Re: [HACKERS] Aborted Transaction During Vacuum

Tom Lane wrote:

>
> Interesting. Is there any additional message appearing in the
> postmaster log? Is a core file being generated? (look in the
> data/base/ subdirectory for the database in question) If there
> is a corefile, a debugger backtrace from it would be helpful.
>
> regards, tom lane

Tom,

I tried the 'vacuum verbose analyze' again today. I get the same error
with the AbortTransaction. There is a core file generated but no pg_vlock
file. The core is over 1 Gig in size (38 Megs gzipped) so I'm not so sure
you'd want to get that (you can have it if you want it though!). Plus,
there seems to be nothing written to the postmaster.log file (I re-started
the postmaster before the vacuum using 'nohup postmaster -i -B 15000 -o -F
> postmaster.log&').

Oliver Elphick (Oliver(dot)Elphick(at)lfix(dot)co(dot)uk) told me that this sounds
like like a bug you patched for him a few days ago:

> It does sound very like the bug that was found a few days back where the

> pg_vlock file gets deleted by a mistaken call to vc_abort(). This only

> gets called in the analyze code.

We're probably talking about the same bug. When I try just 'vacuum verbose'
without the analyze, the vacuum completes just fine. So it must be within
the analyze code.

-Tony