PROBLEM: Postgres fails to start due to kernel issue?

Hi,

I noticed after upgrading to 3.2.0-08812-gc2bc3a3 that Postgres-9.1 server fails to start with error:

FATAL:  could not create shared memory segment: Function not implemented
DETAIL:  Failed system call was shmget(key=5432001, size=30523392, 03600).

After some googling, I thought it was a SYSV shared memory sizing issue but manually changing kernel.shmmax and kernel.shmall had no effect.

The problem has been introduced between 3.2.0-08694-g53999bf and 3.2.0-08812-gc2bc3a3.

Let me know if you need any more information. Bisecting is *very* slow on this machine!

Broken version (latest):

$ sudo sh scripts/ver_linux

Linux s1 3.2.0-09429-g90a4c0f #133 SMP Thu Jan 19 15:17:03 GMT 2012 i686 Pentium III (Coppermine) GenuineIntel GNU/Linux

Gnu C                  4.6.2
Gnu make               3.82
binutils               2.21.1
util-linux             2.19.1
mount                  support
module-init-tools      3.16
e2fsprogs              1.41.14
Linux C Library        2.14.1
Dynamic linker (ldd)   2.14.1
Procps                 3.2.8
Net-tools              1.60_p20110409135728
Kbd                    1.15.3wip
Sh-utils               8.7

$ sudo sysctl -a | grep shm
kernel.shmmax = 33554432
kernel.shmall = 2097152
kernel.shmmni = 4096
kernel.shm_rmid_forced = 0
vm.hugetlb_shm_group = 0

Working version (last good backup):

$ sudo sh scripts/ver_linux

Linux s1 3.2.0 #124 SMP Thu Jan 5 08:06:00 GMT 2012 i686 Pentium III (Coppermine) GenuineIntel GNU/Linux

Gnu C                  4.6.2
Gnu make               3.82
binutils               2.21.1
util-linux             2.19.1
mount                  support
module-init-tools      3.16
e2fsprogs              1.41.14
Linux C Library        2.14.1
Dynamic linker (ldd)   2.14.1
Procps                 3.2.8
Net-tools              1.60_p20110409135728
Kbd                    1.15.3wip
Sh-utils               8.7

$ sudo sysctl -a | grep shm
kernel.shmmax = 33554432
kernel.shmall = 2097152
kernel.shmmni = 4096
kernel.shm_rmid_forced = 0
vm.hugetlb_shm_group = 0

Dave.

If you run strace on it, does it fail with ENOSYS? This might be related to an issue Peter just fixed.

josh

Yes, it almost certainly is. Fixed in rc1.

Sent from my Android phone with K-9 Mail. Please excuse my brevity.

Correct, working again under 3.3.0-rc1. Thanks!

Linux s1 3.3.0-rc1 #135 SMP Fri Jan 20 09:59:56 GMT 2012 i686 Pentium III (Coppermine) GenuineIntel GNU/Linux