Generate less WAL during GiST, GIN and SP-GiST index build.

Instead of WAL-logging every modification during the build separately,
first build the index without any WAL-logging, and make a separate pass
through the index at the end, to write all pages to the WAL. This
significantly reduces the amount of WAL generated, and is usually also
faster, despite the extra I/O needed for the extra scan through the index.
WAL generated this way is also faster to replay.

For GiST, the LSN-NSN interlock makes this a little tricky. All pages must
be marked with a valid (i.e. non-zero) LSN, so that the parent-child
LSN-NSN interlock works correctly. We now use magic value 1 for that during
index build. Change the fake LSN counter to begin from 1000, so that 1 is
safely smaller than any real or fake LSN. 2 would've been enough for our
purposes, but let's reserve a bigger range, in case we need more special
values in the future.

Author: Anastasia Lubennikova, Andrey V. Lepikhov
Reviewed-by: Heikki Linnakangas, Dmitry Dolgov

Author: hlinnaka

src/backend/access/gin/ginbtree.c

@@ -396,7 +396,7 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
 		/* It will fit, perform the insertion */
 		START_CRIT_SECTION();
 
-		if (RelationNeedsWAL(btree->index))
+		if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 		{
 			XLogBeginInsert();
 			XLogRegisterBuffer(0, stack->buffer, REGBUF_STANDARD);
@@ -417,7 +417,7 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
 			MarkBufferDirty(childbuf);
 		}
 
-		if (RelationNeedsWAL(btree->index))
+		if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 		{
 			XLogRecPtr	recptr;
 			ginxlogInsert xlrec;
@@ -595,7 +595,7 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
 		}
 
 		/* write WAL record */
-		if (RelationNeedsWAL(btree->index))
+		if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 		{
 			XLogRecPtr	recptr;
 

src/backend/access/gin/gindatapage.c

@@ -593,7 +593,7 @@ dataBeginPlaceToPageLeaf(GinBtree btree, Buffer buf, GinBtreeStack *stack,
 		 * Great, all the items fit on a single page.  If needed, prepare data
 		 * for a WAL record describing the changes we'll make.
 		 */
-		if (RelationNeedsWAL(btree->index))
+		if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 			computeLeafRecompressWALData(leaf);
 
 		/*
@@ -719,7 +719,7 @@ dataExecPlaceToPageLeaf(GinBtree btree, Buffer buf, GinBtreeStack *stack,
 	dataPlaceToPageLeafRecompress(buf, leaf);
 
 	/* If needed, register WAL data built by computeLeafRecompressWALData */
-	if (RelationNeedsWAL(btree->index))
+	if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 	{
 		XLogRegisterBufData(0, leaf->walinfo, leaf->walinfolen);
 	}
@@ -1152,7 +1152,7 @@ dataExecPlaceToPageInternal(GinBtree btree, Buffer buf, GinBtreeStack *stack,
 	pitem = (PostingItem *) insertdata;
 	GinDataPageAddPostingItem(page, pitem, off);
 
-	if (RelationNeedsWAL(btree->index))
+	if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 	{
 		/*
 		 * This must be static, because it has to survive until XLogInsert,
@@ -1773,6 +1773,7 @@ createPostingTree(Relation index, ItemPointerData *items, uint32 nitems,
 	Pointer		ptr;
 	int			nrootitems;
 	int			rootsize;
+	bool		is_build = (buildStats != NULL);
 
 	/* Construct the new root page in memory first. */
 	tmppage = (Page) palloc(BLCKSZ);
@@ -1826,7 +1827,7 @@ createPostingTree(Relation index, ItemPointerData *items, uint32 nitems,
 	PageRestoreTempPage(tmppage, page);
 	MarkBufferDirty(buffer);
 
-	if (RelationNeedsWAL(index))
+	if (RelationNeedsWAL(index) && !is_build)
 	{
 		XLogRecPtr	recptr;
 		ginxlogCreatePostingTree data;

src/backend/access/gin/ginentrypage.c

@@ -571,7 +571,7 @@ entryExecPlaceToPage(GinBtree btree, Buffer buf, GinBtreeStack *stack,
 		elog(ERROR, "failed to add item to index page in \"%s\"",
 			 RelationGetRelationName(btree->index));
 
-	if (RelationNeedsWAL(btree->index))
+	if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 	{
 		/*
 		 * This must be static, because it has to survive until XLogInsert,

src/backend/access/gin/gininsert.c

@@ -195,6 +195,7 @@ ginEntryInsert(GinState *ginstate,
 		buildStats->nEntries++;
 
 	ginPrepareEntryScan(&btree, attnum, key, category, ginstate);
+	btree.isBuild = (buildStats != NULL);
 
 	stack = ginFindLeafPage(&btree, false, false, NULL);
 	page = BufferGetPage(stack->buffer);
@@ -347,23 +348,6 @@ ginbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 	GinInitBuffer(RootBuffer, GIN_LEAF);
 	MarkBufferDirty(RootBuffer);
 
-	if (RelationNeedsWAL(index))
-	{
-		XLogRecPtr	recptr;
-		Page		page;
-
-		XLogBeginInsert();
-		XLogRegisterBuffer(0, MetaBuffer, REGBUF_WILL_INIT | REGBUF_STANDARD);
-		XLogRegisterBuffer(1, RootBuffer, REGBUF_WILL_INIT);
-
-		recptr = XLogInsert(RM_GIN_ID, XLOG_GIN_CREATE_INDEX);
-
-		page = BufferGetPage(RootBuffer);
-		PageSetLSN(page, recptr);
-
-		page = BufferGetPage(MetaBuffer);
-		PageSetLSN(page, recptr);
-	}
 
 	UnlockReleaseBuffer(MetaBuffer);
 	UnlockReleaseBuffer(RootBuffer);
@@ -419,7 +403,18 @@ ginbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 	 * Update metapage stats
 	 */
 	buildstate.buildStats.nTotalPages = RelationGetNumberOfBlocks(index);
-	ginUpdateStats(index, &buildstate.buildStats);
+	ginUpdateStats(index, &buildstate.buildStats, true);
+
+	/*
+	 * We didn't write WAL records as we built the index, so if WAL-logging is
+	 * required, write all pages to the WAL now.
+	 */
+	if (RelationNeedsWAL(index))
+	{
+		log_newpage_range(index, MAIN_FORKNUM,
+						  0, RelationGetNumberOfBlocks(index),
+						  true);
+	}
 
 	/*
 	 * Return statistics

src/backend/access/gin/ginutil.c

@@ -662,7 +662,7 @@ ginGetStats(Relation index, GinStatsData *stats)
  * Note: nPendingPages and ginVersion are *not* copied over
  */
 void
-ginUpdateStats(Relation index, const GinStatsData *stats)
+ginUpdateStats(Relation index, const GinStatsData *stats, bool is_build)
 {
 	Buffer		metabuffer;
 	Page		metapage;
@@ -692,7 +692,7 @@ ginUpdateStats(Relation index, const GinStatsData *stats)
 
 	MarkBufferDirty(metabuffer);
 
-	if (RelationNeedsWAL(index))
+	if (RelationNeedsWAL(index) && !is_build)
 	{
 		XLogRecPtr	recptr;
 		ginxlogUpdateMeta data;

src/backend/access/gin/ginvacuum.c

@@ -759,7 +759,7 @@ ginvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
 
 	/* Update the metapage with accurate page and entry counts */
 	idxStat.nTotalPages = npages;
-	ginUpdateStats(info->index, &idxStat);
+	ginUpdateStats(info->index, &idxStat, false);
 
 	/* Finally, vacuum the FSM */
 	IndexFreeSpaceMapVacuum(info->index);

src/backend/access/gin/ginxlog.c

@@ -40,36 +40,6 @@ ginRedoClearIncompleteSplit(XLogReaderState *record, uint8 block_id)
 		UnlockReleaseBuffer(buffer);
 }
 
-static void
-ginRedoCreateIndex(XLogReaderState *record)
-{
-	XLogRecPtr	lsn = record->EndRecPtr;
-	Buffer		RootBuffer,
-				MetaBuffer;
-	Page		page;
-
-	MetaBuffer = XLogInitBufferForRedo(record, 0);
-	Assert(BufferGetBlockNumber(MetaBuffer) == GIN_METAPAGE_BLKNO);
-	page = (Page) BufferGetPage(MetaBuffer);
-
-	GinInitMetabuffer(MetaBuffer);
-
-	PageSetLSN(page, lsn);
-	MarkBufferDirty(MetaBuffer);
-
-	RootBuffer = XLogInitBufferForRedo(record, 1);
-	Assert(BufferGetBlockNumber(RootBuffer) == GIN_ROOT_BLKNO);
-	page = (Page) BufferGetPage(RootBuffer);
-
-	GinInitBuffer(RootBuffer, GIN_LEAF);
-
-	PageSetLSN(page, lsn);
-	MarkBufferDirty(RootBuffer);
-
-	UnlockReleaseBuffer(RootBuffer);
-	UnlockReleaseBuffer(MetaBuffer);
-}
-
 static void
 ginRedoCreatePTree(XLogReaderState *record)
 {
@@ -767,9 +737,6 @@ gin_redo(XLogReaderState *record)
 	oldCtx = MemoryContextSwitchTo(opCtx);
 	switch (info)
 	{
-		case XLOG_GIN_CREATE_INDEX:
-			ginRedoCreateIndex(record);
-			break;
 		case XLOG_GIN_CREATE_PTREE:
 			ginRedoCreatePTree(record);
 			break;

src/backend/access/gist/gist.c

@@ -173,7 +173,7 @@ gistinsert(Relation r, Datum *values, bool *isnull,
 						 values, isnull, true /* size is currently bogus */ );
 	itup->t_tid = *ht_ctid;
 
-	gistdoinsert(r, itup, 0, giststate, heapRel);
+	gistdoinsert(r, itup, 0, giststate, heapRel, false);
 
 	/* cleanup */
 	MemoryContextSwitchTo(oldCxt);
@@ -220,7 +220,8 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 				Buffer leftchildbuf,
 				List **splitinfo,
 				bool markfollowright,
-				Relation heapRel)
+				Relation heapRel,
+				bool is_build)
 {
 	BlockNumber blkno = BufferGetBlockNumber(buffer);
 	Page		page = BufferGetPage(buffer);
@@ -459,7 +460,7 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 		 * insertion for that. NB: The number of pages and data segments
 		 * specified here must match the calculations in gistXLogSplit()!
 		 */
-		if (RelationNeedsWAL(rel))
+		if (!is_build && RelationNeedsWAL(rel))
 			XLogEnsureRecordSpace(npage, 1 + npage * 2);
 
 		START_CRIT_SECTION();
@@ -480,18 +481,30 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 		PageRestoreTempPage(dist->page, BufferGetPage(dist->buffer));
 		dist->page = BufferGetPage(dist->buffer);
 
-		/* Write the WAL record */
-		if (RelationNeedsWAL(rel))
-			recptr = gistXLogSplit(is_leaf,
-								   dist, oldrlink, oldnsn, leftchildbuf,
-								   markfollowright);
+		/*
+		 * Write the WAL record.
+		 *
+		 * If we're building a new index, however, we don't WAL-log changes
+		 * yet. The LSN-NSN interlock between parent and child requires that
+		 * LSNs never move backwards, so set the LSNs to a value that's
+		 * smaller than any real or fake unlogged LSN that might be generated
+		 * later. (There can't be any concurrent scans during index build, so
+		 * we don't need to be able to detect concurrent splits yet.)
+		 */
+		if (is_build)
+			recptr = GistBuildLSN;
 		else
-			recptr = gistGetFakeLSN(rel);
+		{
+			if (RelationNeedsWAL(rel))
+				recptr = gistXLogSplit(is_leaf,
+									   dist, oldrlink, oldnsn, leftchildbuf,
+									   markfollowright);
+			else
+				recptr = gistGetFakeLSN(rel);
+		}
 
 		for (ptr = dist; ptr; ptr = ptr->next)
-		{
 			PageSetLSN(ptr->page, recptr);
-		}
 
 		/*
 		 * Return the new child buffers to the caller.
@@ -545,28 +558,29 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 		if (BufferIsValid(leftchildbuf))
 			MarkBufferDirty(leftchildbuf);
 
-		if (RelationNeedsWAL(rel))
+		if (is_build)
+			recptr = GistBuildLSN;
+		else
 		{
-			OffsetNumber ndeloffs = 0,
-						deloffs[1];
-
-			if (OffsetNumberIsValid(oldoffnum))
+			if (RelationNeedsWAL(rel))
 			{
-				deloffs[0] = oldoffnum;
-				ndeloffs = 1;
-			}
+				OffsetNumber ndeloffs = 0,
+							deloffs[1];
 
-			recptr = gistXLogUpdate(buffer,
-									deloffs, ndeloffs, itup, ntup,
-									leftchildbuf);
+				if (OffsetNumberIsValid(oldoffnum))
+				{
+					deloffs[0] = oldoffnum;
+					ndeloffs = 1;
+				}
 
-			PageSetLSN(page, recptr);
-		}
-		else
-		{
-			recptr = gistGetFakeLSN(rel);
-			PageSetLSN(page, recptr);
+				recptr = gistXLogUpdate(buffer,
+										deloffs, ndeloffs, itup, ntup,
+										leftchildbuf);
+			}
+			else
+				recptr = gistGetFakeLSN(rel);
 		}
+		PageSetLSN(page, recptr);
 
 		if (newblkno)
 			*newblkno = blkno;
@@ -607,7 +621,7 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
  */
 void
 gistdoinsert(Relation r, IndexTuple itup, Size freespace,
-			 GISTSTATE *giststate, Relation heapRel)
+			 GISTSTATE *giststate, Relation heapRel, bool is_build)
 {
 	ItemId		iid;
 	IndexTuple	idxtuple;
@@ -620,6 +634,7 @@ gistdoinsert(Relation r, IndexTuple itup, Size freespace,
 	state.freespace = freespace;
 	state.r = r;
 	state.heapRel = heapRel;
+	state.is_build = is_build;
 
 	/* Start from the root */
 	firststack.blkno = GIST_ROOT_BLKNO;
@@ -1252,7 +1267,8 @@ gistinserttuples(GISTInsertState *state, GISTInsertStack *stack,
 							   leftchild,
 							   &splitinfo,
 							   true,
-							   state->heapRel);
+							   state->heapRel,
+							   state->is_build);
 
 	/*
 	 * Before recursing up in case the page was split, release locks on the