Hi, On 2026-06-10 17:36:24 +0000, Pavlo Golub wrote: > @@ -1490,6 +1509,10 @@ pgss_store(const char *query, int64 queryId, > else if (planOrigin == PLAN_STMT_CACHE_CUSTOM) > entry->counters.custom_plan_calls++; > > + /* Record the start time of this execution, if provided */ > + if (exec_start != 0) > + entry->last_execution_start = exec_start; > + > SpinLockRelease(&entry->mutex); > }
FWIW, I remain opposed to adding *any* additional thing under the spinlock. We first need to fix the design of pgss, then we can start discussing expanding it even further. I've seen way too many instances being brought to their knee because of pgss contention. We shouldn't make it even worse. Greetings, Andres Freund
