WAL

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)