Tom Lane writes:
Also, increasing checkpoint_segments and possibly wal_buffers helps a
lot for write-intensive loads.
Following up on those two recomendations from Tom.
Tom mentioned in a different message that if the inserst are small that
increasing wal_buffers would not help.
How about checkpoint_segments?
Also commit_delays seems like may be helpfull. Is it too much a risk to set
commit_delays to 1 second? If the machine has any problems while having a
higher commit_delay will the end result simply be that a larger amount of
transactions will be rolled back?
p.s. did not CC Tom because he uses an RBL which is rather "selective" about
what it lets through (one of the worst in my opinion).
