trx0undo.h, trx0undo.c, trx0trx.c, trx0roll.c:

  Make InnoDB-5.0.3 to process log records of 4.1 undo log header create and reuse just like in 4.1; storing XID to the start of the undo log is a separately logged operation; this conforms to Rule 3 of InnoDB redo logging

innobase/include/trx0undo.h

@@ -307,23 +307,6 @@ trx_undo_parse_discard_latest(
 	byte*	end_ptr,/* in: buffer end */
 	page_t*	page,	/* in: page or NULL */
 	mtr_t*	mtr);	/* in: mtr or NULL */
-/************************************************************************
-Write X/Open XA Transaction Identification (XID) to undo log header */
-
-void
-trx_undo_write_xid(
-/*===============*/
-	trx_ulogf_t*	log_hdr,/* in: undo log header */
-	XID*		xid);	/* in: X/Open XA Transaction Identification */
-
-/************************************************************************
-Read X/Open XA Transaction Identification (XID) from undo log header */
-
-void
-trx_undo_read_xid(
-/*==============*/
-	trx_ulogf_t*	log_hdr,/* in: undo log header */
-	XID*		xid);	/* out: X/Open XA Transaction Identification */
 
 /* Types of an undo log segment */
 #define	TRX_UNDO_INSERT		1	/* contains undo entries for inserts */
@@ -419,7 +402,8 @@ struct trx_undo_struct{
 #define TRX_UNDO_PAGE_HDR_SIZE	(6 + FLST_NODE_SIZE)
 
 /* An update undo segment with just one page can be reused if it has
-< this number bytes used */
+< this number bytes used; we must leave space at least for one new undo
+log header on the page */
 
 #define TRX_UNDO_PAGE_REUSE_LIMIT	(3 * UNIV_PAGE_SIZE / 4)
 
@@ -488,17 +472,25 @@ page of an update undo log segment. */
 #define TRX_UNDO_HISTORY_NODE	34	/* If the log is put to the history
 					list, the file list node is here */
 /*-------------------------------------------------------------*/
+#define TRX_UNDO_LOG_OLD_HDR_SIZE (34 + FLST_NODE_SIZE)
+
+/* Note: the writing of the undo log old header is coded by a log record
+MLOG_UNDO_HDR_CREATE or MLOG_UNDO_HDR_REUSE. The appending of an XID to the
+header is logged separately. In this sense, the XID is not really a member
+of the undo log header. TODO: do not append the XID to the log header if XA
+is not needed by the user. The XID wastes about 150 bytes of space in every
+undo log. In the history list we may have millions of undo logs, which means
+quite a large overhead. */
+
 /* X/Open XA Transaction Identification (XID)                  */
 
-#define	TRX_UNDO_XA_FORMAT	(34 + FLST_NODE_SIZE)
+#define	TRX_UNDO_XA_FORMAT	(TRX_UNDO_LOG_OLD_HDR_SIZE)
 #define	TRX_UNDO_XA_TRID_LEN	(TRX_UNDO_XA_FORMAT + 4)
 #define	TRX_UNDO_XA_BQUAL_LEN	(TRX_UNDO_XA_TRID_LEN + 4)
 #define	TRX_UNDO_XA_XID		(TRX_UNDO_XA_BQUAL_LEN + 4)
-#define	TRX_UNDO_XA_LEN		(TRX_UNDO_XA_XID + XIDDATASIZE)
-
-/*-------------------------------------------------------------*/
-#define	TRX_UNDO_LOG_HDR_SIZE	(TRX_UNDO_XA_LEN)
-/*-------------------------------------------------------------*/
+/*--------------------------------------------------------------*/
+#define TRX_UNDO_LOG_XA_HDR_SIZE (TRX_UNDO_XA_XID + XIDDATASIZE)
+				/* Total size of the header with the XA XID */
 
 #ifndef UNIV_NONINL
 #include "trx0undo.ic"

innobase/trx/trx0roll.c

@@ -950,7 +950,7 @@ try_again:
 		if (progress_pct != trx_roll_progress_printed_pct) {
 			if (trx_roll_progress_printed_pct == 0) {
 				fprintf(stderr,
-"\nInnoDB: Progress in percents: %lu\n", (ulong) progress_pct);
+"\nInnoDB: Progress in percents: %lu", (ulong) progress_pct);
 			} else {
 				fprintf(stderr,
 				" %lu", (ulong) progress_pct);

innobase/trx/trx0trx.c

@@ -162,7 +162,7 @@ trx_create(
 	trx->read_view = NULL;
 
 	/* Set X/Open XA transaction identification to NULL */
-	memset(&trx->xid,0,sizeof(trx->xid));
+	memset(&trx->xid, 0, sizeof(trx->xid));
 	trx->xid.formatID = -1;
 
 	return(trx);
@@ -436,12 +436,8 @@ trx_lists_init_at_db_start(void)
 				if (undo->state == TRX_UNDO_PREPARED) {
 
 					fprintf(stderr,
-"InnoDB: Transaction %lu %lu was in the XA prepared state. We change it to\n"
-"InnoDB: the 'active' state, so that InnoDB's true-and-tested crash\n"
-"InnoDB: recovery will roll it back. If mysqld refuses to start after\n"
-"InnoDB: this, you may be able to resolve the problem by moving the binlog\n"
-"InnoDB: files to a safe place, and deleting all binlog files and the binlog\n"
-"InnoDB: .index file from the datadir.\n", ut_dulint_get_high(trx->id),
+"InnoDB: Transaction %lu %lu was in the XA prepared state.\n",
+					ut_dulint_get_high(trx->id),
 					ut_dulint_get_low(trx->id));
 
 				/*	trx->conc_state = TRX_PREPARED; */
@@ -504,12 +500,8 @@ trx_lists_init_at_db_start(void)
 					if (undo->state == TRX_UNDO_PREPARED) {
 
 					fprintf(stderr,
-"InnoDB: Transaction %lu %lu was in the XA prepared state. We change it to\n"
-"InnoDB: the 'active' state, so that InnoDB's true-and-tested crash\n"
-"InnoDB: recovery will roll it back. If mysqld refuses to start after\n"
-"InnoDB: this, you may be able to resolve the problem by moving the binlog\n"
-"InnoDB: files to a safe place, and deleting all binlog files and the binlog\n"
-"InnoDB: .index file from the datadir.\n", ut_dulint_get_high(trx->id),
+"InnoDB: Transaction %lu %lu was in the XA prepared state.\n",
+					ut_dulint_get_high(trx->id),
 					ut_dulint_get_low(trx->id));
 
 					/* trx->conc_state = TRX_PREPARED; */
@@ -786,8 +778,8 @@ trx_commit_off_kernel(
 		mutex_enter(&kernel_mutex);
 	}
 
-	ut_ad(trx->conc_state == TRX_ACTIVE || trx->conc_state == TRX_PREPARED);
-
+	ut_ad(trx->conc_state == TRX_ACTIVE
+					|| trx->conc_state == TRX_PREPARED);
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&kernel_mutex));
 #endif /* UNIV_SYNC_DEBUG */
@@ -1744,10 +1736,10 @@ trx_prepare_off_kernel(
 	trx_t*	trx)	/* in: transaction */
 {
 	page_t*		update_hdr_page;
-	dulint		lsn;
 	trx_rseg_t*	rseg;
 	trx_undo_t*	undo;
 	ibool		must_flush_log	= FALSE;
+	dulint		lsn;
 	mtr_t		mtr;
 	
 #ifdef UNIV_SYNC_DEBUG
@@ -1765,10 +1757,9 @@ trx_prepare_off_kernel(
 		must_flush_log = TRUE;
 
 		/* Change the undo log segment states from TRX_UNDO_ACTIVE
-		to some other state: these modifications to the file data
-		structure define the transaction as prepared in the file
-		based world, at the serialization point of the log sequence
-		number lsn obtained below. */
+		to TRX_UNDO_PREPARED: these modifications to the file data
+		structure define the transaction as prepared in the
+		file-based world, at the serialization point of lsn. */
 
 		mutex_enter(&(rseg->mutex));
 			
@@ -1780,18 +1771,20 @@ trx_prepare_off_kernel(
 		undo = trx->update_undo;
 
 		if (undo) {
-
 			/* It is not necessary to obtain trx->undo_mutex here
 			because only a single OS thread is allowed to do the
 			transaction prepare for this transaction. */
 					
-			update_hdr_page = trx_undo_set_state_at_prepare(trx,								undo, &mtr);
+			update_hdr_page = trx_undo_set_state_at_prepare(trx,
+								undo, &mtr);
 		}
 
 		mutex_exit(&(rseg->mutex));
 
 		/*--------------*/
-		mtr_commit(&mtr);
+		mtr_commit(&mtr);	/* This mtr commit makes the
+					transaction prepared in the file-based
+					world */
 		/*--------------*/
  		lsn = mtr.end_lsn;
 
@@ -1806,13 +1799,6 @@ trx_prepare_off_kernel(
 	trx->conc_state = TRX_PREPARED;
 	/*--------------------------------------*/
 
-	if (trx->read_view) {
-		read_view_close(trx->read_view);
-
-		mem_heap_empty(trx->read_view_heap);
-		trx->read_view = NULL;
-	}
-
 	if (must_flush_log) {
 
 		mutex_exit(&kernel_mutex);
@@ -1834,7 +1820,7 @@ Does the transaction prepare for MySQL. */
 
 ulint
 trx_prepare_for_mysql(
-/*=================*/
+/*====-=============*/
 			/* out: 0 or error number */
 	trx_t*	trx)	/* in: trx handle */
 {
@@ -1881,9 +1867,8 @@ trx_recover_for_mysql(
 	fprintf(stderr,
 		"  InnoDB: Starting recovery for XA transactions...\n");
 
-
-	/* We should set those transactions which are in
-	the prepared state to the xid_list */
+	/* We should set those transactions which are in the prepared state
+	to the xid_list */
 
 	mutex_enter(&kernel_mutex);
 
@@ -1891,14 +1876,7 @@ trx_recover_for_mysql(
 
 	while (trx) {
 		if (trx->conc_state == TRX_PREPARED) {
-			xid_list[count].formatID = trx->xid.formatID;
-			xid_list[count].gtrid_length = trx->xid.gtrid_length;
-			xid_list[count].bqual_length = trx->xid.bqual_length;
-
-			memcpy(xid_list[count].data,
-					trx->xid.data,
-					trx->xid.gtrid_length + 
-					trx->xid.bqual_length);
+			xid_list[count] = trx->xid;
 
 			ut_print_timestamp(stderr);
 			fprintf(stderr,
@@ -1935,7 +1913,7 @@ trx_recover_for_mysql(
 This function is used to find one X/Open XA distributed transaction
 which is in the prepared state */
 
-trx_t * 
+trx_t*
 trx_get_trx_by_xid(
 /*===============*/
 			/* out: trx or NULL */
@@ -1944,6 +1922,7 @@ trx_get_trx_by_xid(
 	trx_t*	trx;
 
 	if (xid == NULL) {
+
 		return (NULL);
 	}
  
@@ -1972,6 +1951,7 @@ trx_get_trx_by_xid(
 
 	if (trx) {
 		if (trx->conc_state != TRX_PREPARED) {
+
 			return(NULL);
 		}
 
@@ -1980,4 +1960,3 @@ trx_get_trx_by_xid(
 		return(NULL);
 	}
 }
-