"Dan Austin" <DAustin@xxxxxxxxxxxxxxxx> writes: > Solaris 10 has a tool called dtrace which reveals that pid #687 (a > postgres backend) is doing a LOT of writing to a single file (iosnoop.d > for you dtrace hackers): > UID PID D BLOCK SIZE COMM PATHNAME > 103 687 W 120384 98304 postgres > /export/data/pgsql/data/global/pgstat.tmp.687 Hmm. That's the stats collector creating a new current-stats file, which it does every half second or so. You might think about disabling stats collection if you don't have a need for it. But I'm having a hard time believing that that's the main cause of your bottleneck. The backends aren't supposed to block waiting for the stats collector ... regards, tom lane
