diff options
Diffstat (limited to 'extras/geo-rep/schedule_georep.py.in')
| -rw-r--r-- | extras/geo-rep/schedule_georep.py.in | 111 |
1 files changed, 65 insertions, 46 deletions
diff --git a/extras/geo-rep/schedule_georep.py.in b/extras/geo-rep/schedule_georep.py.in index c931111b365..48b2b507060 100644 --- a/extras/geo-rep/schedule_georep.py.in +++ b/extras/geo-rep/schedule_georep.py.in @@ -1,4 +1,4 @@ -#!/usr/bin/env python +#!/usr/bin/python3 """ Schedule Geo-replication ------------------------ @@ -43,7 +43,7 @@ SESSION_MOUNT_LOG_FILE = ("/var/log/glusterfs/geo-replication" "/schedule_georep.mount.log") USE_CLI_COLOR = True - +mnt_list = [] class GlusterBadXmlFormat(Exception): """ @@ -83,13 +83,15 @@ def execute(cmd, success_msg="", failure_msg="", exitcode=-1): On success it can print message in stdout if specified. On failure, exits after writing to stderr. """ - p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE) + p = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, universal_newlines=True) out, err = p.communicate() if p.returncode == 0: if success_msg: output_ok(success_msg) return out else: + if exitcode == 0: + return err_msg = err if err else out output_notok(failure_msg, err=err_msg, exitcode=exitcode) @@ -112,12 +114,12 @@ def cleanup(hostname, volname, mnt): """ Unmount the Volume and Remove the temporary directory """ - execute(["umount", mnt], + execute(["umount", "-l", mnt], failure_msg="Unable to Unmount Gluster Volume " "{0}:{1}(Mounted at {2})".format(hostname, volname, mnt)) execute(["rmdir", mnt], failure_msg="Unable to Remove temp directory " - "{0}".format(mnt)) + "{0}".format(mnt), exitcode=0) @contextmanager @@ -130,6 +132,7 @@ def glustermount(hostname, volname): Automatically unmounts it in case of Exceptions/out of context """ mnt = tempfile.mkdtemp(prefix="georepsetup_") + mnt_list.append(mnt) execute(["@SBIN_DIR@/glusterfs", "--volfile-server", hostname, "--volfile-id", volname, @@ -297,6 +300,7 @@ def get_summary(mastervol, slave_url): status_data = get(mastervol, slave_url) for session in status_data: + session_name = "" summary = { "active": 0, "passive": 0, @@ -339,7 +343,8 @@ def get_summary(mastervol, slave_url): if summary["faulty"] == 0 and summary["offline"] == 0: summary["ok"] = True - out.append([session_name, summary, faulty_rows, down_rows]) + if session_name != "": + out.append([session_name, summary, faulty_rows, down_rows]) return out @@ -347,7 +352,7 @@ def get_summary(mastervol, slave_url): def touch_mount_root(mastervol): # Create a Mount and Touch the Mount point root, # Hack to make sure some event available after - # setting Checkpoint. Without this their is a chance of + # setting Checkpoint. Without this there is a chance of # Checkpoint never completes. with glustermount("localhost", mastervol) as mnt: execute(["touch", mnt]) @@ -376,14 +381,14 @@ def main(args): output_ok("Started Geo-replication and watching Status for " "Checkpoint completion") - touch_mount_root(args.mastervol) - start_time = int(time.time()) duration = 0 # Sleep till Geo-rep initializes time.sleep(60) + touch_mount_root(args.mastervol) + slave_url = "{0}::{1}".format(args.slave, args.slavevol) # Loop to Check the Geo-replication Status and Checkpoint @@ -397,41 +402,39 @@ def main(args): # or any other error. Gluster cmd still produces XML output # with different message output_warning("Unable to get Geo-replication Status") - time.sleep(1) - continue - - session_name, summary, faulty_rows, down_rows = session_summary[0] - chkpt_status = "COMPLETE" if summary["checkpoints_ok"] else \ - "NOT COMPLETE" - ok_status = "OK" if summary["ok"] else "NOT OK" - - if summary["ok"]: - output_ok("All Checkpoints {1}, " - "All status {2} (Turns {0:>3})".format( - turns, chkpt_status, ok_status)) - else: - output_warning("All Checkpoints {1}, " - "All status {2} (Turns {0:>3})".format( - turns, chkpt_status, ok_status)) - - output_warning("Geo-rep workers Faulty/Offline, " - "Faulty: {0} Offline: {1}".format( - repr(faulty_rows), - repr(down_rows))) - - if summary["checkpoints_ok"]: - output_ok("Stopping Geo-replication session now") - cmd = ["@SBIN_DIR@/gluster", "volume", "geo-replication", - args.mastervol, - "%s::%s" % (args.slave, args.slavevol), "stop"] - execute(cmd) - break else: - # If Checkpoint is not complete after a iteration means brick - # was down and came online now. SETATTR on mount is not - # recorded, So again issue touch on mount root So that - # Stime will increase and Checkpoint will complete. - touch_mount_root(args.mastervol) + session_name, summary, faulty_rows, down_rows = session_summary[0] + chkpt_status = "COMPLETE" if summary["checkpoints_ok"] else \ + "NOT COMPLETE" + ok_status = "OK" if summary["ok"] else "NOT OK" + + if summary["ok"]: + output_ok("All Checkpoints {1}, " + "All status {2} (Turns {0:>3})".format( + turns, chkpt_status, ok_status)) + else: + output_warning("All Checkpoints {1}, " + "All status {2} (Turns {0:>3})".format( + turns, chkpt_status, ok_status)) + + output_warning("Geo-rep workers Faulty/Offline, " + "Faulty: {0} Offline: {1}".format( + repr(faulty_rows), + repr(down_rows))) + + if summary["checkpoints_ok"]: + output_ok("Stopping Geo-replication session now") + cmd = ["@SBIN_DIR@/gluster", "volume", "geo-replication", + args.mastervol, + "%s::%s" % (args.slave, args.slavevol), "stop"] + execute(cmd) + break + else: + # If Checkpoint is not complete after a iteration means brick + # was down and came online now. SETATTR on mount is not + # recorded, So again issue touch on mount root So that + # Stime will increase and Checkpoint will complete. + touch_mount_root(args.mastervol) # Increment the turns and Sleep for 10 sec turns += 1 @@ -446,13 +449,18 @@ def main(args): time.sleep(args.interval) + for mnt in mnt_list: + execute(["rmdir", mnt], + failure_msg="Unable to Remove temp directory " + "{0}".format(mnt), exitcode=0) + if __name__ == "__main__": parser = ArgumentParser(formatter_class=RawDescriptionHelpFormatter, description=__doc__) parser.add_argument("mastervol", help="Master Volume Name") parser.add_argument("slave", - help="SLAVEHOST or root@SLAVEHOST " - "or user@SLAVEHOST", + help="Slave hostname " + "(<username>@SLAVEHOST or SLAVEHOST)", metavar="SLAVE") parser.add_argument("slavevol", help="Slave Volume Name") parser.add_argument("--interval", help="Interval in Seconds. " @@ -462,12 +470,23 @@ if __name__ == "__main__": "stop Geo-replication if Checkpoint is not complete " "in the specified timeout time", type=int, default=0) - parser.add_argument("--no-color", help="Use Color in CLI output", + parser.add_argument("--no-color", help="Don't use Color in CLI output", action="store_true") args = parser.parse_args() if args.no_color: USE_CLI_COLOR = False try: + # Check for session existence + cmd = ["@SBIN_DIR@/gluster", "volume", "geo-replication", + args.mastervol, "%s::%s" % (args.slave, args.slavevol), "status"] + execute(cmd) main(args) except KeyboardInterrupt: + for mnt in mnt_list: + execute(["umount", "-l", mnt], + failure_msg="Unable to Unmount Gluster Volume " + "Mounted at {0}".format(mnt), exitcode=0) + execute(["rmdir", mnt], + failure_msg="Unable to Remove temp directory " + "{0}".format(mnt), exitcode=0) output_notok("Exiting...") |