Speedup truncations of relation forks.

When a relation is truncated, shared_buffers needs to be scanned
so that any buffers for the relation forks are invalidated in it.
Previously, shared_buffers was scanned for each relation forks, i.e.,
MAIN, FSM and VM, when VACUUM truncated off any empty pages
at the end of relation or TRUNCATE truncated the relation in place.
Since shared_buffers needed to be scanned multiple times,
it could take a long time to finish those commands especially
when shared_buffers was large.

This commit changes the logic so that shared_buffers is scanned only
one time for those three relation forks.

Author: Kirk Jamison
Reviewed-by: Masahiko Sawada, Thomas Munro, Alvaro Herrera, Takayuki Tsunakawa and Fujii Masao
Discussion: https://postgr.es/m/D09B13F772D2274BB348A310EE3027C64E2067@g01jpexmbkw24

1 files changed, 13 insertions, 29 deletions

diff --git a/src/backend/storage/freespace/freespace.c b/src/backend/storage/freespace/freespace.c
index 2383094cfd1..7e8e7e79a3d 100644
--- a/src/backend/storage/freespace/freespace.c
+++ b/src/backend/storage/freespace/freespace.c
@@ -247,16 +247,18 @@ GetRecordedFreeSpace(Relation rel, BlockNumber heapBlk)
 }
 
 /*
- * FreeSpaceMapTruncateRel - adjust for truncation of a relation.
- *
- * The caller must hold AccessExclusiveLock on the relation, to ensure that
- * other backends receive the smgr invalidation event that this function sends
- * before they access the FSM again.
+ * FreeSpaceMapPrepareTruncateRel - prepare for truncation of a relation.
  *
  * nblocks is the new size of the heap.
+ *
+ * Return the number of blocks of new FSM.
+ * If it's InvalidBlockNumber, there is nothing to truncate;
+ * otherwise the caller is responsible for calling smgrtruncate()
+ * to truncate the FSM pages, and FreeSpaceMapVacuumRange()
+ * to update upper-level pages in the FSM.
  */
-void
-FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
+BlockNumber
+FreeSpaceMapPrepareTruncateRel(Relation rel, BlockNumber nblocks)
 {
 	BlockNumber new_nfsmblocks;
 	FSMAddress	first_removed_address;
@@ -270,7 +272,7 @@ FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
 	 * truncate.
 	 */
 	if (!smgrexists(rel->rd_smgr, FSM_FORKNUM))
-		return;
+		return InvalidBlockNumber;
 
 	/* Get the location in the FSM of the first removed heap block */
 	first_removed_address = fsm_get_location(nblocks, &first_removed_slot);
@@ -285,7 +287,7 @@ FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
 	{
 		buf = fsm_readbuf(rel, first_removed_address, false);
 		if (!BufferIsValid(buf))
-			return;				/* nothing to do; the FSM was already smaller */
+			return InvalidBlockNumber;	/* nothing to do; the FSM was already smaller */
 		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 
 		/* NO EREPORT(ERROR) from here till changes are logged */
@@ -315,28 +317,10 @@ FreeSpaceMapTruncateRel(Relation rel, BlockNumber nblocks)
 	{
 		new_nfsmblocks = fsm_logical_to_physical(first_removed_address);
 		if (smgrnblocks(rel->rd_smgr, FSM_FORKNUM) <= new_nfsmblocks)
-			return;				/* nothing to do; the FSM was already smaller */
+			return InvalidBlockNumber;	/* nothing to do; the FSM was already smaller */
 	}
 
-	/* Truncate the unused FSM pages, and send smgr inval message */
-	smgrtruncate(rel->rd_smgr, FSM_FORKNUM, new_nfsmblocks);
-
-	/*
-	 * We might as well update the local smgr_fsm_nblocks setting.
-	 * smgrtruncate sent an smgr cache inval message, which will cause other
-	 * backends to invalidate their copy of smgr_fsm_nblocks, and this one too
-	 * at the next command boundary.  But this ensures it isn't outright wrong
-	 * until then.
-	 */
-	if (rel->rd_smgr)
-		rel->rd_smgr->smgr_fsm_nblocks = new_nfsmblocks;
-
-	/*
-	 * Update upper-level FSM pages to account for the truncation.  This is
-	 * important because the just-truncated pages were likely marked as
-	 * all-free, and would be preferentially selected.
-	 */
-	FreeSpaceMapVacuumRange(rel, nblocks, InvalidBlockNumber);
+	return new_nfsmblocks;
 }
 
 /*