Apply CORE-367-WAL patch

Author: dmpgpro

src/backend/access/brin/brin_pageops.c

@@ -497,6 +497,7 @@ brin_metapage_init(Page page, BlockNumber pagesPerRange, uint16 version)
 	 * revmap page to be created when the index is.
 	 */
 	metadata->lastRevmapPage = 0;
+	((PageHeader) page)->pd_lower += sizeof(BrinMetaPageData);
 }
 
 /*

src/backend/access/gin/ginbtree.c

@@ -387,7 +387,7 @@ 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);
@@ -408,7 +408,7 @@ ginPlaceToPage(GinBtree btree, GinBtreeStack *stack,
 			MarkBufferDirty(childbuf);
 		}
 
-		if (RelationNeedsWAL(btree->index))
+		if (RelationNeedsWAL(btree->index) && !btree->isBuild)
 		{
 			XLogRecPtr	recptr;
 			ginxlogInsert xlrec;
@@ -565,7 +565,7 @@ 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

@@ -591,7 +591,7 @@ 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);
 
 		/*
@@ -628,6 +628,7 @@ dataBeginPlaceToPageLeaf(GinBtree btree, Buffer buf, GinBtreeStack *stack,
 		 * subsequent insertions will probably also go to the end. This packs
 		 * the index somewhat tighter when appending to a table, which is very
 		 * common.
+		 *
 		 */
 		if (!btree->isBuild)
 		{
@@ -717,7 +718,7 @@ 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);
 	}
@@ -1150,7 +1151,7 @@ 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,
@@ -1767,6 +1768,7 @@ 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);
@@ -1814,7 +1816,7 @@ 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

@@ -570,7 +570,7 @@ 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

@@ -16,6 +16,7 @@
 
 #include "access/gin_private.h"
 #include "access/xloginsert.h"
+#include "access/generic_xlog.h"
 #include "catalog/index.h"
 #include "miscadmin.h"
 #include "storage/bufmgr.h"
@@ -191,6 +192,7 @@ ginEntryInsert(GinState *ginstate,
 		buildStats->nEntries++;
 
 	ginPrepareEntryScan(&btree, attnum, key, category, ginstate);
+	btree.isBuild = (buildStats != NULL);
 
 	stack = ginFindLeafPage(&btree, false, NULL);
 	page = BufferGetPage(stack->buffer);
@@ -341,24 +343,6 @@ 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);
-		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);
 	END_CRIT_SECTION();
@@ -412,7 +396,10 @@ ginbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 	 * Update metapage stats
 	 */
 	buildstate.buildStats.nTotalPages = RelationGetNumberOfBlocks(index);
-	ginUpdateStats(index, &buildstate.buildStats);
+	ginUpdateStats(index, &buildstate.buildStats, true);
+
+	if (RelationNeedsWAL(index))
+		generate_xlog_for_rel(index);
 
 	/*
 	 * Return statistics

src/backend/access/gin/ginutil.c

@@ -342,6 +342,8 @@ GinInitMetabuffer(Buffer b)
 	metadata->nDataPages = 0;
 	metadata->nEntries = 0;
 	metadata->ginVersion = GIN_CURRENT_VERSION;
+
+	((PageHeader) page)->pd_lower += sizeof(GinMetaPageData);
 }
 
 /*
@@ -626,7 +628,7 @@ 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;
@@ -646,7 +648,7 @@ 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

@@ -731,7 +731,7 @@ 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/gist/gist.c

@@ -162,7 +162,7 @@ gistinsert(Relation r, Datum *values, bool *isnull,
 						 values, isnull, true /* size is currently bogus */ );
 	itup->t_tid = *ht_ctid;
 
-	gistdoinsert(r, itup, 0, giststate);
+	gistdoinsert(r, itup, 0, giststate, false);
 
 	/* cleanup */
 	MemoryContextSwitchTo(oldCxt);
@@ -208,7 +208,8 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 				BlockNumber *newblkno,
 				Buffer leftchildbuf,
 				List **splitinfo,
-				bool markfollowright)
+				bool markfollowright,
+				bool is_build)
 {
 	BlockNumber blkno = BufferGetBlockNumber(buffer);
 	Page		page = BufferGetPage(buffer);
@@ -444,7 +445,7 @@ 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 (RelationNeedsWAL(rel) && !is_build)
 			XLogEnsureRecordSpace(npage, 1 + npage * 2);
 
 		START_CRIT_SECTION();
@@ -465,18 +466,20 @@ 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))
+		/*
+		 * Write the WAL record.
+		 * Do not write XLog entry if the insertion is caused by
+		 * index build process.
+		 */
+		if (RelationNeedsWAL(rel) && !is_build)
 			recptr = gistXLogSplit(is_leaf,
-								   dist, oldrlink, oldnsn, leftchildbuf,
-								   markfollowright);
+								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.
@@ -512,7 +515,8 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 		if (BufferIsValid(leftchildbuf))
 			MarkBufferDirty(leftchildbuf);
 
-		if (RelationNeedsWAL(rel))
+
+		if (RelationNeedsWAL(rel) && !is_build)
 		{
 			OffsetNumber ndeloffs = 0,
 						deloffs[1];
@@ -535,6 +539,7 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 			PageSetLSN(page, recptr);
 		}
 
+
 		if (newblkno)
 			*newblkno = blkno;
 	}
@@ -551,17 +556,28 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
 	 * the full page image. There's a chicken-and-egg problem: if we updated
 	 * the child pages first, we wouldn't know the recptr of the WAL record
 	 * we're about to write.
+	 *
+	 * We use fakeLSNs for inserions caused by index build. And when it is
+	 * finished, we write generic_xlog entry for each index page and update
+	 * all LSNs. In order to keep NSNs less then LSNs after this update, we
+	 * set NSN to InvalidXLogRecPtr, which is the smallest possible NSN.
 	 */
+
 	if (BufferIsValid(leftchildbuf))
 	{
 		Page		leftpg = BufferGetPage(leftchildbuf);
+		XLogRecPtr	fakerecptr = InvalidXLogRecPtr;
 
-		GistPageSetNSN(leftpg, recptr);
-		GistClearFollowRight(leftpg);
+		if (!is_build)
+			GistPageSetNSN(leftpg, recptr);
+		else
+			GistPageSetNSN(leftpg, fakerecptr);
 
+		GistClearFollowRight(leftpg);
 		PageSetLSN(leftpg, recptr);
 	}
 
+
 	END_CRIT_SECTION();
 
 	return is_split;
@@ -573,7 +589,8 @@ gistplacetopage(Relation rel, Size freespace, GISTSTATE *giststate,
  * so it does not bother releasing palloc'd allocations.
  */
 void
-gistdoinsert(Relation r, IndexTuple itup, Size freespace, GISTSTATE *giststate)
+gistdoinsert(Relation r, IndexTuple itup, Size freespace,
+			 GISTSTATE *giststate, bool is_build)
 {
 	ItemId		iid;
 	IndexTuple	idxtuple;
@@ -585,6 +602,7 @@ gistdoinsert(Relation r, IndexTuple itup, Size freespace, GISTSTATE *giststate)
 	memset(&state, 0, sizeof(GISTInsertState));
 	state.freespace = freespace;
 	state.r = r;
+	state.is_build = is_build;
 
 	/* Start from the root */
 	firststack.blkno = GIST_ROOT_BLKNO;
@@ -1194,7 +1212,7 @@ gistinserttuples(GISTInsertState *state, GISTInsertStack *stack,
 							   oldoffnum, NULL,
 							   leftchild,
 							   &splitinfo,
-							   true);
+							   true, state->is_build);
 
 	/*
 	 * Before recursing up in case the page was split, release locks on the

src/backend/access/gist/gistbuild.c

@@ -19,6 +19,7 @@
 #include "access/genam.h"
 #include "access/gist_private.h"
 #include "access/xloginsert.h"
+#include "access/generic_xlog.h"
 #include "catalog/index.h"
 #include "miscadmin.h"
 #include "optimizer/cost.h"
@@ -177,18 +178,12 @@ gistbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 
 	MarkBufferDirty(buffer);
 
-	if (RelationNeedsWAL(index))
-	{
-		XLogRecPtr	recptr;
-
-		XLogBeginInsert();
-		XLogRegisterBuffer(0, buffer, REGBUF_WILL_INIT);
-
-		recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_CREATE_INDEX);
-		PageSetLSN(page, recptr);
-	}
-	else
-		PageSetLSN(page, gistGetFakeLSN(heap));
+	/*
+	 * Do not write index pages to WAL unitl index build is finished.
+	 * But we still need increasing LSNs on each page, so use FakeLSN,
+	 * even for relations which eventually need WAL.
+	 */
+	PageSetLSN(page, gistGetFakeLSN(heap));
 
 	UnlockReleaseBuffer(buffer);
 
@@ -221,6 +216,15 @@ gistbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 
 	freeGISTstate(buildstate.giststate);
 
+	/*
+	 * Create generic wal records for all pages of relation, if necessary.
+	 * It seems reasonable not to generate WAL, if we recieved interrupt
+	 * signal.
+	 */
+	CHECK_FOR_INTERRUPTS();
+	if (RelationNeedsWAL(index))
+		generate_xlog_for_rel(index);
+
 	/*
 	 * Return statistics
 	 */
@@ -483,7 +487,7 @@ gistBuildCallback(Relation index,
 		 * locked, we call gistdoinsert directly.
 		 */
 		gistdoinsert(index, itup, buildstate->freespace,
-					 buildstate->giststate);
+					 buildstate->giststate, true);
 	}
 
 	/* Update tuple count and total size. */
@@ -689,7 +693,7 @@ gistbufferinginserttuples(GISTBuildState *buildstate, Buffer buffer, int level,
 							   itup, ntup, oldoffnum, &placed_to_blk,
 							   InvalidBuffer,
 							   &splitinfo,
-							   false);
+							   false, true);
 
 	/*
 	 * If this is a root split, update the root path item kept in memory. This

src/backend/access/gist/gistutil.c

@@ -914,6 +914,7 @@ gistproperty(Oid index_oid, int attno,
  * Temporary and unlogged GiST indexes are not WAL-logged, but we need LSNs
  * to detect concurrent page splits anyway. This function provides a fake
  * sequence of LSNs for that purpose.
+ * Persistent relations are also not WAL-logged while we build index.
  */
 XLogRecPtr
 gistGetFakeLSN(Relation rel)
@@ -934,7 +935,6 @@ gistGetFakeLSN(Relation rel)
 		 * Unlogged relations are accessible from other backends, and survive
 		 * (clean) restarts. GetFakeLSNForUnloggedRel() handles that for us.
 		 */
-		Assert(rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED);
 		return GetFakeLSNForUnloggedRel();
 	}
 }