More GIN refactoring.

Separate the insertion payload from the more static portions of GinBtree.
GinBtree now only contains information related to searching the tree, and
the information of what to insert is passed separately.

Add root block number to GinBtree, instead of passing it around all the
functions as argument.

Split off ginFinishSplit() from ginInsertValue(). ginFinishSplit is
responsible for finding the parent and inserting the downlink to it.

src/backend/access/gin/ginentrypage.c

@@ -112,6 +112,7 @@ GinFormTuple(GinState *ginstate,
 	if (newsize != IndexTupleSize(itup))
 	{
 		itup = repalloc(itup, newsize);
+
 		/*
 		 * PostgreSQL 9.3 and earlier did not clear this new space, so we
 		 * might find uninitialized padding when reading tuples from disk.
@@ -431,22 +432,26 @@ entryGetLeftMostPage(GinBtree btree, Page page)
 }
 
 static bool
-entryIsEnoughSpace(GinBtree btree, Buffer buf, OffsetNumber off)
+entryIsEnoughSpace(GinBtree btree, Buffer buf, OffsetNumber off,
+				   GinBtreeEntryInsertData *insertData)
 {
-	Size		itupsz = 0;
+	Size		releasedsz = 0;
+	Size		addedsz;
 	Page		page = BufferGetPage(buf);
 
-	Assert(btree->entry);
+	Assert(insertData->entry);
 	Assert(!GinPageIsData(page));
 
-	if (btree->isDelete)
+	if (insertData->isDelete)
 	{
 		IndexTuple	itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, off));
 
-		itupsz = MAXALIGN(IndexTupleSize(itup)) + sizeof(ItemIdData);
+		releasedsz = MAXALIGN(IndexTupleSize(itup)) + sizeof(ItemIdData);
 	}
 
-	if (PageGetFreeSpace(page) + itupsz >= MAXALIGN(IndexTupleSize(btree->entry)) + sizeof(ItemIdData))
+	addedsz = MAXALIGN(IndexTupleSize(insertData->entry)) + sizeof(ItemIdData);
+
+	if (PageGetFreeSpace(page) + releasedsz >= addedsz)
 		return true;
 
 	return false;
@@ -457,42 +462,42 @@ entryIsEnoughSpace(GinBtree btree, Buffer buf, OffsetNumber off)
  * should update it, update old child blkno to new right page
  * if child split occurred
  */
-static BlockNumber
-entryPreparePage(GinBtree btree, Page page, OffsetNumber off)
+static void
+entryPreparePage(GinBtree btree, Page page, OffsetNumber off,
+				 GinBtreeEntryInsertData *insertData, BlockNumber updateblkno)
 {
-	BlockNumber ret = InvalidBlockNumber;
-
-	Assert(btree->entry);
+	Assert(insertData->entry);
 	Assert(!GinPageIsData(page));
 
-	if (btree->isDelete)
+	if (insertData->isDelete)
 	{
 		Assert(GinPageIsLeaf(page));
 		PageIndexTupleDelete(page, off);
 	}
 
-	if (!GinPageIsLeaf(page) && btree->rightblkno != InvalidBlockNumber)
+	if (!GinPageIsLeaf(page) && updateblkno != InvalidBlockNumber)
 	{
 		IndexTuple	itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, off));
 
-		GinSetDownlink(itup, btree->rightblkno);
-		ret = btree->rightblkno;
+		GinSetDownlink(itup, updateblkno);
 	}
-
-	btree->rightblkno = InvalidBlockNumber;
-
-	return ret;
 }
 
 /*
  * Place tuple on page and fills WAL record
  *
  * If the tuple doesn't fit, returns false without modifying the page.
+ *
+ * On insertion to an internal node, in addition to inserting the given item,
+ * the downlink of the existing item at 'off' is updated to point to
+ * 'updateblkno'.
  */
 static bool
 entryPlaceToPage(GinBtree btree, Buffer buf, OffsetNumber off,
+				 void *insertPayload, BlockNumber updateblkno,
 				 XLogRecData **prdata)
 {
+	GinBtreeEntryInsertData *insertData = insertPayload;
 	Page		page = BufferGetPage(buf);
 	OffsetNumber placed;
 	int			cnt = 0;
@@ -502,13 +507,17 @@ entryPlaceToPage(GinBtree btree, Buffer buf, OffsetNumber off,
 	static ginxlogInsert data;
 
 	/* quick exit if it doesn't fit */
-	if (!entryIsEnoughSpace(btree, buf, off))
+	if (!entryIsEnoughSpace(btree, buf, off, insertData))
 		return false;
 
 	*prdata = rdata;
-	data.updateBlkno = entryPreparePage(btree, page, off);
+	entryPreparePage(btree, page, off, insertData, updateblkno);
+	data.updateBlkno = updateblkno;
 
-	placed = PageAddItem(page, (Item) btree->entry, IndexTupleSize(btree->entry), off, false, false);
+	placed = PageAddItem(page,
+						 (Item) insertData->entry,
+						 IndexTupleSize(insertData->entry),
+						 off, false, false);
 	if (placed != off)
 		elog(ERROR, "failed to add item to index page in \"%s\"",
 			 RelationGetRelationName(btree->index));
@@ -517,7 +526,7 @@ entryPlaceToPage(GinBtree btree, Buffer buf, OffsetNumber off,
 	data.blkno = BufferGetBlockNumber(buf);
 	data.offset = off;
 	data.nitem = 1;
-	data.isDelete = btree->isDelete;
+	data.isDelete = insertData->isDelete;
 	data.isData = false;
 	data.isLeaf = GinPageIsLeaf(page) ? TRUE : FALSE;
 
@@ -545,12 +554,10 @@ entryPlaceToPage(GinBtree btree, Buffer buf, OffsetNumber off,
 	cnt++;
 
 	rdata[cnt].buffer = InvalidBuffer;
-	rdata[cnt].data = (char *) btree->entry;
-	rdata[cnt].len = IndexTupleSize(btree->entry);
+	rdata[cnt].data = (char *) insertData->entry;
+	rdata[cnt].len = IndexTupleSize(insertData->entry);
 	rdata[cnt].next = NULL;
 
-	btree->entry = NULL;
-
 	return true;
 }
 
@@ -561,8 +568,11 @@ entryPlaceToPage(GinBtree btree, Buffer buf, OffsetNumber off,
  * an equal number!
  */
 static Page
-entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogRecData **prdata)
+entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off,
+			   void *insertPayload,
+			   BlockNumber updateblkno, XLogRecData **prdata)
 {
+	GinBtreeEntryInsertData *insertData = insertPayload;
 	OffsetNumber i,
 				maxoff,
 				separator = InvalidOffsetNumber;
@@ -583,8 +593,9 @@ entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogR
 
 	*prdata = rdata;
 	data.leftChildBlkno = (GinPageIsLeaf(lpage)) ?
-		InvalidOffsetNumber : GinGetDownlink(btree->entry);
-	data.updateBlkno = entryPreparePage(btree, lpage, off);
+		InvalidOffsetNumber : GinGetDownlink(insertData->entry);
+	data.updateBlkno = updateblkno;
+	entryPreparePage(btree, lpage, off, insertData, updateblkno);
 
 	maxoff = PageGetMaxOffsetNumber(lpage);
 	ptr = tupstore;
@@ -593,8 +604,8 @@ entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogR
 	{
 		if (i == off)
 		{
-			size = MAXALIGN(IndexTupleSize(btree->entry));
-			memcpy(ptr, btree->entry, size);
+			size = MAXALIGN(IndexTupleSize(insertData->entry));
+			memcpy(ptr, insertData->entry, size);
 			ptr += size;
 			totalsize += size + sizeof(ItemIdData);
 		}
@@ -608,8 +619,8 @@ entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogR
 
 	if (off == maxoff + 1)
 	{
-		size = MAXALIGN(IndexTupleSize(btree->entry));
-		memcpy(ptr, btree->entry, size);
+		size = MAXALIGN(IndexTupleSize(insertData->entry));
+		memcpy(ptr, insertData->entry, size);
 		ptr += size;
 		totalsize += size + sizeof(ItemIdData);
 	}
@@ -667,20 +678,23 @@ entrySplitPage(GinBtree btree, Buffer lbuf, Buffer rbuf, OffsetNumber off, XLogR
 }
 
 /*
- * Prepare the state in 'btree' for inserting a downlink for given buffer.
+ * Construct insertion payload for inserting the downlink for given buffer.
  */
-static void
+static void *
 entryPrepareDownlink(GinBtree btree, Buffer lbuf)
 {
+	GinBtreeEntryInsertData *insertData;
 	Page		lpage = BufferGetPage(lbuf);
+	BlockNumber lblkno = BufferGetBlockNumber(lbuf);
 	IndexTuple	itup;
 
 	itup = getRightMostTuple(lpage);
 
-	btree->entry = GinFormInteriorTuple(itup,
-										lpage,
-										BufferGetBlockNumber(lbuf));
-	btree->rightblkno = GinPageGetOpaque(lpage)->rightlink;
+	insertData = palloc(sizeof(GinBtreeEntryInsertData));
+	insertData->entry = GinFormInteriorTuple(itup, lpage, lblkno);
+	insertData->isDelete = false;
+
+	return insertData;
 }
 
 /*
@@ -724,6 +738,7 @@ ginPrepareEntryScan(GinBtree btree, OffsetNumber attnum,
 	memset(btree, 0, sizeof(GinBtreeData));
 
 	btree->index = ginstate->index;
+	btree->rootBlkno = GIN_ROOT_BLKNO;
 	btree->ginstate = ginstate;
 
 	btree->findChildPage = entryLocateEntry;
@@ -743,5 +758,4 @@ ginPrepareEntryScan(GinBtree btree, OffsetNumber attnum,
 	btree->entryAttnum = attnum;
 	btree->entryKey = key;
 	btree->entryCategory = category;
-	btree->isDelete = FALSE;
 }

src/backend/access/gin/ginget.c

@@ -374,7 +374,7 @@ restartScanEntry:
 	ginPrepareEntryScan(&btreeEntry, entry->attnum,
 						entry->queryKey, entry->queryCategory,
 						ginstate);
-	stackEntry = ginFindLeafPage(&btreeEntry, GIN_ROOT_BLKNO, true);
+	stackEntry = ginFindLeafPage(&btreeEntry, true);
 	page = BufferGetPage(stackEntry->buffer);
 	needUnlock = TRUE;
 

src/backend/access/gin/gininsert.c

@@ -163,17 +163,20 @@ ginEntryInsert(GinState *ginstate,
 			   GinStatsData *buildStats)
 {
 	GinBtreeData btree;
+	GinBtreeEntryInsertData insertdata;
 	GinBtreeStack *stack;
 	IndexTuple	itup;
 	Page		page;
 
+	insertdata.isDelete = FALSE;
+
 	/* During index build, count the to-be-inserted entry */
 	if (buildStats)
 		buildStats->nEntries++;
 
 	ginPrepareEntryScan(&btree, attnum, key, category, ginstate);
 
-	stack = ginFindLeafPage(&btree, GIN_ROOT_BLKNO, false);
+	stack = ginFindLeafPage(&btree, false);
 	page = BufferGetPage(stack->buffer);
 
 	if (btree.findItem(&btree, stack))
@@ -201,7 +204,7 @@ ginEntryInsert(GinState *ginstate,
 		itup = addItemPointersToLeafTuple(ginstate, itup,
 										  items, nitem, buildStats);
 
-		btree.isDelete = TRUE;
+		insertdata.isDelete = TRUE;
 	}
 	else
 	{
@@ -211,8 +214,8 @@ ginEntryInsert(GinState *ginstate,
 	}
 
 	/* Insert the new or modified leaf tuple */
-	btree.entry = itup;
-	ginInsertValue(&btree, stack, buildStats);
+	insertdata.entry = itup;
+	ginInsertValue(&btree, stack, &insertdata, buildStats);
 	pfree(itup);
 }
 

src/backend/access/gin/ginxlog.c

@@ -774,7 +774,7 @@ ginContinueSplit(ginIncompleteSplit *split)
 	GinState	ginstate;
 	Relation	reln;
 	Buffer		buffer;
-	GinBtreeStack stack;
+	GinBtreeStack *stack;
 
 	/*
 	 * elog(NOTICE,"ginContinueSplit root:%u l:%u r:%u",  split->rootBlkno,
@@ -802,22 +802,22 @@ ginContinueSplit(ginIncompleteSplit *split)
 	}
 	else
 	{
-		ginPrepareDataScan(&btree, reln);
+		ginPrepareDataScan(&btree, reln, split->rootBlkno);
 	}
 
-	stack.blkno = split->leftBlkno;
-	stack.buffer = buffer;
-	stack.off = InvalidOffsetNumber;
-	stack.parent = NULL;
+	stack = palloc(sizeof(GinBtreeStack));
+	stack->blkno = split->leftBlkno;
+	stack->buffer = buffer;
+	stack->off = InvalidOffsetNumber;
+	stack->parent = NULL;
 
-	ginFindParents(&btree, &stack, split->rootBlkno);
+	ginFindParents(&btree, stack);
+	LockBuffer(stack->parent->buffer, GIN_UNLOCK);
+	ginFinishSplit(&btree, stack, NULL);
 
-	btree.prepareDownlink(&btree, buffer);
-	ginInsertValue(&btree, stack.parent, NULL);
+	/* buffer is released by ginFinishSplit */
 
 	FreeFakeRelcacheEntry(reln);
-
-	UnlockReleaseBuffer(buffer);
 }
 
 void

src/include/access/gin_private.h

@@ -485,41 +485,59 @@ typedef struct GinBtreeData
 
 	/* insert methods */
 	OffsetNumber (*findChildPtr) (GinBtree, Page, BlockNumber, OffsetNumber);
-	bool		(*placeToPage) (GinBtree, Buffer, OffsetNumber, XLogRecData **);
-	Page		(*splitPage) (GinBtree, Buffer, Buffer, OffsetNumber, XLogRecData **);
-	void		(*prepareDownlink) (GinBtree, Buffer);
+	bool		(*placeToPage) (GinBtree, Buffer, OffsetNumber, void *, BlockNumber, XLogRecData **);
+	Page		(*splitPage) (GinBtree, Buffer, Buffer, OffsetNumber, void *, BlockNumber, XLogRecData **);
+	void	   *(*prepareDownlink) (GinBtree, Buffer);
 	void		(*fillRoot) (GinBtree, Buffer, Buffer, Buffer);
 
 	bool		isData;
 
 	Relation	index;
+	BlockNumber rootBlkno;
 	GinState   *ginstate;		/* not valid in a data scan */
 	bool		fullScan;
 	bool		isBuild;
 
-	BlockNumber rightblkno;
-
-	/* Entry options */
+	/* Search key for Entry tree */
 	OffsetNumber entryAttnum;
 	Datum		entryKey;
 	GinNullCategory entryCategory;
-	IndexTuple	entry;
-	bool		isDelete;
 
-	/* Data (posting tree) options */
+	/* Search key for data tree (posting tree) */
+	ItemPointerData itemptr;
+} GinBtreeData;
+
+/* This represents a tuple to be inserted to entry tree. */
+typedef struct
+{
+	IndexTuple	entry;			/* tuple to insert */
+	bool		isDelete;		/* delete old tuple at same offset? */
+} GinBtreeEntryInsertData;
+
+/*
+ * This represents an itempointer, or many itempointers, to be inserted to
+ * a data (posting tree) leaf page
+ */
+typedef struct
+{
 	ItemPointerData *items;
 	uint32		nitem;
 	uint32		curitem;
+} GinBtreeDataLeafInsertData;
 
-	PostingItem pitem;
-} GinBtreeData;
+/*
+ * For internal data (posting tree) pages, the insertion payload is a
+ * PostingItem
+ */
 
-extern GinBtreeStack *ginFindLeafPage(GinBtree btree, BlockNumber rootBlkno, bool searchMode);
+extern GinBtreeStack *ginFindLeafPage(GinBtree btree, bool searchMode);
 extern Buffer ginStepRight(Buffer buffer, Relation index, int lockmode);
 extern void freeGinBtreeStack(GinBtreeStack *stack);
 extern void ginInsertValue(GinBtree btree, GinBtreeStack *stack,
+			   void *insertdata, GinStatsData *buildStats);
+extern void ginFindParents(GinBtree btree, GinBtreeStack *stack);
+extern void ginFinishSplit(GinBtree btree, GinBtreeStack *stack,
 			   GinStatsData *buildStats);
-extern void ginFindParents(GinBtree btree, GinBtreeStack *stack, BlockNumber rootBlkno);
 
 /* ginentrypage.c */
 extern IndexTuple GinFormTuple(GinState *ginstate,
@@ -543,7 +561,7 @@ extern void ginInsertItemPointers(Relation index, BlockNumber rootBlkno,
 					  GinStatsData *buildStats);
 extern GinBtreeStack *ginScanBeginPostingTree(Relation index, BlockNumber rootBlkno);
 extern void ginDataFillRoot(GinBtree btree, Buffer root, Buffer lbuf, Buffer rbuf);
-extern void ginPrepareDataScan(GinBtree btree, Relation index);
+extern void ginPrepareDataScan(GinBtree btree, Relation index, BlockNumber rootBlkno);
 
 /* ginscan.c */