[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bacula-devel] Bug 1083 again...maybe

On Tuesday 13 May 2008 16:18:12 Josh Fisher wrote:
> Kern,
> Below is a condensed version of the log from a sequence of jobs using
> bacula-dir and bacula-sd v. 2.2.10-b2. I believe that the reason job
> 7272 below failed has something to do with bug 1083 (mult-drive disk
> autochanger and volume swapping between drives).
> I purposefully reset the job start times to prevent simultaneous jobs
> from starting. I inadvertently missed one and jobs 7273 and 7274 ran in
> parallel. This turned out to be interesting, because when jobs 7273 and
> 7274 were simultaneously started the drives were  nearly in the same
> state as when job 7272 started. A usable volume was in drive 1 and an
> unusable volume (or no volume) was in drive 0. For the simultaneous jobs
> it worked. I think this might have been by chance, however, because it
> appears that job 7273 (using drive 0) did not have to swap from drive 1
> because job 7274 (using drive 1) had already unloaded drive 1 to slot 34???
> Also, both Dir and SD are running in the same Xen domU with only one
> virtual CPU assigned. Would job 7273 (the one using drive 0) had failed
> in the same way as job 7272 if there would have been another CPU
> available and the two jobs "really" started simultaneously?
> Anyway, I did not have debug level set in the SD, so I don't have much
> more for you at this time. I am now running SD with debug level 150 and
> will let you know what happens. Should I file another bug report if this
> can be reproduced?

To answer the above question.  If you can reproduce this on version 2.2.10-b3, 
yes, I think you should file a new bug report.  

However, your summary item that I have included below looks *exactly* like the 
problem I found.  Bacula "forgets" to unload element 34 that is in drive 1 
because of a bug (I stored the slot number in the wrong device).  This 
*should* be fixed in version 2.2.10-b3.  

If you know how to access the Branch-2.2 SVN, you can pull it directly from 
there.  The official release that I am currently uploading to Source Forge 
looks like it may be stalled :-(   Often it only works very early in the 
morning when neither Europe nor the US are using Source Forge.


> 12-May 23:51  drive 0 = slot-39 (pool=full) - drive 1 = slot-34 (pool=inc)
>   job 7272
>     using drive 0
>     finds slot-39 (pool=full) in drive 0
>     unloads drive 0 into slot 39
>     attempts to load slot-34 (pool=inc) into drive 0
>     fails - autochanger script returns (Storage Element 34 Empty (loaded in
> drive 1))

This SF.net email is sponsored by: Microsoft 
Defy all challenges. Microsoft(R) Visual Studio 2008. 
Bacula-devel mailing list

This mailing list archive is a service of Copilotco.