Bruce Momjian wrote:
> Interesting. So it times out a transaction to release its locks. Does
> it time from the start of the transaction? I assume so. We do have
> statment_timeout so maybe transaction_timeout would be the proper name.
> Timing out on just the idle state seems strange to me.
Timing out on the start of the transaction isn't a good idea; what about
long running transactions? You'd have to set the timeout very high (lets
say to have nightly imports) which would make the "idle in transaction
and not responding any more" errors detected very late in online situations.
So the timeout should start after each statement's completion.
Regards,
Andreas
