Menu
▾
▴
SF.net SVN: safekeep:[775] safekeep/trunk/safekeep
|
From: <fcr...@us...> - 2011-11-28 12:38:43
|
Revision: 775
http://safekeep.svn.sourceforge.net/safekeep/?rev=775&view=rev
Author: fcrawford
Date: 2011-11-28 12:38:37 +0000 (Mon, 28 Nov 2011)
Log Message:
-----------
- includes the PID of the creating safekeep in the snapshot names,
- handles the client side cleanup, if snapshots are left around,
- cleans up the backup files if the client fails,
- intentionally block run a "safekeep --client --cleanup" when other safekeep is running.
- clean up of messages and some control flow,
- suppress tracebacks, only printing them on debug requests,
- ensure that the client is closed before starting some possibly long running server processes.
Modified Paths:
--------------
safekeep/trunk/safekeep
Modified: safekeep/trunk/safekeep
===================================================================
--- safekeep/trunk/safekeep 2011-11-28 12:38:02 UTC (rev 774)
+++ safekeep/trunk/safekeep 2011-11-28 12:38:37 UTC (rev 775)
@@ -64,6 +64,7 @@
backup_user = None
home_dir = None
base_dir = None
+current_pid = os.getpid()
default_bandwidth = {}
cmd = "<Missing>"
@@ -140,7 +141,7 @@
def error(msg, ex=None):
extra = ""
- if ex:
+ if ex and verbosity_level > 2:
extra = stacktrace()
log(msg + extra, 'ERR')
@@ -595,7 +596,7 @@
continue
try:
os.remove(dump['file'])
- except Exception, e:
+ except OSError, e:
warn('Unable to remove dump file: %s for database %s because: %s' %
(dump['file'], dump['db'], e))
@@ -738,20 +739,20 @@
debug('Checking FS snapshots')
for snap in cfg['snaps']:
device = snap['device']
- if check_lvm_information(device):
- raise Exception("Previous snapshots found for %s: run 'safekeep --server --cleanup' to correct" % device)
+ if check_lvm_information(device) and not do_client_scrub():
+ raise Exception("Previous snapshots found for %s and automatic correction failed: run 'safekeep --server --cleanup' to correct" % device)
ret = spawn(['modprobe', 'dm-snapshot'])
if ret:
warn('modprobe dm-snapshot failed, continuing')
- bdir = tempfile.mkdtemp("-rbind", "safekeep-", "/mnt")
+ bdir = tempfile.mkdtemp("-rbind", "safekeep-%d-" % current_pid, "/mnt")
ret = spawn(['mount', '--rbind', '/', bdir])
if ret:
warn('mount --rbind failed, snapshotting will be disabled')
try:
os.rmdir(bdir)
- except Exception, e:
- warn('Failed to remove: %s' % bdir)
+ except OSError, e:
+ warn('Failed to remove: %s: %s' % (bdir, e))
bdir = '/'
else:
do_client_snap(cfg, bdir)
@@ -772,8 +773,8 @@
else:
try:
os.rmdir(bdir)
- except Exception, e:
- warn('Unable to remove: ' + bdir)
+ except OSError, e:
+ warn('Unable to remove: %s: %s' % (bdir, e))
do_client_dbdump_teardown(cfg)
@@ -792,6 +793,22 @@
else:
scrubbed = False
+ # Go through and see if any come from existing safekeep processes
+ pattern = re.compile(r"_snap_safekeep-(\d+)-")
+ lvm_snap_list = lvm_snap_information()
+ for (volume, group) in lvm_snap_list:
+ matches = pattern.search(volume)
+ if matches is not None:
+ pid = matches.group(1)
+ # Look up /proc/<pid>/cmdline to see what process is running
+ proc_file = "/proc/" + pid + "/cmdline"
+ if pid != current_pid and os.path.exists(proc_file):
+ fin = open(proc_file, "r")
+ (cmd, arg0, args) = fin.read().split('\0', 2)
+ fin.close()
+ if os.path.basename(arg0) == "safekeep":
+ raise Exception('another safekeep process running: pid %s' % pid)
+
if os.environ['PATH'][-1] == ':':
os.environ['PATH'] += '/sbin:/usr/sbin:/usr/local/sbin:'
else:
@@ -811,8 +828,8 @@
else:
try:
os.rmdir(mountpoint)
- except Exception, e:
- warn('Failed to remove: %s' % mountpoint)
+ except OSError, e:
+ warn('Failed to remove: %s: %s' % (mountpoint, e))
else:
ret = spawn(['umount', mountpoint])
if ret:
@@ -827,7 +844,7 @@
# Now cleanup any snapshots still hanging around
debug("Cleaning up remaining snapshots")
- for (volume, group) in lvm_snap_information():
+ for (volume, group) in lvm_snap_list:
device = os.path.join('/dev', group, volume)
info("Removing snapshot %s" % device)
ret = do_lvremove(device)
@@ -845,12 +862,19 @@
info("Removing rbind directory %s" % mountpoint)
try:
os.rmdir(mountpoint)
- except Exception, e:
- warn('Failed to remove: %s' % mountpoint)
+ except OSError, e:
+ warn('Failed to remove: %s: %s' % (mountpoint, e))
if not scrubbed:
info('No cleanup required')
+ # This has to be rerun to see if it has been successful
+ if lvm_snap_information():
+ return False
+ return True
+
+ return False
+
def do_client():
debug("Do client main loop")
should_cleanup = True
@@ -914,7 +938,7 @@
raise ClientException(line[5:].strip())
elif line.startswith('TRACEBACK'):
i = line.find('>>>')
- raise ClientException(line[10:i].strip(), line[i+3:].replace('###', '\n'))
+ raise ClientException(line[10:i].strip(), line[i+3:].replace('###', '\n').rstrip())
elif not line:
raise Exception('client died unexpectedly')
else:
@@ -1064,7 +1088,7 @@
try:
os.makedirs(datadir)
except EnvironmentError, ex:
- raise Exception('Can not create data store dir: %s' % datadir)
+ raise Exception('Can not create data store dir: %s: %s' % (datadir, ex))
rdiff_logdir = os.path.join(datadir, 'rdiff-backup-data')
if cfg['retention'] and os.path.isdir(rdiff_logdir) and not cleanup:
@@ -1098,8 +1122,6 @@
cin.flush()
do_server_getanswer(cout)
bdir = '/' # Fake directory for the rest of the cleanup
- do_server_rdiff_cleanup(cfg)
- cleaned_up = True
errs = 0
else:
cin.write('SETUP\n')
@@ -1121,6 +1143,7 @@
backup_marker = None
do_server_rdiff(cfg, bdir, nice, ionice, force)
+ cleaned_up = True
errs = 0
if os.path.isdir(rdiff_logdir):
@@ -1142,16 +1165,23 @@
info('Server backup for client %s: OK (%d WARNINGS)' % (id, errs))
except Exception, ex:
- if cleanup and not cleaned_up:
+ if cleanup:
info('Client-side cleanup for client %s: FAILED' % id)
- do_server_rdiff_cleanup(cfg)
else:
if isinstance(ex, ClientException):
error('Client %s: FAILED due to: %s' % (id, ex or ''))
- if ex.traceback: error(ex.traceback)
+ if ex.traceback and verbosity_level > 2: error(ex.traceback)
else:
error('Server backup for client %s: FAILED' % id, ex)
+ # Shutdown client
+ cout.close()
+ cin.close()
+
+ if not cleaned_up:
+ do_server_rdiff_cleanup(cfg)
+ cleaned_up = True
+
if output_done:
info('------------------------------------------------------------------')
debug('Server backup done')
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
