Stephen Tyler <stephen(at)stephen-tyler(dot)com> writes:
> My expectation would be that postgresql would issue an fsync() (or perhaps a
> few such calls) at the end of the transaction. But this does not seem to be
> the case:
fsyncs mostly happen as a result of checkpoint activity. I wonder
whether you have done any tuning of checkpoint intervals,
shared_buffers, etc.
The very long times to complete the fsyncs sound like a kernel or driver
issue; not much we can do to help you with that.
regards, tom lane