Re: UTC4115FATAL: the database system is in recovery mode

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Fri, May 27, 2011 at 2:26 PM, Mathew Samuel
> <Mathew(dot)Samuel(at)entrust(dot)com> wrote:
>> 2011-03-28 10:44:42 UTC3932PANIC: cannot abort transaction 1827110275, it
>> was already committed
>>
>> Not sure if this is a known bug (or if it is a bug at all or something I can
>> address using different configuration) but I thought I would post here first
>> if any one might be familiar with this issue and suggest a possible
>> solution. Any ideas?

> Not really, although I've seen a few other crashes attributable to
> trying to cancel VACUUM FULL. As of PostgreSQL 9.0, VACUUM FULL has
> been completely rewritten, so whatever bug is causing this problem is
> almost certainly gone in more recent versions.

That was in fact one of the primary motivations for the rewrite.

> It would be nice to
> track done and fix whatever is wrong in the older branches, but we'd
> probably need a reproducible test case of some sort. I don't think it
> craps out like that EVERY time you interrupt it...

It does if you interrupt it after the midstream commit :-(. See around
line 2800 in commands/vacuum.c in 8.4, for example, and note that the
lame HOLD_INTERRUPTS hack isn't in the version the OP is running.

regards, tom lane