Re: Issue with pg_stat_subscription_stats

Hi,

On 2022-07-04 11:01:01 +0900, Masahiko Sawada wrote:
> I've attached the patch, fix_drop_subscriptions_stats.patch, to fix it.

LGTM. Unless somebody sees a reason not to, I'm planning to commit that to 15
and HEAD.

> I've also attached another PoC patch,
> poc_create_subscription_stats.patch, to create the stats entry when
> creating the subscription, which address the issue reported in this
> thread; the pg_stat_reset_subscription_stats() doesn't update the
> stats_reset if no error is reported yet.

It'd be good for this to include a test.

> diff --git a/src/backend/utils/activity/pgstat_subscription.c b/src/backend/utils/activity/pgstat_subscription.c
> index e1072bd5ba..ef318b7422 100644
> --- a/src/backend/utils/activity/pgstat_subscription.c
> +++ b/src/backend/utils/activity/pgstat_subscription.c
> @@ -47,8 +47,20 @@ pgstat_report_subscription_error(Oid subid, bool is_apply_error)
> void
> pgstat_create_subscription(Oid subid)
> {
> + PgStat_EntryRef *entry_ref;
> + PgStatShared_Subscription *shstatent;
> +
> pgstat_create_transactional(PGSTAT_KIND_SUBSCRIPTION,
> InvalidOid, subid);
> +
> + entry_ref = pgstat_get_entry_ref_locked(PGSTAT_KIND_SUBSCRIPTION,
> + InvalidOid, subid,
> + false);
> + shstatent = (PgStatShared_Subscription *) entry_ref->shared_stats;
> +
> + memset(&shstatent->stats, 0, sizeof(shstatent->stats));
> +
> + pgstat_unlock_entry(entry_ref);
> }
>
> /*

I think most of this could just be pgstat_reset_entry().

Greetings,

Andres Freund