While putting a test server (dual PIII 1GHz, 512M ram, ide single disk for
system and $PGDATA) under high load, I keep running into:
Mar 18 12:33:19 c257lx postgres[25277]: [6] NOTICE: ALTER TABLE / ADD
PRIMARY KEY will create implicit index 'test_pkey' for table 'test'
Mar 18 12:33:22 c257lx postgres[25273]: [22] LOG: recycled transaction log
file 0000000F00000068
Mar 18 12:33:45 c257lx postgres[25277]: [7] PANIC: open of
/home/db/pgsql/data/pg_clog/0406 failed: No such file o
r directory
Mar 18 12:33:46 c257lx postgres[15667]: [1] LOG: server process (pid 25277)
was terminated by signal 6
This always happens several passes into a loop (second or third with default
shared memory settings, much later with with 4096 shared buffers and 32
checkpoint segments). The workload in question goes like this: for each file
in a zip archive: unzip $file | copy to test from stdin; index
'dry_quad_hold'; insert into table1 from select ... where ... (inner joins
to "denormalize" the data) <3 of these in parallel subshells> ; add indexes
to the "sub-tables" in 3 parallel subshells; loop over roughly 100 text
files of 340,000 rows and 80 columns each. (It is designed to test my "LVM
snapshots" script.)
The workload being tested is handled without problem on production servers
running 7.2.3 -- quad (dual) PIII Xeon 500 MHz w, 6 GB (2 GB) RAM, SCSI
(IDE; "3ware") RAID5 decicated partition for $PGDATA.
Searching for <str>postgresql "terminated by signal 6"</str> led me to this
thread:
http://archives.postgresql.org/pgsql-performance/2003-01/msg00322.php
Is this a 7.3 regression? ... a memory size thing? I'm installing 7.2.4 on
the test machine and will re-run the same script there.
Thanks for help/suggestions,
Murthy