WAL

src/backend/access/transam/transsup.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/Attic/transsup.c,v 1.25 2000/01/26 05:56:04 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/Attic/transsup.c,v 1.26 2000/10/28 16:20:53 vadim Exp $
  *
  * NOTES
  *	  This file contains support functions for the high
@@ -186,6 +186,10 @@ TransBlockGetXidStatus(Block tblock,
 	bits8		bit2;
 	BitIndex	offset;
 
+#ifdef XLOG
+	tblock = (Block) ((char*) tblock + sizeof(XLogRecPtr));
+#endif
+
 	/* ----------------
 	 *	calculate the index into the transaction data where
 	 *	our transaction status is located
@@ -227,6 +231,10 @@ TransBlockSetXidStatus(Block tblock,
 	Index		index;
 	BitIndex	offset;
 
+#ifdef XLOG
+	tblock = (Block) ((char*) tblock + sizeof(XLogRecPtr));
+#endif
+
 	/* ----------------
 	 *	calculate the index into the transaction data where
 	 *	we sould store our transaction status.

src/backend/access/transam/varsup.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/varsup.c,v 1.29 2000/07/25 20:18:19 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/varsup.c,v 1.30 2000/10/28 16:20:53 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -125,7 +125,11 @@ VariableRelationPutNextXid(TransactionId xid)
 
 	TransactionIdStore(xid, &(var->nextXidData));
 
+#ifdef XLOG
+	WriteBuffer(buf);	/* temp */
+#else
 	FlushBuffer(buf, TRUE);
+#endif
 }
 
 /* --------------------------------

src/backend/access/transam/xact.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.77 2000/10/24 20:06:39 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.78 2000/10/28 16:20:53 vadim Exp $
  *
  * NOTES
  *		Transaction aborts can now occur two ways:
@@ -176,6 +176,8 @@
 
 extern bool SharedBufferChanged;
 
+void RecordTransactionCommit(void);
+
 static void AbortTransaction(void);
 static void AtAbort_Cache(void);
 static void AtAbort_Locks(void);
@@ -191,7 +193,6 @@ static void AtStart_Memory(void);
 static void CleanupTransaction(void);
 static void CommitTransaction(void);
 static void RecordTransactionAbort(void);
-static void RecordTransactionCommit(void);
 static void StartTransaction(void);
 
 /* ----------------
@@ -220,7 +221,7 @@ int			XactIsoLevel;
 #ifdef XLOG
 #include "access/xlogutils.h"
 
-int			CommitDelay = 100;
+int			CommitDelay = 5;	/* 1/200 sec */
 
 void		xact_redo(XLogRecPtr lsn, XLogRecord *record);
 void		xact_undo(XLogRecPtr lsn, XLogRecord *record);
@@ -658,8 +659,8 @@ AtStart_Memory(void)
  *			  -cim 3/18/90
  * --------------------------------
  */
-static void
-RecordTransactionCommit(void)
+void
+RecordTransactionCommit()
 {
 	TransactionId xid;
 	int			leak;
@@ -683,6 +684,8 @@ RecordTransactionCommit(void)
 		struct timeval	delay;
 		XLogRecPtr		recptr;
 
+		BufmgrCommit();
+
 		xlrec.xtime = time(NULL);
 		/*
 		 * MUST SAVE ARRAY OF RELFILENODE-s TO DROP

src/backend/access/transam/xlog.c

@@ -6,7 +6,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.21 2000/10/24 09:56:09 vadim Exp $
+ * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.22 2000/10/28 16:20:54 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -220,6 +220,8 @@ static uint32 readOff = 0;
 static char readBuf[BLCKSZ];
 static XLogRecord *nextRecord = NULL;
 
+static bool InRedo = false;
+
 XLogRecPtr
 XLogInsert(RmgrId rmid, uint8 info, char *hdr, uint32 hdrlen, char *buf, uint32 buflen)
 {
@@ -481,6 +483,19 @@ XLogFlush(XLogRecPtr record)
 	unsigned	i = 0;
 	bool		force_lgwr = false;
 
+	if (XLOG_DEBUG)
+	{
+		fprintf(stderr, "XLogFlush%s%s: rqst %u/%u; wrt %u/%u; flsh %u/%u\n",
+			(IsBootstrapProcessingMode()) ? "(bootstrap)" : "",
+			(InRedo) ? "(redo)" : "",
+			record.xlogid, record.xrecoff,
+			LgwrResult.Write.xlogid, LgwrResult.Write.xrecoff,
+			LgwrResult.Flush.xlogid, LgwrResult.Flush.xrecoff);
+		fflush(stderr);
+	}
+
+	if (IsBootstrapProcessingMode() || InRedo)
+		return;
 	if (XLByteLE(record, LgwrResult.Flush))
 		return;
 	WriteRqst = LgwrRqst.Write;
@@ -894,7 +909,7 @@ ReadRecord(XLogRecPtr *RecPtr, char *buffer)
 	record = (XLogRecord *) ((char *) readBuf + RecPtr->xrecoff % BLCKSZ);
 
 got_record:;
-	if (record->xl_len == 0 || record->xl_len >
+	if (record->xl_len >
 		(BLCKSZ - RecPtr->xrecoff % BLCKSZ - SizeOfXLogRecord))
 	{
 		elog(emode, "ReadRecord: invalid record len %u in (%u, %u)",
@@ -1259,7 +1274,6 @@ StartupXLOG()
 				LastRec;
 	XLogRecord *record;
 	char		buffer[MAXLOGRECSZ + SizeOfXLogRecord];
-	int			recovery = 0;
 	bool		sie_saved = false;
 
 #endif
@@ -1380,16 +1394,15 @@ StartupXLOG()
 			elog(STOP, "Invalid Redo/Undo record in shutdown checkpoint");
 		if (ControlFile->state == DB_SHUTDOWNED)
 			elog(STOP, "Invalid Redo/Undo record in Shutdowned state");
-		recovery = 1;
+		InRecovery = true;
 	}
 	else if (ControlFile->state != DB_SHUTDOWNED)
 	{
-		if (checkPoint.Shutdown)
-			elog(STOP, "Invalid state in control file");
-		recovery = 1;
+		InRecovery = true;
 	}
 
-	if (recovery)
+	/* REDO */
+	if (InRecovery)
 	{
 		elog(LOG, "The DataBase system was not properly shut down\n"
 			 "\tAutomatic recovery is in progress...");
@@ -1401,6 +1414,7 @@ StartupXLOG()
 		StopIfError = true;
 
 		XLogOpenLogRelation();	/* open pg_log */
+		XLogInitRelationCache();
 
 		/* Is REDO required ? */
 		if (XLByteLT(checkPoint.redo, RecPtr))
@@ -1409,9 +1423,9 @@ StartupXLOG()
 /* read past CheckPoint record */
 			record = ReadRecord(NULL, buffer);
 
-		/* REDO */
 		if (record->xl_len != 0)
 		{
+			InRedo = true;
 			elog(LOG, "Redo starts at (%u, %u)",
 				 ReadRecPtr.xlogid, ReadRecPtr.xrecoff);
 			do
@@ -1441,12 +1455,40 @@ StartupXLOG()
 			elog(LOG, "Redo done at (%u, %u)",
 				 ReadRecPtr.xlogid, ReadRecPtr.xrecoff);
 			LastRec = ReadRecPtr;
+			InRedo = false;
 		}
 		else
 			elog(LOG, "Redo is not required");
+	}
+
+	/* Init xlog buffer cache */
+	record = ReadRecord(&LastRec, buffer);
+	logId = EndRecPtr.xlogid;
+	logSeg = (EndRecPtr.xrecoff - 1) / XLogSegSize;
+	logOff = 0;
+	logFile = XLogFileOpen(logId, logSeg, false);
+	XLogCtl->xlblocks[0].xlogid = logId;
+	XLogCtl->xlblocks[0].xrecoff =
+		((EndRecPtr.xrecoff - 1) / BLCKSZ + 1) * BLCKSZ;
+	Insert = &XLogCtl->Insert;
+	memcpy((char *) (Insert->currpage), readBuf, BLCKSZ);
+	Insert->currpos = ((char *) Insert->currpage) +
+		(EndRecPtr.xrecoff + BLCKSZ - XLogCtl->xlblocks[0].xrecoff);
+	Insert->PrevRecord = LastRec;
+
+	LgwrRqst.Write = LgwrRqst.Flush =
+	LgwrResult.Write = LgwrResult.Flush = EndRecPtr;
+
+	XLogCtl->Write.LgwrResult = LgwrResult;
+	Insert->LgwrResult = LgwrResult;
+
+	XLogCtl->LgwrRqst = LgwrRqst;
+	XLogCtl->LgwrResult = LgwrResult;
 
 #ifdef NOT_USED
 	/* UNDO */
+	if (InRecovery)
+	{
 		RecPtr = ReadRecPtr;
 		if (XLByteLT(checkPoint.undo, RecPtr))
 		{
@@ -1465,29 +1507,16 @@ StartupXLOG()
 		}
 		else
 			elog(LOG, "Undo is not required");
-#endif
 	}
+#endif
 
-	/* Init xlog buffer cache */
-	record = ReadRecord(&LastRec, buffer);
-	logId = EndRecPtr.xlogid;
-	logSeg = (EndRecPtr.xrecoff - 1) / XLogSegSize;
-	logOff = 0;
-	logFile = XLogFileOpen(logId, logSeg, false);
-	XLogCtl->xlblocks[0].xlogid = logId;
-	XLogCtl->xlblocks[0].xrecoff =
-		((EndRecPtr.xrecoff - 1) / BLCKSZ + 1) * BLCKSZ;
-	Insert = &XLogCtl->Insert;
-	memcpy((char *) (Insert->currpage), readBuf, BLCKSZ);
-	Insert->currpos = ((char *) Insert->currpage) +
-		(EndRecPtr.xrecoff + BLCKSZ - XLogCtl->xlblocks[0].xrecoff);
-	Insert->PrevRecord = ControlFile->checkPoint;
-
-	if (recovery)
+	if (InRecovery)
 	{
 		CreateCheckPoint(true);
 		StopIfError = sie_saved;
+		XLogCloseRelationCache();
 	}
+	InRecovery = false;
 
 #endif	 /* XLOG */
 

src/backend/access/transam/xlogutils.c

@@ -22,6 +22,7 @@
 #include "access/htup.h"
 #include "access/xlogutils.h"
 #include "catalog/pg_database.h"
+#include "lib/hasht.h"
 
 /*
  * ---------------------------------------------------------------
@@ -240,28 +241,6 @@ static int					_xlcnt = 0;
 #define	_XLOG_INITRELCACHESIZE	32
 #define	_XLOG_MAXRELCACHESIZE	512
 
-void
-XLogCloseRelationCache(void)
-{
-	int i;
-
-	if (!_xlrelarr)
-		return;
-
-	for (i = 1; i < _xlast; i++)
-	{
-		Relation	reln = &(_xlrelarr[i].reldata);
-		if (reln->rd_fd >= 0)
-			smgrclose(DEFAULT_SMGR, reln);
-	}
-
-	free(_xlrelarr);
-	free(_xlpgcarr);
-
-	hash_destroy(_xlrelcache);
-	_xlrelarr = NULL;
-}
-
 static void
 _xl_init_rel_cache(void)
 {
@@ -286,6 +265,35 @@ _xl_init_rel_cache(void)
 								HASH_ELEM | HASH_FUNCTION);
 }
 
+static void
+_xl_remove_hash_entry(XLogRelDesc **edata, int dummy)
+{
+	XLogRelCacheEntry	   *hentry;
+	bool					found;
+	XLogRelDesc			   *rdesc = *edata;
+	Form_pg_class			tpgc = rdesc->reldata.rd_rel;
+
+	rdesc->lessRecently->moreRecently = rdesc->moreRecently;
+	rdesc->moreRecently->lessRecently = rdesc->lessRecently;
+
+	hentry = (XLogRelCacheEntry*) hash_search(_xlrelcache, 
+		(char*)&(rdesc->reldata.rd_node), HASH_REMOVE, &found);
+
+	if (hentry == NULL)
+		elog(STOP, "_xl_remove_hash_entry: can't delete from cache");
+	if (!found)
+		elog(STOP, "_xl_remove_hash_entry: file was not found in cache");
+
+	if (rdesc->reldata.rd_fd >= 0)
+		smgrclose(DEFAULT_SMGR, &(rdesc->reldata));
+
+	memset(rdesc, 0, sizeof(XLogRelDesc));
+	memset(tpgc, 0, sizeof(FormData_pg_class));
+	rdesc->reldata.rd_rel = tpgc;
+
+	return;
+}
+
 static XLogRelDesc*
 _xl_new_reldesc(void)
 {
@@ -310,34 +318,43 @@ _xl_new_reldesc(void)
 	}
 	else /* reuse */
 	{
-		XLogRelCacheEntry	   *hentry;
-		bool					found;
 		XLogRelDesc	   *res = _xlrelarr[0].moreRecently;
-		Form_pg_class			tpgc = res->reldata.rd_rel;
 
-		res->lessRecently->moreRecently = res->moreRecently;
-		res->moreRecently->lessRecently = res->lessRecently;
-
-		hentry = (XLogRelCacheEntry*) hash_search(_xlrelcache, 
-			(char*)&(res->reldata.rd_node), HASH_REMOVE, &found);
-
-		if (hentry == NULL)
-			elog(STOP, "XLogOpenRelation: can't delete from cache");
-		if (!found)
-			elog(STOP, "XLogOpenRelation: file was not found in cache");
-
-		if (res->reldata.rd_fd >= 0)
-			smgrclose(DEFAULT_SMGR, &(res->reldata));
-
-		memset(res, 0, sizeof(XLogRelDesc));
-		memset(tpgc, 0, sizeof(FormData_pg_class));
-		res->reldata.rd_rel = tpgc;
+		_xl_remove_hash_entry(&res, 0);
 
 		_xlast--;
 		return(res);
 	}
 }
 
+extern void CreateDummyCaches(void);
+extern void DestroyDummyCaches(void);
+
+void
+XLogInitRelationCache(void)
+{
+	CreateDummyCaches();
+	_xl_init_rel_cache();
+}
+
+void
+XLogCloseRelationCache(void)
+{
+
+	DestroyDummyCaches();
+
+	if (!_xlrelarr)
+		return;
+
+	HashTableWalk(_xlrelcache, (HashtFunc)_xl_remove_hash_entry, 0);
+	hash_destroy(_xlrelcache);
+
+	free(_xlrelarr);
+	free(_xlpgcarr);
+
+	_xlrelarr = NULL;
+}
+
 Relation
 XLogOpenRelation(bool redo, RmgrId rmid, RelFileNode rnode)
 {
@@ -345,9 +362,6 @@ XLogOpenRelation(bool redo, RmgrId rmid, RelFileNode rnode)
 	XLogRelCacheEntry	   *hentry;
 	bool					found;
 
-	if (!_xlrelarr)
-		_xl_init_rel_cache();
-
 	hentry = (XLogRelCacheEntry*) 
 			hash_search(_xlrelcache, (char*)&rnode, HASH_FIND, &found);
 

src/backend/commands/dbcommands.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/commands/dbcommands.c,v 1.62 2000/10/22 17:55:36 pjw Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/commands/dbcommands.c,v 1.63 2000/10/28 16:20:54 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -193,6 +193,9 @@ createdb(const char *dbname, const char *dbpath, int encoding)
 			elog(ERROR, "CREATE DATABASE: Could not initialize database directory. Delete failed as well");
 	}
 
+#ifdef XLOG
+	BufferSync();
+#endif
 }
 
 

src/backend/commands/vacuum.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.170 2000/10/24 09:56:15 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/commands/vacuum.c,v 1.171 2000/10/28 16:20:54 vadim Exp $
  *
 
  *-------------------------------------------------------------------------
@@ -1787,7 +1787,9 @@ failed to add item with len = %u to page %u (free space %u, nusd %u, noff %u)",
 
 	if (num_moved > 0)
 	{
-
+#ifdef XLOG
+		RecordTransactionCommit();
+#else
 		/*
 		 * We have to commit our tuple' movings before we'll truncate
 		 * relation, but we shouldn't lose our locks. And so - quick hack:
@@ -1797,6 +1799,7 @@ failed to add item with len = %u to page %u (free space %u, nusd %u, noff %u)",
 		FlushBufferPool();
 		TransactionIdCommit(myXID);
 		FlushBufferPool();
+#endif
 	}
 
 	/*

src/backend/storage/buffer/bufmgr.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.91 2000/10/23 04:10:06 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.92 2000/10/28 16:20:55 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -42,6 +42,13 @@
  *		freelist.c -- chooses victim for buffer replacement
  *		buf_table.c -- manages the buffer lookup table
  */
+
+#ifdef XLOG
+
+#include "xlog_bufmgr.c"
+
+#else
+
 #include <sys/types.h>
 #include <sys/file.h>
 #include <math.h>
@@ -2512,3 +2519,5 @@ MarkBufferForCleanup(Buffer buffer, void (*CleanupFunc)(Buffer))
 	SpinRelease(BufMgrLock);
 	return;
 }
+
+#endif	/* ! XLOG */

src/backend/storage/buffer/localbuf.c

@@ -16,10 +16,17 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/localbuf.c,v 1.32 2000/10/23 04:10:06 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/localbuf.c,v 1.33 2000/10/28 16:20:56 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
+   
+#ifdef XLOG
+
+#include "xlog_localbuf.c"
+
+#else
+
 #include <sys/types.h>
 #include <sys/file.h>
 #include <math.h>
@@ -247,10 +254,11 @@ InitLocalBuffer(void)
 }
 
 /*
- * LocalBufferSync -
- *	  flush all dirty buffers in the local buffer cache. Since the buffer
- *	  cache is only used for keeping relations visible during a transaction,
- *	  we will not need these buffers again.
+ * LocalBufferSync
+ *
+ * Flush all dirty buffers in the local buffer cache at commit time.
+ * Since the buffer cache is only used for keeping relations visible
+ * during a transaction, we will not need these buffers again.
  */
 void
 LocalBufferSync(void)
@@ -303,3 +311,5 @@ ResetLocalBufferPool(void)
 	MemSet(LocalRefCount, 0, sizeof(long) * NLocBuffer);
 	nextFreeLocalBuf = 0;
 }
+
+#endif	/* XLOG */

src/backend/storage/buffer/xlog_localbuf.c

@@ -0,0 +1,274 @@
+/*-------------------------------------------------------------------------
+ *
+ * localbuf.c
+ *	  local buffer manager. Fast buffer manager for temporary tables
+ *	  or special cases when the operation is not visible to other backends.
+ *
+ *	  When a relation is being created, the descriptor will have rd_islocal
+ *	  set to indicate that the local buffer manager should be used. During
+ *	  the same transaction the relation is being created, any inserts or
+ *	  selects from the newly created relation will use the local buffer
+ *	  pool. rd_islocal is reset at the end of a transaction (commit/abort).
+ *	  This is useful for queries like SELECT INTO TABLE and create index.
+ *
+ * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
+ * Portions Copyright (c) 1994-5, Regents of the University of California
+ *
+ *
+ * IDENTIFICATION
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/Attic/xlog_localbuf.c,v 1.1 2000/10/28 16:20:56 vadim Exp $
+ *
+ *-------------------------------------------------------------------------
+ */
+#include <sys/types.h>
+#include <sys/file.h>
+#include <math.h>
+#include <signal.h>
+
+#include "postgres.h"
+
+#include "executor/execdebug.h"
+#include "storage/smgr.h"
+#include "utils/relcache.h"
+
+extern long int LocalBufferFlushCount;
+
+int			NLocBuffer = 64;
+BufferDesc *LocalBufferDescriptors = NULL;
+long	   *LocalRefCount = NULL;
+
+static int	nextFreeLocalBuf = 0;
+
+/*#define LBDEBUG*/
+
+/*
+ * LocalBufferAlloc -
+ *	  allocate a local buffer. We do round robin allocation for now.
+ */
+BufferDesc *
+LocalBufferAlloc(Relation reln, BlockNumber blockNum, bool *foundPtr)
+{
+	int			i;
+	BufferDesc *bufHdr = (BufferDesc *) NULL;
+
+	if (blockNum == P_NEW)
+	{
+		blockNum = reln->rd_nblocks;
+		reln->rd_nblocks++;
+	}
+
+	/* a low tech search for now -- not optimized for scans */
+	for (i = 0; i < NLocBuffer; i++)
+	{
+		if (LocalBufferDescriptors[i].tag.rnode.relNode == 
+			reln->rd_node.relNode &&
+			LocalBufferDescriptors[i].tag.blockNum == blockNum)
+		{
+
+#ifdef LBDEBUG
+			fprintf(stderr, "LB ALLOC (%u,%d) %d\n",
+					RelationGetRelid(reln), blockNum, -i - 1);
+#endif
+			LocalRefCount[i]++;
+			*foundPtr = TRUE;
+			return &LocalBufferDescriptors[i];
+		}
+	}
+
+#ifdef LBDEBUG
+	fprintf(stderr, "LB ALLOC (%u,%d) %d\n",
+			RelationGetRelid(reln), blockNum, -nextFreeLocalBuf - 1);
+#endif
+
+	/* need to get a new buffer (round robin for now) */
+	for (i = 0; i < NLocBuffer; i++)
+	{
+		int			b = (nextFreeLocalBuf + i) % NLocBuffer;
+
+		if (LocalRefCount[b] == 0)
+		{
+			bufHdr = &LocalBufferDescriptors[b];
+			LocalRefCount[b]++;
+			nextFreeLocalBuf = (b + 1) % NLocBuffer;
+			break;
+		}
+	}
+	if (bufHdr == NULL)
+		elog(ERROR, "no empty local buffer.");
+
+	/*
+	 * this buffer is not referenced but it might still be dirty (the last
+	 * transaction to touch it doesn't need its contents but has not
+	 * flushed it).  if that's the case, write it out before reusing it!
+	 */
+	if (bufHdr->flags & BM_DIRTY || bufHdr->cntxDirty)
+	{
+		Relation	bufrel = RelationNodeCacheGetRelation(bufHdr->tag.rnode);
+
+		Assert(bufrel != NULL);
+
+		/* flush this page */
+		smgrwrite(DEFAULT_SMGR, bufrel, bufHdr->tag.blockNum,
+				  (char *) MAKE_PTR(bufHdr->data));
+		LocalBufferFlushCount++;
+
+		/*
+		 * drop relcache refcount incremented by
+		 * RelationIdCacheGetRelation
+		 */
+		RelationDecrementReferenceCount(bufrel);
+	}
+
+	/*
+	 * it's all ours now.
+	 *
+	 * We need not in tblNode currently but will in future I think,
+	 * when we'll give up rel->rd_fd to fmgr cache.
+	 */
+	bufHdr->tag.rnode = reln->rd_node;
+	bufHdr->tag.blockNum = blockNum;
+	bufHdr->flags &= ~BM_DIRTY;
+	bufHdr->cntxDirty = false;
+
+	/*
+	 * lazy memory allocation. (see MAKE_PTR for why we need to do
+	 * MAKE_OFFSET.)
+	 */
+	if (bufHdr->data == (SHMEM_OFFSET) 0)
+	{
+		char	   *data = (char *) malloc(BLCKSZ);
+
+		bufHdr->data = MAKE_OFFSET(data);
+	}
+
+	*foundPtr = FALSE;
+	return bufHdr;
+}
+
+/*
+ * WriteLocalBuffer -
+ *	  writes out a local buffer
+ */
+int
+WriteLocalBuffer(Buffer buffer, bool release)
+{
+	int			bufid;
+
+	Assert(BufferIsLocal(buffer));
+
+#ifdef LBDEBUG
+	fprintf(stderr, "LB WRITE %d\n", buffer);
+#endif
+
+	bufid = -(buffer + 1);
+	LocalBufferDescriptors[bufid].flags |= BM_DIRTY;
+
+	if (release)
+	{
+		Assert(LocalRefCount[bufid] > 0);
+		LocalRefCount[bufid]--;
+	}
+
+	return true;
+}
+
+/*
+ * InitLocalBuffer -
+ *	  init the local buffer cache. Since most queries (esp. multi-user ones)
+ *	  don't involve local buffers, we delay allocating memory for actual the
+ *	  buffer until we need it.
+ */
+void
+InitLocalBuffer(void)
+{
+	int			i;
+
+	/*
+	 * these aren't going away. I'm not gonna use palloc.
+	 */
+	LocalBufferDescriptors =
+		(BufferDesc *) malloc(sizeof(BufferDesc) * NLocBuffer);
+	MemSet(LocalBufferDescriptors, 0, sizeof(BufferDesc) * NLocBuffer);
+	nextFreeLocalBuf = 0;
+
+	for (i = 0; i < NLocBuffer; i++)
+	{
+		BufferDesc *buf = &LocalBufferDescriptors[i];
+
+		/*
+		 * negative to indicate local buffer. This is tricky: shared
+		 * buffers start with 0. We have to start with -2. (Note that the
+		 * routine BufferDescriptorGetBuffer adds 1 to buf_id so our first
+		 * buffer id is -1.)
+		 */
+		buf->buf_id = -i - 2;
+	}
+
+	LocalRefCount = (long *) malloc(sizeof(long) * NLocBuffer);
+	MemSet(LocalRefCount, 0, sizeof(long) * NLocBuffer);
+}
+
+/*
+ * LocalBufferSync
+ *
+ * Flush all dirty buffers in the local buffer cache at commit time.
+ * Since the buffer cache is only used for keeping relations visible
+ * during a transaction, we will not need these buffers again.
+ *
+ * Note that we have to *flush* local buffers because of them are not
+ * visible to checkpoint makers. But we can skip XLOG flush check.
+ */
+void
+LocalBufferSync(void)
+{
+	int			i;
+
+	for (i = 0; i < NLocBuffer; i++)
+	{
+		BufferDesc *buf = &LocalBufferDescriptors[i];
+		Relation	bufrel;
+
+		if (buf->flags & BM_DIRTY || buf->cntxDirty)
+		{
+#ifdef LBDEBUG
+			fprintf(stderr, "LB SYNC %d\n", -i - 1);
+#endif
+			bufrel = RelationNodeCacheGetRelation(buf->tag.rnode);
+
+			Assert(bufrel != NULL);
+
+			smgrwrite(DEFAULT_SMGR, bufrel, buf->tag.blockNum,
+						(char *) MAKE_PTR(buf->data));
+			smgrmarkdirty(DEFAULT_SMGR, bufrel, buf->tag.blockNum);
+			LocalBufferFlushCount++;
+
+			/* drop relcache refcount from RelationIdCacheGetRelation */
+			RelationDecrementReferenceCount(bufrel);
+
+			buf->flags &= ~BM_DIRTY;
+			buf->cntxDirty = false;
+		}
+	}
+
+	MemSet(LocalRefCount, 0, sizeof(long) * NLocBuffer);
+	nextFreeLocalBuf = 0;
+}
+
+void
+ResetLocalBufferPool(void)
+{
+	int			i;
+
+	for (i = 0; i < NLocBuffer; i++)
+	{
+		BufferDesc *buf = &LocalBufferDescriptors[i];
+
+		buf->tag.rnode.relNode = InvalidOid;
+		buf->flags &= ~BM_DIRTY;
+		buf->cntxDirty = false;
+		buf->buf_id = -i - 2;
+	}
+
+	MemSet(LocalRefCount, 0, sizeof(long) * NLocBuffer);
+	nextFreeLocalBuf = 0;
+}

src/backend/storage/file/fd.c

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/file/fd.c,v 1.64 2000/10/02 19:42:47 petere Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/file/fd.c,v 1.65 2000/10/28 16:20:56 vadim Exp $
  *
  * NOTES:
  *
@@ -823,8 +823,10 @@ FileWrite(File file, char *buffer, int amount)
 	if (returnCode > 0)
 	{
 		VfdCache[file].seekPos += returnCode;
+#ifndef XLOG
 		/* mark the file as needing fsync */
 		VfdCache[file].fdstate |= FD_DIRTY;
+#endif
 	}
 	else
 		VfdCache[file].seekPos = FileUnknownPos;

src/backend/storage/smgr/md.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.76 2000/10/20 11:01:11 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.77 2000/10/28 16:20:57 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -76,12 +76,7 @@ static int	_mdfd_getrelnfd(Relation reln);
 static MdfdVec *_mdfd_openseg(Relation reln, int segno, int oflags);
 static MdfdVec *_mdfd_getseg(Relation reln, int blkno);
 
-#ifdef OLD_FILE_NAMING
-static int _mdfd_blind_getseg(char *dbname, char *relname,
-				   Oid dbid, Oid relid, int blkno);
-#else
 static int _mdfd_blind_getseg(RelFileNode rnode, int blkno);
-#endif
 
 static int	_fdvec_alloc(void);
 static void _fdvec_free(int);
@@ -134,11 +129,7 @@ mdcreate(Relation reln)
 
 	Assert(reln->rd_unlinked && reln->rd_fd < 0);
 
-#ifdef OLD_FILE_NAMING
-	path = relpath(RelationGetPhysicalRelationName(reln));
-#else
 	path = relpath(reln->rd_node);
-#endif
 	fd = FileNameOpenFile(path, O_RDWR | O_CREAT | O_EXCL | PG_BINARY, 0600);
 
 	/*
@@ -336,11 +327,7 @@ mdopen(Relation reln)
 	int			vfd;
 
 	Assert(reln->rd_fd < 0);
-#ifdef OLD_FILE_NAMING
-	path = relpath(RelationGetPhysicalRelationName(reln));
-#else
 	path = relpath(reln->rd_node);
-#endif
 
 	fd = FileNameOpenFile(path, O_RDWR | PG_BINARY, 0600);
 	if (fd < 0)
@@ -579,30 +566,16 @@ mdflush(Relation reln, BlockNumber blocknum, char *buffer)
  *		the file, making it more like mdflush().
  */
 int
-#ifdef OLD_FILE_NAMING
-mdblindwrt(char *dbname,
-		   char *relname,
-		   Oid dbid,
-		   Oid relid,
-		   BlockNumber blkno,
-		   char *buffer,
-		   bool dofsync)
-#else
 mdblindwrt(RelFileNode rnode,
 		   BlockNumber blkno,
 		   char *buffer,
 		   bool dofsync)
-#endif
 {
 	int			status;
 	long		seekpos;
 	int			fd;
 
-#ifdef OLD_FILE_NAMING
-	fd = _mdfd_blind_getseg(dbname, relname, dbid, relid, blkno);
-#else
 	fd = _mdfd_blind_getseg(rnode, blkno);
-#endif
 
 	if (fd < 0)
 		return SM_FAIL;
@@ -676,25 +649,13 @@ mdmarkdirty(Relation reln, BlockNumber blkno)
  *		rather than building md/fd datastructures to postpone it till later.
  */
 int
-#ifdef OLD_FILE_NAMING
-mdblindmarkdirty(char *dbname,
-				 char *relname,
-				 Oid dbid,
-				 Oid relid,
-				 BlockNumber blkno)
-#else
 mdblindmarkdirty(RelFileNode rnode,
 				 BlockNumber blkno)
-#endif
 {
 	int			status;
 	int			fd;
 
-#ifdef OLD_FILE_NAMING
-	fd = _mdfd_blind_getseg(dbname, relname, dbid, relid, blkno);
-#else
 	fd = _mdfd_blind_getseg(rnode, blkno);
-#endif
 
 	if (fd < 0)
 		return SM_FAIL;
@@ -915,6 +876,22 @@ mdabort()
 	return SM_SUCCESS;
 }
 
+#ifdef XLOG
+/*
+ *	mdsync() -- Sync storage.
+ *
+ */
+int
+mdsync()
+{
+	sync();
+	if (IsUnderPostmaster)
+		sleep(2);
+	sync();
+	return SM_SUCCESS;
+}
+#endif
+
 /*
  *	_fdvec_alloc () -- grab a free (or new) md file descriptor vector.
  *
@@ -996,11 +973,7 @@ _mdfd_openseg(Relation reln, int segno, int oflags)
 			   *fullpath;
 
 	/* be sure we have enough space for the '.segno', if any */
-#ifdef OLD_FILE_NAMING
-	path = relpath(RelationGetPhysicalRelationName(reln));
-#else
 	path = relpath(reln->rd_node);
-#endif
 
 	if (segno > 0)
 	{
@@ -1115,12 +1088,7 @@ _mdfd_getseg(Relation reln, int blkno)
  */
 
 static int
-#ifdef OLD_FILE_NAMING
-_mdfd_blind_getseg(char *dbname, char *relname, Oid dbid, Oid relid,
-				   int blkno)
-#else
 _mdfd_blind_getseg(RelFileNode rnode, int blkno)
-#endif
 {
 	char	   *path;
 	int			fd;
@@ -1130,12 +1098,7 @@ _mdfd_blind_getseg(RelFileNode rnode, int blkno)
 
 #endif
 
-#ifdef OLD_FILE_NAMING
-	/* construct the path to the relation */
-	path = relpath_blind(dbname, relname, dbid, relid);
-#else
 	path = relpath(rnode);
-#endif
 
 #ifndef LET_OS_MANAGE_FILESIZE
 	/* append the '.segno', if needed */

src/backend/storage/smgr/smgr.c

@@ -11,7 +11,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/smgr.c,v 1.41 2000/10/21 15:43:31 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/smgr.c,v 1.42 2000/10/28 16:20:57 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -36,27 +36,17 @@ typedef struct f_smgr
 										   char *buffer);
 	int			(*smgr_flush) (Relation reln, BlockNumber blocknum,
 										   char *buffer);
-#ifdef OLD_FILE_NAMING
-	int			(*smgr_blindwrt) (char *dbname, char *relname,
-											  Oid dbid, Oid relid,
-										 BlockNumber blkno, char *buffer,
-											  bool dofsync);
-#else
 	int			(*smgr_blindwrt) (RelFileNode rnode, BlockNumber blkno, 
 										char *buffer, bool dofsync);
-#endif
 	int			(*smgr_markdirty) (Relation reln, BlockNumber blkno);
-#ifdef OLD_FILE_NAMING
-	int			(*smgr_blindmarkdirty) (char *dbname, char *relname,
-													Oid dbid, Oid relid,
-													BlockNumber blkno);
-#else
 	int			(*smgr_blindmarkdirty) (RelFileNode, BlockNumber blkno);
-#endif
 	int			(*smgr_nblocks) (Relation reln);
 	int			(*smgr_truncate) (Relation reln, int nblocks);
 	int			(*smgr_commit) (void);	/* may be NULL */
 	int			(*smgr_abort) (void);	/* may be NULL */
+#ifdef XLOG
+	int			(*smgr_sync) (void);
+#endif
 } f_smgr;
 
 /*
@@ -69,7 +59,11 @@ static f_smgr smgrsw[] = {
 	/* magnetic disk */
 	{mdinit, NULL, mdcreate, mdunlink, mdextend, mdopen, mdclose,
 		mdread, mdwrite, mdflush, mdblindwrt, mdmarkdirty, mdblindmarkdirty,
+#ifdef XLOG
+	mdnblocks, mdtruncate, mdcommit, mdabort, mdsync},
+#else
 	mdnblocks, mdtruncate, mdcommit, mdabort},
+#endif
 
 #ifdef STABLE_MEMORY_STORAGE
 	/* main memory */
@@ -310,40 +304,6 @@ smgrflush(int16 which, Relation reln, BlockNumber blocknum, char *buffer)
  *		this page down to stable storage in this circumstance.	The
  *		write should be synchronous if dofsync is true.
  */
-#ifdef OLD_FILE_NAMING
-int
-smgrblindwrt(int16 which,
-			 char *dbname,
-			 char *relname,
-			 Oid dbid,
-			 Oid relid,
-			 BlockNumber blkno,
-			 char *buffer,
-			 bool dofsync)
-{
-	char	   *dbstr;
-	char	   *relstr;
-	int			status;
-
-	/* strdup here is probably redundant */
-	dbstr = pstrdup(dbname);
-	relstr = pstrdup(relname);
-
-	status = (*(smgrsw[which].smgr_blindwrt)) (dbstr, relstr, dbid, relid,
-											   blkno, buffer, dofsync);
-
-	if (status == SM_FAIL)
-		elog(ERROR, "cannot write block %d of %s [%s] blind: %m",
-			 blkno, relstr, dbstr);
-
-	pfree(dbstr);
-	pfree(relstr);
-
-	return status;
-}
-
-#else
-
 int
 smgrblindwrt(int16 which,
 			 RelFileNode rnode,
@@ -361,7 +321,6 @@ smgrblindwrt(int16 which,
 
 	return status;
 }
-#endif
 
 /*
  *	smgrmarkdirty() -- Mark a page dirty (needs fsync).
@@ -394,39 +353,6 @@ smgrmarkdirty(int16 which,
  *
  *		Just like smgrmarkdirty, except we don't have a reldesc.
  */
-#ifdef OLD_FILE_NAMING
-int
-smgrblindmarkdirty(int16 which,
-				   char *dbname,
-				   char *relname,
-				   Oid dbid,
-				   Oid relid,
-				   BlockNumber blkno)
-{
-	char	   *dbstr;
-	char	   *relstr;
-	int			status;
-
-	/* strdup here is probably redundant */
-	dbstr = pstrdup(dbname);
-	relstr = pstrdup(relname);
-
-	status = (*(smgrsw[which].smgr_blindmarkdirty)) (dbstr, relstr,
-													 dbid, relid,
-													 blkno);
-
-	if (status == SM_FAIL)
-		elog(ERROR, "cannot mark block %d of %s [%s] blind: %m",
-			 blkno, relstr, dbstr);
-
-	pfree(dbstr);
-	pfree(relstr);
-
-	return status;
-}
-
-#else
-
 int
 smgrblindmarkdirty(int16 which,
 				   RelFileNode rnode,
@@ -442,7 +368,6 @@ smgrblindmarkdirty(int16 which,
 
 	return status;
 }
-#endif
 
 /*
  *	smgrnblocks() -- Calculate the number of POSTGRES blocks in the
@@ -528,6 +453,27 @@ smgrabort()
 	return SM_SUCCESS;
 }
 
+#ifdef XLOG
+int
+smgrsync()
+{
+	int			i;
+
+	for (i = 0; i < NSmgr; i++)
+	{
+		if (smgrsw[i].smgr_sync)
+		{
+			if ((*(smgrsw[i].smgr_sync)) () == SM_FAIL)
+				elog(STOP, "storage sync failed on %s: %m",
+					 DatumGetCString(DirectFunctionCall1(smgrout,
+														 Int16GetDatum(i))));
+		}
+	}
+
+	return SM_SUCCESS;
+}
+#endif
+
 #ifdef NOT_USED
 bool
 smgriswo(int16 smgrno)

src/backend/utils/cache/relcache.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/relcache.c,v 1.113 2000/10/23 04:10:08 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/cache/relcache.c,v 1.114 2000/10/28 16:20:57 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -2064,7 +2064,62 @@ RelationCacheInitializePhase2(void)
 	}
 }
 
+#ifdef XLOG		/* used by XLogInitCache */
 
+void CreateDummyCaches(void);
+void DestroyDummyCaches(void);
+
+void
+CreateDummyCaches(void)
+{
+	MemoryContext	oldcxt;
+	HASHCTL			ctl;
+
+	if (!CacheMemoryContext)
+		CreateCacheMemoryContext();
+
+	oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
+
+	MemSet(&ctl, 0, (int) sizeof(ctl));
+	ctl.keysize = sizeof(NameData);
+	ctl.datasize = sizeof(Relation);
+	RelationNameCache = hash_create(INITRELCACHESIZE, &ctl, HASH_ELEM);
+
+	ctl.keysize = sizeof(Oid);
+	ctl.hash = tag_hash;
+	RelationIdCache = hash_create(INITRELCACHESIZE, &ctl,
+								  HASH_ELEM | HASH_FUNCTION);
+
+	ctl.keysize = sizeof(RelFileNode);
+	ctl.hash = tag_hash;
+	RelationNodeCache = hash_create(INITRELCACHESIZE, &ctl,
+								  HASH_ELEM | HASH_FUNCTION);
+	MemoryContextSwitchTo(oldcxt);
+}
+
+void
+DestroyDummyCaches(void)
+{
+	MemoryContext	oldcxt;
+
+	if (!CacheMemoryContext)
+		return;
+
+	oldcxt = MemoryContextSwitchTo(CacheMemoryContext);
+
+	if (RelationNameCache)
+		hash_destroy(RelationNameCache);
+	if (RelationIdCache)
+		hash_destroy(RelationIdCache);
+	if (RelationNodeCache)
+		hash_destroy(RelationNodeCache);
+
+	RelationNameCache = RelationIdCache = RelationNodeCache = NULL;
+
+	MemoryContextSwitchTo(oldcxt);
+}
+
+#endif	/* XLOG */
 
 static void
 AttrDefaultFetch(Relation relation)

src/backend/utils/init/postinit.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/utils/init/postinit.c,v 1.68 2000/10/16 14:52:15 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/utils/init/postinit.c,v 1.69 2000/10/28 16:20:58 vadim Exp $
  *
  *
  *-------------------------------------------------------------------------
@@ -231,9 +231,6 @@ InitPostgres(const char *dbname, const char *username)
 {
 	bool		bootstrap = IsBootstrapProcessingMode();
 
-	/* initialize the local buffer manager */
-	InitLocalBuffer();
-
 #ifndef XLOG
 	if (!TransactionFlushEnabled())
 		on_shmem_exit(FlushBufferPool, 0);
@@ -414,4 +411,8 @@ BaseInit(void)
 	smgrinit();
 
 	EnablePortalManager();		/* memory for portal/transaction stuff */
+
+	/* initialize the local buffer manager */
+	InitLocalBuffer();
+
 }

src/include/access/transam.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: transam.h,v 1.24 2000/01/26 05:57:51 momjian Exp $
+ * $Id: transam.h,v 1.25 2000/10/28 16:20:59 vadim Exp $
  *
  *	 NOTES
  *		Transaction System Version 101 now support proper oid
@@ -67,7 +67,11 @@ typedef unsigned char XidStatus;/* (2 bits) */
  *		transaction page definitions
  * ----------------
  */
+#ifdef XLOG
+#define TP_DataSize				(BLCKSZ - sizeof(XLogRecPtr))
+#else
 #define TP_DataSize				BLCKSZ
+#endif
 #define TP_NumXidStatusPerBlock (TP_DataSize * 4)
 
 /* ----------------
@@ -84,6 +88,10 @@ typedef unsigned char XidStatus;/* (2 bits) */
  */
 typedef struct LogRelationContentsData
 {
+#ifdef XLOG
+	XLogRecPtr	LSN;		/* temp hack: LSN is member of any block */
+							/* so should be described in bufmgr */
+#endif
 	int			TransSystemVersion;
 } LogRelationContentsData;
 
@@ -107,6 +115,9 @@ typedef LogRelationContentsData *LogRelationContents;
  */
 typedef struct VariableRelationContentsData
 {
+#ifdef XLOG
+	XLogRecPtr	LSN;
+#endif
 	int			TransSystemVersion;
 	TransactionId nextXidData;
 	TransactionId lastXidData;	/* unused */

src/include/access/xact.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: xact.h,v 1.28 2000/10/20 11:01:14 vadim Exp $
+ * $Id: xact.h,v 1.29 2000/10/28 16:20:59 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -135,6 +135,8 @@ extern bool IsTransactionBlock(void);
 extern void UserAbortTransactionBlock(void);
 extern void AbortOutOfAnyTransaction(void);
 
+extern void RecordTransactionCommit(void);
+
 extern TransactionId DisabledTransactionId;
 
 extern void XactPushRollback(void (*func) (void *), void* data);

src/include/access/xlog.h

@@ -10,12 +10,7 @@
 
 #include "access/rmgr.h"
 #include "access/transam.h"
-
-typedef struct XLogRecPtr
-{
-	uint32		xlogid;			/* log file #, 0 based */
-	uint32		xrecoff;		/* offset of record in log file */
-} XLogRecPtr;
+#include "access/xlogdefs.h"
 
 typedef struct XLogRecord
 {
@@ -83,12 +78,7 @@ typedef XLogPageHeaderData *XLogPageHeader;
 #define XLByteEQ(left, right)		\
 			(right.xlogid == left.xlogid && right.xrecoff ==  left.xrecoff)
 
-/*
- * StartUpID (SUI) - system startups counter.
- * It's to allow removing pg_log after shutdown.
- */
-typedef	uint32		StartUpID;
-extern	StartUpID	ThisStartUpID;
+extern	StartUpID	ThisStartUpID;	/* current SUI */
 extern	bool		InRecovery;
 extern	XLogRecPtr	MyLastRecPtr;
 

src/include/access/xlogdefs.h

@@ -0,0 +1,24 @@
+/*
+ *
+ * xlogdefs.h
+ *
+ * Postgres transaction log manager record pointer and
+ * system stratup number definitions
+ *
+ */
+#ifndef XLOG_DEFS_H
+#define XLOG_DEFS_H
+
+typedef struct XLogRecPtr
+{
+	uint32		xlogid;			/* log file #, 0 based */
+	uint32		xrecoff;		/* offset of record in log file */
+} XLogRecPtr;
+
+/*
+ * StartUpID (SUI) - system startups counter. It's to allow removing
+ * pg_log after shutdown, in future.
+ */
+typedef	uint32		StartUpID;
+
+#endif	 /* XLOG_DEFS_H */

src/include/access/xlogutils.h

@@ -9,8 +9,10 @@ extern bool XLogIsValidTuple(RelFileNode hnode, ItemPointer iptr);
 
 extern void XLogOpenLogRelation(void);
 
-extern Buffer XLogReadBuffer(bool extend, Relation reln, BlockNumber blkno);
+extern void XLogInitRelationCache(void);
 extern void XLogCloseRelationCache(void);
+
 extern Relation XLogOpenRelation(bool redo, RmgrId rmid, RelFileNode rnode);
+extern Buffer XLogReadBuffer(bool extend, Relation reln, BlockNumber blkno);
 
 #endif

src/include/storage/buf_internals.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: buf_internals.h,v 1.41 2000/10/23 04:10:14 vadim Exp $
+ * $Id: buf_internals.h,v 1.42 2000/10/28 16:21:00 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -109,6 +109,10 @@ typedef struct sbufdesc
 	bool		ri_lock;		/* read-intent lock */
 	bool		w_lock;			/* context exclusively locked */
 
+#ifdef XLOG
+	bool		cntxDirty;		/* new way to mark block as dirty */
+#endif
+
 	BufferBlindId blind;		/* was used to support blind write */
 
 	/*

src/include/storage/bufmgr.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: bufmgr.h,v 1.41 2000/10/20 11:01:21 vadim Exp $
+ * $Id: bufmgr.h,v 1.42 2000/10/28 16:21:00 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -15,7 +15,7 @@
 #define BUFMGR_H
 
 #include "storage/buf_internals.h"
-
+#include "access/xlogdefs.h"
 
 typedef void *Block;
 
@@ -177,4 +177,9 @@ extern void AbortBufferIO(void);
 extern bool BufferIsUpdatable(Buffer buffer);
 extern void MarkBufferForCleanup(Buffer buffer, void (*CleanupFunc)(Buffer));
 
+#ifdef XLOG
+extern void BufmgrCommit(void);
+extern void BufferSync(void);
+#endif
+
 #endif

src/include/storage/bufpage.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: bufpage.h,v 1.34 2000/10/21 15:43:36 vadim Exp $
+ * $Id: bufpage.h,v 1.35 2000/10/28 16:21:00 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -118,7 +118,8 @@ typedef OpaqueData *Opaque;
  */
 typedef struct PageHeaderData
 {
-#ifdef XLOG
+#ifdef XLOG						/* XXX LSN is member of *any* block, not */
+								/* only page-organized - 'll change later */
 	XLogRecPtr	pd_lsn;			/* LSN: next byte after last byte of xlog */
 								/* record for last change of this page */
 	StartUpID	pd_sui;			/* SUI of last changes (currently it's */

src/include/storage/smgr.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: smgr.h,v 1.22 2000/10/16 14:52:28 vadim Exp $
+ * $Id: smgr.h,v 1.23 2000/10/28 16:21:00 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -36,26 +36,19 @@ extern int smgrwrite(int16 which, Relation reln, BlockNumber blocknum,
 		  char *buffer);
 extern int smgrflush(int16 which, Relation reln, BlockNumber blocknum,
 		  char *buffer);
-#ifdef OLD_FILE_NAMING
-extern int smgrblindwrt(int16 which, char *dbname, char *relname,
-			 Oid dbid, Oid relid,
-			 BlockNumber blkno, char *buffer,
-			 bool dofsync);
-extern int smgrblindmarkdirty(int16 which, char *dbname, char *relname,
-				   Oid dbid, Oid relid,
-				   BlockNumber blkno);
-#else
 extern int smgrblindwrt(int16 which, RelFileNode rnode,
 						BlockNumber blkno, char *buffer, bool dofsync);
 extern int smgrblindmarkdirty(int16 which, RelFileNode rnode,
 						BlockNumber blkno);
-#endif
 extern int	smgrmarkdirty(int16 which, Relation reln, BlockNumber blkno);
 extern int	smgrnblocks(int16 which, Relation reln);
 extern int	smgrtruncate(int16 which, Relation reln, int nblocks);
 extern int	smgrcommit(void);
 extern int	smgrabort(void);
 
+#ifdef XLOG
+extern int	smgrsync(void);
+#endif
 
 
 /* internals: move me elsewhere -- ay 7/94 */
@@ -71,22 +64,18 @@ extern int	mdread(Relation reln, BlockNumber blocknum, char *buffer);
 extern int	mdwrite(Relation reln, BlockNumber blocknum, char *buffer);
 extern int	mdflush(Relation reln, BlockNumber blocknum, char *buffer);
 extern int	mdmarkdirty(Relation reln, BlockNumber blkno);
-#ifdef OLD_FILE_NAMING
-extern int mdblindwrt(char *dbname, char *relname, Oid dbid, Oid relid,
-		   BlockNumber blkno, char *buffer,
-		   bool dofsync);
-extern int mdblindmarkdirty(char *dbname, char *relname, Oid dbid, Oid relid,
-				 BlockNumber blkno);
-#else
 extern int mdblindwrt(RelFileNode rnode, BlockNumber blkno,
 						char *buffer, bool dofsync);
 extern int mdblindmarkdirty(RelFileNode rnode, BlockNumber blkno);
-#endif
 extern int	mdnblocks(Relation reln);
 extern int	mdtruncate(Relation reln, int nblocks);
 extern int	mdcommit(void);
 extern int	mdabort(void);
 
+#ifdef XLOG
+extern int	mdsync(void);
+#endif
+
 /* mm.c */
 extern SPINLOCK MMCacheLock;