>>>>> On Mon, 22 Jun 2020 10:24:13 -0400, Phil Stracchino said: > > Status: > - Director and its FD running 9.6.5, everything else running 9.6.3 > except for the one 9.4.4 Fedora client > - Director now talking directly to its local database node instead of > talking to the cluster via HAproxy (SHOULD be safe most of the time, the > machine obviously won't get rebooted while backup jobs are running) > > No hung jobs overnight. I'll keep it this way for a few more days to > see, but could 9.6.5, or the combination of 9.6.5 plus this > libmysqlclient version, be not getting along well with HAproxy...?
That theory fits the evidience, but there is no direct evidience that it is caused by a bug in 9.6.5. E.g. since it seems to be somewhat random, it could also just be chance due to memory layout or timing. I assume you only have one libmysqlclient library, so does ldd print the same thing for both versions of the Director? Also, the crashes are not necessarily the same problem as the hanging. It might turn out that they are both related to the DB, but the backtraces from all the daemons are needed to show that (before you cancel any jobs). __Martin _______________________________________________ Bacula-devel mailing list Bacula-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-devel
