Add more debugging information when dropping twice pgstats entry
authorMichael Paquier <[email protected]>
Fri, 7 Jun 2024 09:46:13 +0000 (18:46 +0900)
committerMichael Paquier <[email protected]>
Fri, 7 Jun 2024 09:46:13 +0000 (18:46 +0900)
Floris Van Nee has reported a bug in the pgstats facility where a stats
entry already dropped would get again dropped.  This case should not
happen, still the error generated did not offer any details about the
stats entry getting dropped.

This commit improves the error message generated to inform about the
stats entry kind, database OID, object OID and refcount, which should
help to debug more the problem reported.  Bertrand Drouvot has been
independently able to reach this error path while writing a new feature,
and more details about the failure would have been helpful for
debugging.

Author: Andres Freund, Bertrand Drouvot
Discussion: https://postgr.es/m/20240505160915[email protected]
Discussion: https://postgr.es/m/ZkM30paAD8Cr/[email protected]
Backpatch-through: 15

src/backend/utils/activity/pgstat_shmem.c

index 4a4b69891d3cfcbad66ab5e2ed5d49c9da688283..fb79c5e771b45a372c0c78a828fade8b155dd6c7 100644 (file)
@@ -785,7 +785,11 @@ pgstat_drop_entry_internal(PgStatShared_HashEntry *shent,
     * backends to release their references.
     */
    if (shent->dropped)
-       elog(ERROR, "can only drop stats once");
+       elog(ERROR,
+            "trying to drop stats entry already dropped: kind=%s dboid=%u objoid=%u refcount=%u",
+            pgstat_get_kind_info(shent->key.kind)->name,
+            shent->key.dboid, shent->key.objoid,
+            pg_atomic_read_u32(&shent->refcount));
    shent->dropped = true;
 
    /* release refcount marking entry as not dropped */