Add a comment documenting the question of whether PrefetchBuffer should

tglsfdc · tglsfdc · commit 1b2bb33a54a8 · 2009-04-03T18:17:43.000Z
try to protect an already-existing buffer from being evicted.  This was
left as an open issue when the posix_fadvise patch was committed.  I'm
not sure there's any evidence to justify more work in this area, but we
should have some record about it in the source code.
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/storage/buffer/bufmgr.c,v 1.250 2009/03/31 22:12:48 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/storage/buffer/bufmgr.c,v 1.251 2009/04/03 18:17:43 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -153,6 +153,18 @@ PrefetchBuffer(Relation reln, ForkNumber forkNum, BlockNumber blockNum)
 		/* If not in buffers, initiate prefetch */
 		if (buf_id < 0)
 			smgrprefetch(reln->rd_smgr, forkNum, blockNum);
+
+		/*
+		 * If the block *is* in buffers, we do nothing.  This is not really
+		 * ideal: the block might be just about to be evicted, which would
+		 * be stupid since we know we are going to need it soon.  But the
+		 * only easy answer is to bump the usage_count, which does not seem
+		 * like a great solution: when the caller does ultimately touch the
+		 * block, usage_count would get bumped again, resulting in too much
+		 * favoritism for blocks that are involved in a prefetch sequence.
+		 * A real fix would involve some additional per-buffer state, and
+		 * it's not clear that there's enough of a problem to justify that.
+		 */
 	}
 #endif /* USE_PREFETCH */
 }

Original file line number	Diff line number	Diff line change
`@@ -8,7 +8,7 @@`
`8`	`8`	`*`
`9`	`9`	`*`
`10`	`10`	`* IDENTIFICATION`
`11`		`- * $PostgreSQL: pgsql/src/backend/storage/buffer/bufmgr.c,v 1.250 2009/03/31 22:12:48 tgl Exp $`
	`11`	`+ * $PostgreSQL: pgsql/src/backend/storage/buffer/bufmgr.c,v 1.251 2009/04/03 18:17:43 tgl Exp $`
`12`	`12`	`*`
`13`	`13`	`*-------------------------------------------------------------------------`
`14`	`14`	`*/`
`@@ -153,6 +153,18 @@ PrefetchBuffer(Relation reln, ForkNumber forkNum, BlockNumber blockNum)`
`153`	`153`	`/* If not in buffers, initiate prefetch */`
`154`	`154`	`if (buf_id < 0)`
`155`	`155`	`smgrprefetch(reln->rd_smgr, forkNum, blockNum);`
	`156`	`+`
	`157`	`+ /*`
	`158`	`+ * If the block is in buffers, we do nothing. This is not really`
	`159`	`+ * ideal: the block might be just about to be evicted, which would`
	`160`	`+ * be stupid since we know we are going to need it soon. But the`
	`161`	`+ * only easy answer is to bump the usage_count, which does not seem`
	`162`	`+ * like a great solution: when the caller does ultimately touch the`
	`163`	`+ * block, usage_count would get bumped again, resulting in too much`
	`164`	`+ * favoritism for blocks that are involved in a prefetch sequence.`
	`165`	`+ * A real fix would involve some additional per-buffer state, and`
	`166`	`+ * it's not clear that there's enough of a problem to justify that.`
	`167`	`+ */`
`156`	`168`	`}`
`157`	`169`	`#endif /* USE_PREFETCH */`
`158`	`170`	`}`