On Mon, 14 Jun 2021 16:06:10 +0900
Yugo NAGATA <[email protected]> wrote:
> On Mon, 14 Jun 2021 08:47:40 +0200 (CEST)
> Fabien COELHO <[email protected]> wrote:
> > > I attached the fixed patch that uses return instead of break, and I
> > > confirmed
> > > that this made the progress reporting work property.
> >
> > I'm hesitating to do such a strictural change for a degenerate case linked
> > to "insane" parameters, as pg is unlikely to reach 100 million tps, ever.
> > It seems to me enough that the command is not blocked in such cases.
>
> Sure. The change from "break" to "return" is just for making the progress
> reporting work in the loop, as you mentioned. However, my original intention
> is avoiding stuck in a corner-case where a unrealistic parameter was used, and
> I agree with you that this change　is not so necessary for handling such a
> special situation.
I attached the v2 patch to clarify that I withdrew the v3 patch.
Regards
Yugo Nagata
--
Yugo NAGATA <[email protected]>
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index d7479925cb..fe75533e3e 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -3223,31 +3223,30 @@ advanceConnectionState(TState *thread, CState *st, StatsData *agg)
/*
* If --latency-limit is used, and this slot is already late
* so that the transaction will miss the latency limit even if
- * it completed immediately, skip this time slot and schedule
- * to continue running on the next slot that isn't late yet.
- * But don't iterate beyond the -t limit, if one is given.
+ * it completed immediately, skip this time slot and loop to
+ * reschedule.
*/
if (latency_limit)
{
pg_time_now_lazy(&now);
- while (thread->throttle_trigger < now - latency_limit &&
- (nxacts <= 0 || st->cnt < nxacts))
+ if (thread->throttle_trigger < now - latency_limit)
{
processXactStats(thread, st, &now, true, agg);
- /* next rendez-vous */
- thread->throttle_trigger +=
- getPoissonRand(&thread->ts_throttle_rs, throttle_delay);
- st->txn_scheduled = thread->throttle_trigger;
- }
- /*
- * stop client if -t was exceeded in the previous skip
- * loop
- */
- if (nxacts > 0 && st->cnt >= nxacts)
- {
- st->state = CSTATE_FINISHED;
+ /* stop client if -T/-t was exceeded. */
+ if (timer_exceeded || (nxacts > 0 && st->cnt >= nxacts))
+ /*
+ * For very unrealistic rates under -T, some skipped
+ * transactions are not counted because the catchup
+ * loop is not fast enough just to do the scheduling
+ * and counting at the expected speed.
+ *
+ * We do not bother with such a degenerate case.
+ */
+ st->state = CSTATE_FINISHED;
+
+ /* otherwise loop over PREPARE_THROTTLE */
break;
}
}
