Re: Unresolved error 0xC0000409 on Windows Server

On Fri, Nov 02, 2012 at 02:05:47PM -0500, Merlin Moncure wrote:
> On Fri, Nov 2, 2012 at 1:25 PM, Matthew Gerber <gerber(dot)matthew(at)gmail(dot)com> wrote:
> > I am encountering an error on my Postgres installation for Windows Server
> > 64-bit. The error was posted here a couple months ago; however, no solution
> > was found on the pgsql-bugs list, so I am reposting it to pgsql-hackers in
> > the hopes that someone will be able to help. My error message is identical
> > to the one previously posted:
> >
> > 2012-11-01 22:36:26 EDT LOG: 00000: server process (PID 7060) was
> > terminated by exception 0xC0000409
> > 2012-11-01 22:36:26 EDT DETAIL: Failed process was running: INSERT INTO
> > [snipped SQL command]

Could you post an anonymized query, post an anonymized query plan, and/or
describe the general nature of the query? Does it call functions? About how
many rows does it insert?

What server settings have you customized?
https://wiki.postgresql.org/wiki/Server_Configuration

If you could get a stack trace or minidump, that would be most helpful:
https://wiki.postgresql.org/wiki/Getting_a_stack_trace_of_a_running_PostgreSQL_backend_on_Windows

Magnus's questions for the reporter of bug #7517 are relevant, too. Does the
system have any antivirus software installed?

> > 2012-11-01 22:36:26 EDT LOG: 00000: all server processes terminated;
> > reinitializing
> > 2012-11-01 22:36:26 EDT LOCATION: PostmasterStateMachine,
> > src\backend\postmaster\postmaster.c:3135
> > 2012-11-01 22:36:36 EDT FATAL: XX000: pre-existing shared memory block is
> > still in use
> > 2012-11-01 22:36:36 EDT HINT: Check if there are any old server processes
> > still running, and terminate them.
> > 2012-11-01 22:36:36 EDT LOCATION: PGSharedMemoryCreate,
> > src\backend\port\win32_shmem.c:194

This part smells like a bug in its own right.

> hm, several times over the last couple of months (both on postgres 9.1
> and 9.2), i've seen a similar crash, but on linux. It hits the log
> like this:
>
> Execution halted (~ 200x)
> Error: segfault from C stack overflow
> Execution halted (~ 30x)
> LOG: server process (PID 19882) was terminated by signal 11: Segmentation fault
> LOG: terminating any other active server processes

> note the lack of LOG in 'Execution halted', etc. This has happened
> several times, on different servers using different workloads (but
> always under load). As of yet, I've been unable to get a core but I
> hope to get one next time it happens. I wonder if it's a similar
> cause?

Google suggests those unadorned messages originate in R. Do the affected
systems use PL/R? If so ...

> One thing I've been tempted to try is raising max_stack_depth to see
> if that helps the problem.

... that probably won't help. Depending on the specifics of the situation,
*lowering* max_stack_depth might tend to give you an ERROR instead of a crash.
Or it might just give R a bit more stack space to devour before reaching the
same crash it would otherwise reach.

Thanks,
nm