Re: anole: assorted stability problems

On 2015-05-20 16:21:57 -0300, Alvaro Herrera wrote:
> In HEAD only.  Previous branches seem mostly clean, so there's something
> going wrong.  Spinlocks going wrong perhaps?
>
> http://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=anole&dt=2015-05-20%2016%3A30%3A26&stg=check
> ! PANIC:  stuck spinlock (c00000000d6f4140) detected at lwlock.c:816
> ! server closed the connection unexpectedly
> !     This probably means the server terminated abnormally
> !     before or while processing the request.
> ! connection to server was lost
>
> http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=anole&dt=2015-05-09%2020%3A30%3A29
> ! PANIC:  semop(id=0) failed: Result too large
> ! server closed the connection unexpectedly
> !     This probably means the server terminated abnormally
> !     before or while processing the request.
> ! connection to server was lost
>
> http://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=anole&dt=2015-05-05%2018%3A39%3A38&stg=check
> ! FATAL:  semop(id=0) failed: File too large
> ! server closed the connection unexpectedly
> !     This probably means the server terminated abnormally
> !     before or while processing the request.
> ! connection to server was lost
>
> http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=anole&dt=2015-05-03%2012%3A30%3A18
> ! PANIC:  semop(id=-1073741824) failed: Invalid argument
> ! server closed the connection unexpectedly
> !     This probably means the server terminated abnormally
> !     before or while processing the request.
> ! connection to server was lost
>
> http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=anole&dt=2015-04-29%2004%3A30%3A25
> ! PANIC:  stuck spinlock (c00000000d335360) detected at lwlock.c:767
> ! server closed the connection unexpectedly
> !     This probably means the server terminated abnormally
> !     before or while processing the request.
> ! connection to server was lost

And now:

! FATAL:  semop(id=-2013265921) failed: Invalid argument
! CONTEXT:  SQL statement "CREATE TEMP TABLE brin_result (cid tid)"
! PL/pgSQL function inline_code_block line 20 at SQL statement
! server closed the connection unexpectedly
!     This probably means the server terminated abnormally
!     before or while processing the request.
! connection to server was lost

Uhm:

void
s_init_lock_sema(volatile slock_t *lock)
{static int    counter = 0;
*lock = (++counter) % NUM_SPINLOCK_SEMAPHORES;
}

One problem here might be that counter is signed. Once s_init_lock_sema
has been called often enough for counter to wrap around strange things
will happen.  But - I don't see why this codepatch would even be hit
once on this platform? It's only built !HAVE_SPINLOCKS which isn't the
case on anole.  So this appears to be an independent bug (9.4+).

One that has lead me to find an atomics bug (9.5+, stupid forgotten
codepath for atomics on spinlocks on semaphores) - which again should be
independent, because it's again is only relevant when spinlocks aren't
used...

I'll fix both.


But that leaves this problem.