Greg Wooledge ([EMAIL PROTECTED]) wrote:
> PostgreSQL 6.4.2, compiled and installed on AIX 4.3.2 (a PowerPC
> system).
> Everything was going fairly well until this morning.
It's even worse today. *None* of the updates succeeded this morning.
I was able to do queries, but the data looked just like yesterday's.
UID PID PPID C STIME TTY TIME CMD
postgres 4278 39506 0 04:16:24 pts/7 0:06 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 DELETE waiting
postgres 5730 39506 0 04:13:56 pts/7 0:05 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 DELETE waiting
postgres 7738 39506 0 04:19:26 pts/7 0:04 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 DELETE waiting
postgres 19186 39506 60 04:10:02 pts/7 2:52 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 INSERT
postgres 24946 39506 0 04:10:44 pts/7 0:05 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 DELETE waiting
postgres 25442 39506 0 04:11:34 pts/7 0:05 /usr/local/pgsql/bin/postgres daemon
WWW.XXX.YYY.ZZZ sysinfo2 DELETE waiting
[...]
There were 26 postgres-owned processes running. (I killed 2, but a new
runaway process stepped up each time to take over the CPU from the
previous one. As you can see above, process 19186 was running away.)
The errlog in ~postgres was timestamped 06:15, which is when my vacuum
runs for this database. There were no error messages from any other
process but the vacuum; not even after killing two runaways.
There was a pg_vlock file from yesterday morning, when I had to kill
the postmaster to get anything to work. I removed that, but it did not
appear to make any difference. After removing that, and killing off
one or two more runaways, I finally stopped and restarted postmaster
again. I'm running a vacuum now, and hoping that it works. It's been
running for a while....
I'm seriously considering rewriting major hunks of this application so
that it doesn't need to use the PostgreSQL large objects. There are
currently 3462 files in data/base/sysinfo2, and that number is only
going to increase as this application continues to run, unless I do
away with the large objects.
But obviously I'm going to have to do *something*. It's not usable
right now.
Does the 6.5 beta version fix any known problems with large objects, or
with reliability under high loads of simultaneous DELETE/INSERT/UPDATE
statements? Or should I stick with 6.4.2?
Are there any known reliability issues with AIX 4.3.2?
