Reduce overhead of cache-clobber testing in LookupOpclassInfo().

tglsfdc · tglsfdc · commit 2f487116e998 · 2021-07-05T16:51:57.000-04:00
Commit 03ffc4d added logic to bypass all caching behavior in LookupOpclassInfo when CLOBBER_CACHE_ALWAYS is enabled. It doesn't look like I stopped to think much about what that would cost, but recent investigation shows that the cost is enormous: it roughly doubles the time needed for cache-clobber test runs. There does seem to be value in this behavior when trying to test the opclass-cache loading logic itself, but for other purposes the cost is excessive. Hence, let's back off to doing this only when debug_invalidate_system_caches_always is at least 3; or in older branches, when CLOBBER_CACHE_RECURSIVELY is defined. While here, clean up some other minor issues in LookupOpclassInfo. Re-order the code so we aren't left with broken cache entries (leading to later core dumps) in the unlikely case that we suffer OOM while trying to allocate space for a new entry. (That seems to be my oversight in 03ffc4d.) Also, in >= v13, stop allocating one array entry too many. That's evidently left over from sloppy reversion in 851b14b. Back-patch to all supported branches, mainly to reduce the runtime of cache-clobbering buildfarm animals. Discussion: https://postgr.es/m/1370856.1625428625@sss.pgh.pa.us
diff --git a/src/backend/utils/cache/relcache.c b/src/backend/utils/cache/relcache.c
@@ -1622,15 +1622,15 @@ LookupOpclassInfo(Oid operatorClassOid,
 		/* First time through: initialize the opclass cache */
 		HASHCTL		ctl;
 
+		/* Also make sure CacheMemoryContext exists */
+		if (!CacheMemoryContext)
+			CreateCacheMemoryContext();
+
 		MemSet(&ctl, 0, sizeof(ctl));
 		ctl.keysize = sizeof(Oid);
 		ctl.entrysize = sizeof(OpClassCacheEnt);
 		OpClassCache = hash_create("Operator class cache", 64,
 								   &ctl, HASH_ELEM | HASH_BLOBS);
-
-		/* Also make sure CacheMemoryContext exists */
-		if (!CacheMemoryContext)
-			CreateCacheMemoryContext();
 	}
 
 	opcentry = (OpClassCacheEnt *) hash_search(OpClassCache,
@@ -1639,39 +1639,42 @@ LookupOpclassInfo(Oid operatorClassOid,
 
 	if (!found)
 	{
-		/* Need to allocate memory for new entry */
+		/* Initialize new entry */
 		opcentry->valid = false;	/* until known OK */
 		opcentry->numSupport = numSupport;
-
-		if (numSupport > 0)
-			opcentry->supportProcs = (RegProcedure *)
-				MemoryContextAllocZero(CacheMemoryContext,
-									   (numSupport + 1) * sizeof(RegProcedure));
-		else
-			opcentry->supportProcs = NULL;
+		opcentry->supportProcs = NULL;	/* filled below */
 	}
 	else
 	{
 		Assert(numSupport == opcentry->numSupport);
 	}
 
 	/*
-	 * When testing for cache-flush hazards, we intentionally disable the
-	 * operator class cache and force reloading of the info on each call. This
-	 * is helpful because we want to test the case where a cache flush occurs
-	 * while we are loading the info, and it's very hard to provoke that if
-	 * this happens only once per opclass per backend.
+	 * When aggressively testing cache-flush hazards, we disable the operator
+	 * class cache and force reloading of the info on each call.  This models
+	 * no real-world behavior, since the cache entries are never invalidated
+	 * otherwise.  However it can be helpful for detecting bugs in the cache
+	 * loading logic itself, such as reliance on a non-nailed index.  Given
+	 * the limited use-case and the fact that this adds a great deal of
+	 * expense, we enable it only in CLOBBER_CACHE_RECURSIVELY mode.
 	 */
-#if defined(CLOBBER_CACHE_ALWAYS)
+#if defined(CLOBBER_CACHE_RECURSIVELY)
 	opcentry->valid = false;
 #endif
 
 	if (opcentry->valid)
 		return opcentry;
 
 	/*
-	 * Need to fill in new entry.
-	 *
+	 * Need to fill in new entry.  First allocate space, unless we already did
+	 * so in some previous attempt.
+	 */
+	if (opcentry->supportProcs == NULL && numSupport > 0)
+		opcentry->supportProcs = (RegProcedure *)
+			MemoryContextAllocZero(CacheMemoryContext,
+								   numSupport * sizeof(RegProcedure));
+
+	/*
 	 * To avoid infinite recursion during startup, force heap scans if we're
 	 * looking up info for the opclasses used by the indexes we would like to
 	 * reference here.
