread, finalize, memoize

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


class Monitor(object):
    """class which spawns and manages gsyncd workers"""

    ST_INIT     = 'Initializing...'
    ST_STABLE   = 'Stable'
    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:
            logging.info('new state: %s' % state)
            if getattr(gconf, 'state_file', None):
                update_file(gconf.state_file, lambda f: f.write(state + '\n'))

    @staticmethod
    def terminate():
        # relax one SIGTERM by setting a handler that sets back
        # standard handler
        set_term_handler(lambda *a: set_term_handler())
        # give a chance to graceful exit
        os.kill(-os.getpid(), signal.SIGTERM)

    def monitor(self, w, argv, cpids):
        """the monitor loop

        Basic logic is a blantantly simple blunt heuristics:
        if spawned client survives 60 secs, it's considered OK.
        This servers us pretty well as it's not vulneralbe to
        any kind of irregular behavior of the child...

        ... well, except for one: if children is hung up on
        waiting for some event, it can survive aeons, still
        will be defunct. So we tweak the above logic to
        expect the worker to send us a signal within 60 secs
        (in the form of closing its end of a pipe). The worker
        does this when it's done with the setup stage
        ready to enter the service loop (note it's the setup
        stage which is vulnerable to hangs -- the full
        blown worker blows up on EPIPE if the net goes down,
        due to the keep-alive thread)
        """

        self.set_state(self.ST_INIT, w)
        ret = 0
        def nwait(p, o=0):
            p2, r = waitpid(p, o)
            if not p2:
                return
            return r
        def exit_signalled(s):
            """ child teminated due to receipt of SIGUSR1 """
            return (os.WIFSIGNALED(s) and (os.WTERMSIG(s) == signal.SIGUSR1))
        def exit_status(s):
            if os.WIFEXITED(s):
                return os.WEXITSTATUS(s)
            return 1
        conn_timeout = int(gconf.connection_timeout)
        while ret in (0, 1):
            logging.info('-' * conn_timeout)
            logging.info('starting gsyncd worker')
            pr, pw = os.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]),
                                                 '--resource-remote', w[1]])
            self.lock.acquire()
            cpids.add(cpid)
            self.lock.release()
            os.close(pw)
            t0 = time.time()
            so = select((pr,), (), (), conn_timeout)[0]
            os.close(pr)
            if so:
                ret = nwait(cpid, os.WNOHANG)
                if ret != None:
                    logging.debug("worker died before establishing connection")
                else:
                    logging.debug("worker seems to be connected (?? racy check)")
                    while time.time() < t0 + conn_timeout:
                        ret = nwait(cpid, os.WNOHANG)
                        if ret != None:
                            logging.debug("worker died in startup phase")
                            break
                        time.sleep(1)
            else:
                logging.debug("worker not confirmed in %d sec, aborting it" % \
                              conn_timeout)
                self.terminate()
                time.sleep(1)
                os.kill(cpid, signal.SIGKILL)
                ret = nwait(cpid)
            if ret == None:
                self.set_state(self.ST_STABLE, w)
                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)
            time.sleep(10)
        self.set_state(self.ST_INCON, w)
        return ret

    def multiplex(self, wspx, suuid):
        def sigcont_handler(*a):
            """
            Re-init logging and send group kill signal
            """
            md = gconf.log_metadata
            logging.shutdown()
            lcls = logging.getLoggerClass()
            lcls.setup(label=md.get('saved_label'), **md)
            pid = os.getpid()
            os.kill(-pid, signal.SIGUSR1)
        signal.signal(signal.SIGUSR1, lambda *a: ())
        signal.signal(signal.SIGCONT, sigcont_handler)

        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))

        cpids = set()
        ta = []
        for wx in wspx:
            def wmon(w):
                cpid, _ = self.monitor(w, argv, cpids)
                terminate()
                time.sleep(1)
                self.lock.acquire()
                for cpid in cpids:
                    os.kill(cpid, signal.SIGKILL)
                self.lock.release()
                finalize(exval=1)
            t = Thread(target = wmon, args=[wx])
            t.start()
            ta.append(t)
        for t in ta:
            t.join()

def distribute(*resources):
    master, slave = resources
    mvol = Volinfo(master.volume, master.host)
    logging.debug('master bricks: ' + repr(mvol.bricks))
    locmbricks = [ b['dir'] for b in mvol.bricks if is_host_local(b['host']) ]
    prelude  = []
    si = slave
    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, si.host, prelude)
        sbricks = svol.bricks
        suuid = svol.uuid
    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

    # get the proper index of local bricks in master volume,
    # and setup the geo replication with appropriate slave node
    mbricks = [ b['dir'] for b in mvol.bricks ]
    mbricks.sort()
    locmbricks.sort()
    slaves.sort()
    workerspex = []

    locbidx = 0
    for idx, brick in enumerate(mbricks):
        if brick == locmbricks[locbidx]:
            workerspex.append((locmbricks[locbidx], slaves[idx % len(slaves)]))
            locbidx += 1
    logging.info('worker specs: ' + repr(workerspex))
    return workerspex, suuid

def monitor(*resources):
    """oh yeah, actually Monitor is used as singleton, too"""
    return Monitor().multiplex(*distribute(*resources))