ovdb instability with Berkeley DB5/6

boogles at somtow.mit.edu boogles at somtow.mit.edu
Thu Sep 12 21:56:04 UTC 2019


I have an INN environment that I updated recently.  I got it stable, but
I figured I would post my experience in case there's something I can do
to help.  Otherwise, maybe my experience will help anyone else having
this issue.

Env: FreeBSD 12 - amd64, INN 2.6.3

The trouble came when using 'makehistory -O -x -F' to build my ovdb.  I
started with the precompiled Berkeley DB 5.3 from the FreeBSD ports
system.  I couldn't get it to be stable.  At first, I had mmap on.  That
was very unstable.  It would die and corrupt the DB.  It wouldn't
recover.  Then I turned on shared memory instead.  Things were slightly
more stable, but it would eventually hang.  When I did an 'ovdb_stat
-t', I would see there was a single transaction running.  I tried lots
of other parameters randomly.  I figure I have pretty good coverage of
them at this point.

My experience was about the same with BDB 6.2.

I finally switched back to BDB 4.8, which is EoL.  With that and shared
memory, it was stable enough to build ovdb (about 20 hours).  Everything
looks good so far.

I was upgrading from 2.5.3 and I can't remember what version of BDB I
was using way back then.  It was probably another 4.x version.  That was
rock solid for years.  Hopefully I've gotten it back into a good state.

If this happens again, do folks have thoughts on how to debug it?

-boogles


More information about the inn-workers mailing list