[Bucardo-general] KID dies with 'could not serialize access due to concurrent update' (prevents replication)

[Greg Sabino Mullane]

On Fri, Dec 09, 2011 at 12:14:21AM -0500, Bill McGonigle wrote:
> But I have no explanation for why rebooting the vm would make those
> concurrency problems stop happening.  I'm identifying the 'concurrent
> update' errors as not coincidental in this particular case since they
> coincided exactly with replication failure in the logs and stopped
> appearing in the postgresql log files (error statement logging on) after
> the reboot.

I have no explanation either. The only thing that may factor in is a 
genuine Postgres bug, or the fact that the shared buffer and OS cache 
are both purged on restart. Cannot see how either of those last two 
would lead to reduced serialization errors though.

> > If you create
> > a file named /tmp/bucardo-force-lock-foobar, where foobar is the name
> > of your sync, Bucardo will attempt to lock all the tables involved in
> > the sync at the start of the transaction.
> 
> Cool - that's much handier than futzing around with a bunch of iptables
> rules!  If this ever happens again I'll start there.  Is the file
> checked for on each sync run?

Yes - it is checked for at the beginning of each sync. It's designed to 
be a one-time fix: if the file exists and is used, Bucardo removes it 
at the end of the sync.

> P.S.  Thanks for all the suggestions for improvements - I'm tentatively
> planning on a Bucardo 5 upgrade after the holidays and a platform
> upgrade later in the year (newer PostgreSQL then).  Is there a current
> list of reasons why Bucardo 5 might not be ready for production in some
> cases?  I've been following the list traffic on it, and it looks from
> here to have greatly matured (and I love the new architecture).

The TODO lists some of the outstanding items at the top. None of them 
are really showstoppers except for the quoting bug. I think once that 
gets fixed and the test suite passes without errors, we can roll 
out a 5.0.0 GA version.

-- 
Greg Sabino Mullane greg at endpoint.com
End Point Corporation
PGP Key: 0x14964AC8
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 163 bytes
Desc: not available
Url : https://mail.endcrypt.com/pipermail/bucardo-general/attachments/20111209/638e58d2/attachment.bin