Hi Ana, >Did you monitor resource usage during the backups? The list of files generated for accurate backups are >kept in memory (by both director and client), so this should cause resource use (CPU, memory, etc.) to increase in both hosts.
I did monitor these items earlier, but did not notice a huge drop in usage in these items. However, I was using Zabbix to monitor which takes a reading every so many minutes. It's possible it didn't take a reading when the problem occurred. I should've used something like top instead or VMware's performance tools. I'll take a look at the graphs again...I may have missed it. However, as usual your suggestions helped! I increased the amount of memory on Director and added more CPUs. I also reduced the number of concurrent jobs from 50 to 25. So far, two days of incremental backups have not produced any deadlock errors. All my backups finished successfully. Thank you so much for the help! (again) :-) Your advice is always, always so helpful. -craig On Sat, Aug 8, 2015 at 4:11 PM, Ana Emília M. Arruda <emiliaarr...@gmail.com > wrote: > Hi Craig, > > On Fri, Aug 7, 2015 at 3:49 PM, Craig Shiroma <shiroma.crai...@gmail.com> > wrote: > >> Hi Ana, >> >> Thank you for the suggestion! >> > > You're welcome! > > >> >> I'll look into adding more CPU and memory to director, although I didn't >> see much of an impact on either between a non-accurate run and an accurate >> run. For example, there was large depletion of available memory, no >> swapping, or high load. >> > > Did you monitor resource usage during the backups? The list of files > generated for accurate backups are kept in memory (by both director and > client), so this should cause resource use (CPU, memory, etc.) to increase > in both hosts. > > >> >> I did add more memory to the catalog server and turned Accurate back on >> for the same hosts. I had no deadlocks. Last night was mostly Fulls, >> though. Not sure if that makes a difference. Would you know if Bacula >> would uses less resources when fulls are run because it is going to back up >> everything anyway and no comparison of files needs to be made (I'm >> guessing)? When a full is done, does Bacula still need to keep a list of >> the files in memory for hosts using Accurate backups? My first thought is >> no. >> > > Mine too. When using accurate backups, the amount of resources used > should be noticed when running incremental, differential and full+basejobs > backups. > > >> >> Thanks again for the help! Your posts are always so helpful. >> > > Thank you too! > Best regards, > Ana > > >> >> -craig >> >> >> On Thu, Aug 6, 2015 at 3:37 PM, Ana Emília M. Arruda < >> emiliaarr...@gmail.com> wrote: >> >>> Hello Craig, >>> >>> In one of your posts you mentioned Segmentation violation in the >>> director host. Accurate backups requires more resources than normal ones. >>> Have you checked if CPU and memory resources are enough in director and the >>> clients that are configured for using accurate mode? >>> >>> Best regards, >>> Ana >>> >>> On Thu, Aug 6, 2015 at 5:36 PM, Craig Shiroma <shiroma.crai...@gmail.com >>> > wrote: >>> >>>> Thanks Kern! I'll bring in a DBA on our side to have a look. >>>> >>>> Would you have any thoughts on this question posed earlier? >>>> >>>> 3. Why is Bacula spinning off a new job right away after it detects the >>>> deadlock for each affected job instead of waiting until the rescheduled job >>>> runs? I verified that there were no duplicate jobs in the queue before the >>>> backups started running, no jobs were running before the start of the >>>> backups, and I did not start any of these backups manually to cause a >>>> second job to appear. >>>> >>>> This happened on both nights I ran with Accurate turned On on the hosts >>>> that had failed backups because of the deadlock. >>>> >>>> Regards, >>>> -craig >>>> >>>> On Thu, Aug 6, 2015 at 9:48 AM, Kern Sibbald <k...@sibbald.com> wrote: >>>> >>>>> On 06.08.2015 21:44, Craig Shiroma wrote: >>>>> >>>>> Hi Kern, >>>>> >>>>> Thank you for the info! We're using MySQL 5.6 Percona Server, Release >>>>> 68.0, Revision 656. >>>>> >>>>> Would this setting cause the problem? >>>>> innodb_lock_wait_timeout = 100 >>>>> >>>>> Is it too high or too low or has no bearing on the problem? >>>>> >>>>> >>>>> Sorry, I am a Bacula programmer, and I do not know much about >>>>> databases -- especially MySQL since I use PostgreSQL. PostgreSQL is >>>>> harder >>>>> to install and a bit harder to configure than MySQL, but it performs much >>>>> better. >>>>> >>>>> >>>>> >>>>> Thanks again, >>>>> -craig >>>>> >>>>> >>>>> On Thu, Aug 6, 2015 at 9:26 AM, Kern Sibbald <k...@sibbald.com> wrote: >>>>> >>>>>> On 06.08.2015 18:46, Bryn Hughes wrote: >>>>>> >>>>>> I think what Kern is getting at is that your database is what threw >>>>>> the error, not Bacula. Whatever DB you are using is what is having the >>>>>> issue. >>>>>> >>>>>> >>>>>> Yes. That is exactly what I was implying. >>>>>> >>>>>> The rest of this is directed to Craig: >>>>>> If you are using MariaDB (I have no indication that you are), please >>>>>> be aware that it may be a very good database, maybe even better than >>>>>> MySQL, >>>>>> but Bacula is built and tested against MySQL, and if you use binaries >>>>>> that >>>>>> were built for MySQL, you could run into problems by using MariaDB. Even >>>>>> if your binaries were explicitly built with MariaDB, it may not be >>>>>> compatible with the way Bacula works. Bacula has a tendency to push >>>>>> databases to the extreme, and it works well with MySQL and PostgreSQL, >>>>>> but >>>>>> possibly not with other databases. I bring up MariaDB because it has >>>>>> been >>>>>> mentioned in another posting to this list. >>>>>> >>>>>> I would be very surprised if your problem has anything to do with >>>>>> Accurate -- the database routines know nothing about accurate and none of >>>>>> the data is different. It is more likely due to the VM environment or to >>>>>> some build or version problem with MySQL (or MariaDB). >>>>>> >>>>>> Best regards, >>>>>> Kern >>>>>> >>>>>> >>>>>> Bryn >>>>>> >>>>>> On 2015-08-06 09:11 AM, Craig Shiroma wrote: >>>>>> >>>>>> Hi Kern, >>>>>> >>>>>> Thank you very much for the reply! Would you have any suggestions on >>>>>> what may be causing this problem or how I can debug it? Obviously, I'm >>>>>> encountering deadlocks when accurate backup runs on some of our hosts and >>>>>> we want to use accurate backup on all of our hosts if possible. >>>>>> >>>>>> Warmest regards, >>>>>> -craig >>>>>> >>>>>> On Thu, Aug 6, 2015 at 12:11 AM, Kern Sibbald <k...@sibbald.com> >>>>>> wrote: >>>>>> >>>>>>> On 06.08.2015 10:15, Craig Shiroma wrote: >>>>>>> >>>>>>> Hello again, >>>>>>> >>>>>>> I just thought I'd update this post with more information in hopes >>>>>>> of getting some explanation for the deadlocks. >>>>>>> >>>>>>> I ran with Accurate backup on our test VMs (RHEL) for a couple of >>>>>>> days and got the same errors on some VMs that were running accurate and >>>>>>> some that were not. These hosts were running concurrently. I would say >>>>>>> 90% of the hosts that were configured to use Accurate finished >>>>>>> successfully. However, there were a few that failed with the deadlock >>>>>>> error -- some that were configured to use accurate and some that were >>>>>>> not >>>>>>> configured to use accurate. Also, on all of these, a second job started >>>>>>> for each of the affected hosts right after Bacula detected the deadlock >>>>>>> even though it said a reschedule would happen 3600 seconds later (the >>>>>>> 3600 >>>>>>> seconds is correct). >>>>>>> >>>>>>> Tonight, I disabled accurate on all hosts and the deadlocks did not >>>>>>> happen. No errors were detected and all the backups finished >>>>>>> successfully. >>>>>>> >>>>>>> Some questions... >>>>>>> 1. Can I back up multiple hosts concurrently with some hosts >>>>>>> configured to use accurate and some configured not to use accurate? >>>>>>> Or, is >>>>>>> it an all or none thing, meaning all hosts that run concurrently must >>>>>>> either be using accurate backup or not using accurate backup (cannot mix >>>>>>> the two)? >>>>>>> >>>>>>> 2. It seems like the hosts that get out of the starting gate first >>>>>>> are the ones affected. I am configured to run 50 jobs concurrently. >>>>>>> Again, no problems with accurate turned off on all hosts for months now. >>>>>>> >>>>>>> 3. Why is Bacula spinning off a new job right away after it detects >>>>>>> the deadlock for each affected job instead of waiting until the >>>>>>> rescheduled >>>>>>> job runs? I verified that there were no duplicate jobs in the queue >>>>>>> before >>>>>>> the backups started running, no jobs were running before the start of >>>>>>> the >>>>>>> backups, and I did not start any of these backups manually to cause a >>>>>>> second job to appear. >>>>>>> >>>>>>> >>>>>>> Bacula is not aware of any SQL internal deadlocks. >>>>>>> >>>>>>> >>>>>>> From the INNODB Monitor output: >>>>>>> >>>>>>> TRANSACTION: >>>>>>> TRANSACTION 208788977, ACTIVE 1 sec setting auto-inc lock >>>>>>> mysql tables in use 4, locked 4 >>>>>>> 9 lock struct(s), heap size 1184, 5 row lock(s) >>>>>>> MySQL thread id 50808, OS thread handle 0x7f8f2c3b4700, query id >>>>>>> 29558637 <host> 192.168.10.99 bacula Sending data >>>>>>> INSERT INTO File (FileIndex, JobId, PathId, FilenameId, LStat, MD5, >>>>>>> DeltaSeq) SELECT batch.FileIndex, batch.JobId, Path.PathId, >>>>>>> Filename.FilenameId,batch.LStat, batch.MD5, batch.DeltaSeq FROM batch >>>>>>> JOIN >>>>>>> Path ON (batch.Path = Path.Path) JOIN Filename ON (batch.Name = >>>>>>> Filename.Name) >>>>>>> WAITING FOR THIS LOCK TO BE GRANTED: >>>>>>> TABLE LOCK table `bacula`.`File` trx id 208788977 lock mode AUTO-INC >>>>>>> waiting >>>>>>> WE ROLL BACK TRANSACTION (2) >>>>>>> >>>>>>> I am running Bacula 7.0.5 on RHEL 6.6 x64 with Director, Storage and >>>>>>> Catalog running on separate RHEL 6.6 hosts. Our clients are RHEL 6's, >>>>>>> 5's >>>>>>> and Windows Servers 2008 and 2012R2. >>>>>>> >>>>>>> Any help would be much appreciated. >>>>>>> >>>>>>> Warmest regards, >>>>>>> -craig >>>>>>> >>>>>>> On Tue, Aug 4, 2015 at 1:56 PM, Craig Shiroma < >>>>>>> shiroma.crai...@gmail.com> wrote: >>>>>>> >>>>>>>> BTW, I suppose there could've been two jobs for the host(s) in >>>>>>>> scheduling queue. If this was the case, is there a way to find out >>>>>>>> after >>>>>>>> the fact? If this did actually happen, what could cause duplicate >>>>>>>> jobs to >>>>>>>> be scheduled on the same day at the same time? I know no one manually >>>>>>>> ran >>>>>>>> the jobs in question. Again, this only was a problem for a few of the >>>>>>>> jobs >>>>>>>> that ran last night, not all of them and some to do accurate backup and >>>>>>>> some not. >>>>>>>> >>>>>>>> Regards, >>>>>>>> -craig >>>>>>>> >>>>>>>> On Tue, Aug 4, 2015 at 9:27 AM, Craig Shiroma < >>>>>>>> shiroma.crai...@gmail.com> wrote: >>>>>>>> >>>>>>>>> Hello, >>>>>>>>> >>>>>>>>> I had a few backups fail last night with the following error: >>>>>>>>> >>>>>>>>> 2015-08-03 18:02:46bacula-dir JobId 123984: b INTO File >>>>>>>>> (FileIndex, JobId, PathId, FilenameId, LStat, MD5, DeltaSeq) SELECT >>>>>>>>> batch.FileIndex, batch.JobId, Path.PathId, >>>>>>>>> Filename.FilenameId,batch.LStat, >>>>>>>>> batch.MD5, batch.DeltaSeq FROM batch JOIN Path ON (batch.Path = >>>>>>>>> Path.Path) >>>>>>>>> JOIN Filename ON (batch.Name = Filename.Name): ERR=Deadlock found when >>>>>>>>> trying to get lock; try restarting transaction >>>>>>>>> >>>>>>>>> The only thing I did yesterday was switch a bunch of backups to >>>>>>>>> use Accurate backup and restart bacula-dir and bacula-sd after that. >>>>>>>>> However, the above problem also occurred on some hosts that was not >>>>>>>>> set to >>>>>>>>> use Accurate backup. From the log, it seems like two jobs for this >>>>>>>>> host >>>>>>>>> was scheduled to run at 18:00 because the second job started and >>>>>>>>> found a >>>>>>>>> duplicate job (job 123984) and canceled the backup. I know there >>>>>>>>> were no >>>>>>>>> jobs running before 18:00 so 123984 was not an old job still running. >>>>>>>>> Same >>>>>>>>> with the other jobs that were canceled because of the above situation. >>>>>>>>> >>>>>>>>> Anyway, does anyone have an idea what would cause this, especially >>>>>>>>> how the second job got shot into the system. After the deadlock >>>>>>>>> error, >>>>>>>>> Bacula said it would reschedule the job. However the second job >>>>>>>>> started >>>>>>>>> right after the deadlock error instead of one hour later which makes >>>>>>>>> me >>>>>>>>> think that there were two jobs for this host scheduled to run at >>>>>>>>> 18:00. >>>>>>>>> >>>>>>>>> Thank you in advance, >>>>>>>>> -craig >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>> >>>>>>> >>>>>>> ------------------------------------------------------------------------------ >>>>>>> >>>>>>> >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Bacula-users mailing >>>>>>> listBacula-users@lists.sourceforge.nethttps://lists.sourceforge.net/lists/listinfo/bacula-users >>>>>>> >>>>>>> >>>>>>> >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> >>>>>> >>>>>> >>>>>> _______________________________________________ >>>>>> Bacula-users mailing >>>>>> listBacula-users@lists.sourceforge.nethttps://lists.sourceforge.net/lists/listinfo/bacula-users >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> >>>>>> >>>>>> >>>>>> _______________________________________________ >>>>>> Bacula-users mailing >>>>>> listBacula-users@lists.sourceforge.nethttps://lists.sourceforge.net/lists/listinfo/bacula-users >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> >>>>>> _______________________________________________ >>>>>> Bacula-users mailing list >>>>>> Bacula-users@lists.sourceforge.net >>>>>> https://lists.sourceforge.net/lists/listinfo/bacula-users >>>>>> >>>>>> >>>>> >>>>> >>>> >>>> >>>> ------------------------------------------------------------------------------ >>>> >>>> _______________________________________________ >>>> Bacula-users mailing list >>>> Bacula-users@lists.sourceforge.net >>>> https://lists.sourceforge.net/lists/listinfo/bacula-users >>>> >>>> >>> >> >
------------------------------------------------------------------------------
_______________________________________________ Bacula-users mailing list Bacula-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/bacula-users
