On 02/05/14 02:48, bugzilla-daemon at main.slony.info wrote:
> http://www.slony.info/bugzilla/show_bug.cgi?id=289
> 
> --- Comment #5 from Tomasz Karlik <tkarlik at gmail.com> 2014-02-04 23:48:09 PST ---
> (In reply to comment #4)
>> I am still hunting this one.
>> 
>> Running a multi-client pgbench for hours and using txid snapshot in-out
>> functions all the time alongside, I was not yet able to create a single
>> occurrence of this problem. Does your application by any chance use
>> subtransactions, like "exceptions" in PL/pgSQL?
> 
> Yes, we use "exceptions" in pl/pgslq. I'll try to check, if this could be the
> source of problem.
> 
> We also use prepared transactions managed by application server. Maybe
> two-phase commits affects slony operation?

This is not a Slony specific problem.

Slony is "using" the PostgreSQL built in txid features, like Londiste
and Bucardo are, and possibly any number of other products.

What is apparently happening here is that the txid_snapshot data type
output function is producing a string representation of a snapshot, that
the type's input function rejects as invalid input. This is a serious
bug in PostgreSQL itself.

That said, since Slony is heavily depending on this data type, we are of
course very interested in this problem. I will think about an isolated
test case using subtransactions first.


Jan

-- 
Anyone who trades liberty for security deserves neither
liberty nor security. -- Benjamin Franklin