[Yocto-builds] buildbot failure in The Yocto Autobuilder on nightly-oecore

[Richard Purdie]

On Fri, 2013-03-22 at 10:57 +0000, Paul Eggleton wrote:
> On Friday 22 March 2013 01:32:59 yocto-builds at yoctoproject.org wrote:
> > Build status: FAILURE
> > 
> > Buildslave for this Build: ab06
> > | NOTE: Test Result for qemuarm core-image-sato
> > | NOTE: Testcase PASS FAIL NORESULT
> > | NOTE: SSH 1 0 0
> > | NOTE: SCP 1 0 0
> > | NOTE: dmesg 1 0 0
> > | NOTE: smart_help 1 0 0
> > | NOTE: smart_query 0 1 0
> > | NOTE: rpm_query 1 0 0
> > | NOTE: connman 1 0 0
> > | NOTE: shutdown 1 0 0
> 
> And from the log:
> >| 	TARGET: smart command run without problem
> >| 	TARGET: smart command has issue when running, Pls. check the error log
> >| 	TARGET: ##### Error Log #####
> >| rpmdb: BDB0113 Thread/process 663/3069857792 failed: BDB1507 Thread died in 
> >Berkeley DB library
> >| error: db_init:db3.c:1098: dbenv->failchk(-30973): BDB0087 DB_RUNRECOVERY: 
> >Fatal error, run database recovery
> >| Re-opening dbenv with DB_RECOVER ...
> >| BDB2526 Finding last valid log LSN: file: 1 offset 8322133
> >| 
> >recovery 0% complete
> >...
> >recovery 18% completeBDB1518 Recovery complete at Fri Mar 22 01:11:21 2013
> >| BDB1519 Maximum transaction ID 0 recovery checkpoint [1][8328113]
> >| .
> >| recovery succeeded.
> >| Updating cache...               ######################################## 
> >[100%]
> >| 
> >| avahi-daemon-0.6.31-r6.1 at armv5te
> >| avahi-locale-en-gb-0.6.31-r6.1 at armv5te
> >| 
> >| Saving cache...
> >| 
> >| 	TARGET: #####    End    #####
> 
> From this we can determine:
> 
> * The problem is no longer confined to MIPS
> * The RPM database is definitely becoming corrupted
> 
> Even in the face of multiple clients opening the RPM database within the same 
> machine, locking should prevent simultaneous access of the database from 
> resulting in corruption. Is there a chance two qemu instances are being 
> started on the same image file, thus bypassing any internal locking?

That is unlikely as the AB takes a copy of the image file for each qemu
instance and we only run one sanity test per image/machine combo. I'm
open to evidence which says otherwise but having looked at the code
paths recently for other reasons, I can't see it...

Cheers,

Richard