Hello list,
I've been working a bit today to resolve the error I'm seeing mentioned
in the title, and I was hoping you might have some insight. I've
managed to semi-reliably reproduce this based on two different tasks:
- a pg_dumpall from my backup server will throw this, but not always on
the same database. Sometimes it seems to get 'stuck' on one, but later
if I try it again it'll throw the error on a different DB. The
pg_dumpall appears to work perfectly when run locally.
- A unit test for a Rails app I'm developing will consistently throw
this error, though oddly I can run the problem statement in psql, over
the network, without problem.
My biggest problem is the lack of any real error message on the server.
I don't see anything wrong in the system logs, and there's no core
file in the /var/postgresql directory. I did a 'vacuumdb -afz' just as
a shot in the dark, without affect. Pretty much all I see in the logs
is this:
<USER%DB x.y.z.a(51478) 487>LOG: unexpected EOF on client connection
Googling turned up a few reports suggesting bad hardware, or corrupted
indexes, but I don't think that's the case here.
Any starting points or ideas would be greatly appreciated.
Basic server info:
- Less than 3 months old
- OpenBSD 3.8, patched fully, x86 w/ a P4
- 2G of Ram, ~1.5G free usually
- /var/postgresql partition has about 8G free, and is backed by a AMI
hardware raid-1 array.
- PostgreSQL version 8.0.3 on server and clients, built locally from
OpenBSD ports
- ruby-postgres adapter version 0.7.1 used in the rails project
Special OS Config bits:
- kern.seminfo sysctl tree:
kern.seminfo.semmni=256
kern.seminfo.semmns=2048
kern.seminfo.semmnu=30
kern.seminfo.semmsl=60
kern.seminfo.semopm=100
kern.seminfo.semume=10
kern.seminfo.semusz=100
kern.seminfo.semvmx=32767
kern.seminfo.semaem=16384
- Bits from _postgresql's ulimit:
$ ulimit -a
time(cpu-seconds) unlimited
file(blocks) unlimited
coredump(blocks) unlimited
data(kbytes) 1048576
stack(kbytes) 8192
lockedmem(kbytes) 635424
memory(kbytes) 1905136
nofiles(descriptors) 768
processes 532
- postgresql.conf changes:
max_connections = 200
syslog_facility = 'LOCAL0'
syslog_ident = 'postgres'
log_duration = true
stats_start_collector = true
stats_command_string = false
stats_block_level = false
stats_row_level = false
Thanks in advance!
--
Dave Steinberg
http://www.geekisp.com/
http://www.steinbergcomputing.com/
