alex-lists-pgsql(at)yuriev(dot)com writes:
> if the problem is caused by the the acquire lock->modify column->release
> lock on the table 1, then why does it increase significantly increase as the
> number of entries in the table 3 grows? The simulation maintains pretty much
> constant rate of new requests coming to table 3.
Are you vacuuming table 1 on a regular basis while you do this?
Accumulation of dead tuple versions in table 1 is a pretty severe
performance risk in the sort of scenario you are describing.
(I'm not sure why it would manifest as transactions showing "INSERT
waiting" state though.)
regards, tom lane