On 5/16/06, Tom Lane <tgl@xxxxxxxxxxxxx> wrote:
"Tony Wasson" <ajwasson@xxxxxxxxx> writes: > When I saw the same error as you, the stats collector process was > missing. The collector, or the buffer process? The reported message would be emitted by the buffer process, after which it would immediately exit. (The collector would go away too once it noticed EOF on its input.) By and by the postmaster should start a fresh pair of processes.
The stats collector was dead and would not respawn. Our options seemed limited to restarting postmaster or ignoring the error. Here was what the process list looked like: kangaroo:~ twasson$ ps waux | grep post pgsql 574 0.0 -0.0 460104 832 p0 S Wed06AM 10:26.98 /usr/local/pgsql/bin/postmaster -D /Volumes/Vol0/pgsql-data pgsql 578 0.0 -5.2 460356 108620 p0 S Wed06AM 27:43.68 postgres: writer process twasson 23844 0.0 -0.0 18172 688 std S+ 10:05AM 0:00.01 grep post
IIRC, the postmaster's spawning is rate-limited to once a minute, so if the new buffer were immediately dying with the same error, that would explain your observation of once-a-minute messages. This all still leaves us no closer to understanding *why* the recv() is failing, though. What it does suggest is that the problem is a hard, repeatable error when it does occur, which makes me loath to put in the quick-fix "retry on EAGAIN" that I previously suggested. If it is a hard error then that will just convert the problem into a busy-loop that'll eat all your CPU cycles ... not much of an improvement ...
