Allow BufferAccessStrategy to limit pin count.

While pinning extra buffers to look ahead, users of strategies are in
danger of using too many buffers.  For some strategies, that means
"escaping" from the ring, and in others it means forcing dirty data to
disk very frequently with associated WAL flushing.  Since external code
has no insight into any of that, allow individual strategy types to
expose a clamp that should be applied when deciding how many buffers to
pin at once.

Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
Discussion: https://postgr.es/m/CAAKRu_aJXnqsyZt6HwFLnxYEBgE17oypkxbKbT1t1geE_wvH2Q%40mail.gmail.com

3 files changed, 48 insertions, 0 deletions

diff --git a/src/backend/storage/aio/read_stream.c b/src/backend/storage/aio/read_stream.c
index b9e11a28312..9a70a81f7ae 100644
--- a/src/backend/storage/aio/read_stream.c
+++ b/src/backend/storage/aio/read_stream.c
@@ -419,6 +419,7 @@ read_stream_begin_relation(int flags,
 	size_t		size;
 	int16		queue_size;
 	int16		max_ios;
+	int			strategy_pin_limit;
 	uint32		max_pinned_buffers;
 	Oid			tablespace_id;
 	SMgrRelation smgr;
@@ -460,6 +461,10 @@ read_stream_begin_relation(int flags,
 	max_pinned_buffers = Min(max_pinned_buffers,
 							 PG_INT16_MAX - io_combine_limit - 1);
 
+	/* Give the strategy a chance to limit the number of buffers we pin. */
+	strategy_pin_limit = GetAccessStrategyPinLimit(strategy);
+	max_pinned_buffers = Min(strategy_pin_limit, max_pinned_buffers);
+
 	/* Don't allow this backend to pin more than its share of buffers. */
 	if (SmgrIsTemp(smgr))
 		LimitAdditionalLocalPins(&max_pinned_buffers);
diff --git a/src/backend/storage/buffer/freelist.c b/src/backend/storage/buffer/freelist.c
index 3611357fa30..de2ef1dd5e6 100644
--- a/src/backend/storage/buffer/freelist.c
+++ b/src/backend/storage/buffer/freelist.c
@@ -630,6 +630,48 @@ GetAccessStrategyBufferCount(BufferAccessStrategy strategy)
 }
 
 /*
+ * GetAccessStrategyPinLimit -- get cap of number of buffers that should be pinned
+ *
+ * When pinning extra buffers to look ahead, users of a ring-based strategy are
+ * in danger of pinning too much of the ring at once while performing look-ahead.
+ * For some strategies, that means "escaping" from the ring, and in others it
+ * means forcing dirty data to disk very frequently with associated WAL
+ * flushing.  Since external code has no insight into any of that, allow
+ * individual strategy types to expose a clamp that should be applied when
+ * deciding on a maximum number of buffers to pin at once.
+ *
+ * Callers should combine this number with other relevant limits and take the
+ * minimum.
+ */
+int
+GetAccessStrategyPinLimit(BufferAccessStrategy strategy)
+{
+	if (strategy == NULL)
+		return NBuffers;
+
+	switch (strategy->btype)
+	{
+		case BAS_BULKREAD:
+
+			/*
+			 * Since BAS_BULKREAD uses StrategyRejectBuffer(), dirty buffers
+			 * shouldn't be a problem and the caller is free to pin up to the
+			 * entire ring at once.
+			 */
+			return strategy->nbuffers;
+
+		default:
+
+			/*
+			 * Tell caller not to pin more than half the buffers in the ring.
+			 * This is a trade-off between look ahead distance and deferring
+			 * writeback and associated WAL traffic.
+			 */
+			return strategy->nbuffers / 2;
+	}
+}
+
+/*
  * FreeAccessStrategy -- release a BufferAccessStrategy object
  *
  * A simple pfree would do at the moment, but we would prefer that callers
diff --git a/src/include/storage/bufmgr.h b/src/include/storage/bufmgr.h
index f380f9d9a6c..07ba1a60502 100644
--- a/src/include/storage/bufmgr.h
+++ b/src/include/storage/bufmgr.h
@@ -318,6 +318,7 @@ extern BufferAccessStrategy GetAccessStrategy(BufferAccessStrategyType btype);
 extern BufferAccessStrategy GetAccessStrategyWithSize(BufferAccessStrategyType btype,
 													  int ring_size_kb);
 extern int	GetAccessStrategyBufferCount(BufferAccessStrategy strategy);
+extern int	GetAccessStrategyPinLimit(BufferAccessStrategy strategy);
 
 extern void FreeAccessStrategy(BufferAccessStrategy strategy);