Someone else was doing the vacuum that didn't complete this last time
and they started it at night so no other queries were running. I wasn't
monitoring I/O usage at the time and in the past I just always removed
the indexes and vacuumed when this happened.
This is on a Solaris server, would you suggest any additional commands
besides iostat to monitor the i/o?
Ellen
Tom Lane wrote:
Ellen Cyran <ellen@xxxxxxxxxxxxxxxxx> writes:
Hm, that should be OK. What do you have maintenance_work_mem set to?
It's set at the default 16384.
That should be plenty for getting rid of a million or so tuples. I'm
wondering if you are seeing some weird locking effect. Is the VACUUM
constantly busy with I/O or does it sit and wait at points? Do you have
other queries actively accessing the table during the VACUUM?
regards, tom lane
