From: <bac...@li...> - 2004-12-24 10:27:17
|
The following bug has been CLOSED ====================================================================== http://bugs.bacula.org/bug_view_advanced_page.php?bug_id=0000209 ====================================================================== Reported By: abuehler Assigned To: ====================================================================== Project: bacula Bug ID: 209 Category: File Daemon Reproducibility: always Severity: crash Priority: normal Status: closed ====================================================================== Date Submitted: 12-21-2004 09:53 PST Last Modified: 12-24-2004 02:32 PST ====================================================================== Summary: FD crash after sucessfull backup Description: After sucessfull completion of the backup the FD crash. See attached traceback. In output of the "status client=albundy-fd" the job is not listed, but appears in "list jobs". ====================================================================== ---------------------------------------------------------------------- kern - 12-21-2004 10:10 PST ---------------------------------------------------------------------- The traceback doesn't seem to show the FD crashing. I recommend that you manually run it under the debugger. Then when it crashes, enter the following command: thread apply all bt Hopefully that will show where the problem lies. I suspect that the status client is showing nothing because either the job never ran or the FD crashed. If it crashes, any running jobs are lost and not reported if the FD is restarted (i.e. they are only written to the job history at the end of the job). ---------------------------------------------------------------------- abuehler - 12-22-2004 07:22 PST ---------------------------------------------------------------------- I've checked out the cvs 1.37.1 (21Dec04) and ran it with gdb. It seems the FD is getting SIGUSR2. After entering "continue" (>100 times) in gdb the FD terminates with SIGSEGV (gdbout.txt). I got a email notifcation with a failed backup. Next I've tried runing FD in foreground, set the debug level to 150 and redirect stdout and stderr to a file (debug-150.txt, omitted serveral lines). This time I got email notification with a "Backup OK" (debug-email.txt) and a backtrace mail (debug-trace.txt). ---------------------------------------------------------------------- kern - 12-22-2004 09:25 PST ---------------------------------------------------------------------- Thanks for your efforts. Please try running the FD under the debugger again as follows: gdb bacula-fd run -s -f -c bacula-fd.conf when it comes back to the command prompt (after crashing), enter: thread apply all bt Hopefully that will show me where the problem is. From the -dxxx trace, it seems like it is dying in some utility thread because after the seg fault, sending of the data to the SD continues ... ---------------------------------------------------------------------- abuehler - 12-22-2004 11:44 PST ---------------------------------------------------------------------- The attached file "gdbout2.txt" contains the output of the gdb session. ---------------------------------------------------------------------- kern - 12-22-2004 12:58 PST ---------------------------------------------------------------------- The traceback does not clearly show where the seg fault occurs, but from previous similar reports, I can now make a guess (estimate 99% correct). You are probably either mixing old and new syntax FileSets, and for some reason, when you have old style includes/excludes, something gets confused and seg faults. Try switching to using all new style FileSet, and I am 99% sure your problem will go away. Please report back. ---------------------------------------------------------------------- kern - 12-24-2004 02:32 PST ---------------------------------------------------------------------- I believe that I have now found and fixed the FD crash problem (even if you use the old style include/excludes). It is now in the 1.37.1 CVS. I would appreciate it if you would try it -- assuming that you have not already converted your FileSet. I've also attached a 1.36.1 patch for users on 1.36.1. Note, this patch can also be applied to 1.37.1 Bug History Date Modified Username Field Change ====================================================================== 12-21-04 09:53 abuehler New Bug 12-21-04 09:53 abuehler File Added: traceback.txt 12-21-04 10:10 kern Bugnote Added: 0000579 12-22-04 07:22 abuehler Bugnote Added: 0000581 12-22-04 07:23 abuehler File Added: gdbout.txt 12-22-04 07:23 abuehler File Added: debug-150.txt 12-22-04 07:24 abuehler File Added: debug-email.txt 12-22-04 07:24 abuehler File Added: debug-trace.txt 12-22-04 09:25 kern Bugnote Added: 0000583 12-22-04 09:25 kern Status new => feedback 12-22-04 11:40 abuehler File Added: gdbout2.txt 12-22-04 11:44 abuehler Bugnote Added: 0000586 12-22-04 12:58 kern Bugnote Added: 0000587 12-24-04 02:28 kern File Added: 1.36.1-fileset.patch 12-24-04 02:32 kern Bugnote Added: 0000588 12-24-04 02:32 kern Resolution open => fixed 12-24-04 02:32 kern Status feedback => closed ====================================================================== |