Re: In progress INSERT wrecks plans on table

On Sat, Jul 13, 2013 at 1:47 AM, Jeff Janes <jeff(dot)janes(at)gmail(dot)com> wrote:
> But I also tried it with 4 pgbench clients, and ran into a collapse of
> the performance, TPS dropping down to ~8 TPS. It is too variable to
> figure out how reliable the speed-up with this patch is, so far.
> Apparently they are all fighting over the spinlock on the
> ProcArrayLock.
>
> This is a single quad core, "Intel(R) Xeon(R) CPU X3210 @ 2.13GHz"
>
> So I agree with (3) above, about not checking
> TransactionIdIsInProgress repeatedly. Or could we change the order of
> operations so that TransactionIdIsInProgress is checked only after
> XidInMVCCSnapshot?

I haven't checked the patch in detail, but it sounds like my proposal
for CSN based snapshots[1] could help here. Using it
TransactionIdIsInProgress can be done completely lock-free. It would
include a direct dense array lookup, read barrier and a check of the
dense/sparse horizon, and if necessary a binary search in the sparse
array and another read barrier and check for sparse array version
counter.

I plan to start working on the patch next week. I hope to have a first
cut available for CF2.

[1] http://www.postgresql.org/message-id/CA+CSw_tEpJ=md1zgxPkjH6CWDnTDft4gBi=+P9SnoC+Wy3pKdA@mail.gmail.com

Regards,
Ants Aasma
--
Cybertec Schönig & Schönig GmbH
Gröhrmühlgasse 26
A-2700 Wiener Neustadt
Web: http://www.postgresql-support.de