Applying InnoDB snashot 5.0-ss4007, part 2. Fixes

Bug #18828: If InnoDB runs out of undo slots, it returns misleading 'table is full'

This is a backport of code already in 5.1+.  The error message change referred
to in the detailed revision comments is still pending.

Detailed revision comments:

r3937 | calvin | 2009-01-15 03:11:56 +0200 (Thu, 15 Jan 2009) | 17 lines
branches/5.0:

Backport the fix for Bug#18828. Return DB_TOO_MANY_CONCURRENT_TRXS
when we run out of UNDO slots in the rollback segment. The backport
is requested by MySQL under bug#41529 - Safe handling of InnoDB running
out of undo log slots.

This is a partial fix since the MySQL error code requested to properly
report the error condition back to the client has not yet materialized.
Currently we have #ifdef'd the error code translation in ha_innodb.cc.
This will have to be changed as and when MySQl add the new requested
code or an equivalent code that we can then use.

Given the above, currently we will get the old behavior, not the
"fixed" and intended behavior.

Approved by:	Heikki (on IM)

innobase/dict/dict0crea.c

@@ -1249,7 +1249,8 @@ dict_create_or_check_foreign_constraint_tables(void)
 		fprintf(stderr, "InnoDB: error %lu in creation\n",
 			(ulong) error);
 		
-		ut_a(error == DB_OUT_OF_FILE_SPACE);
+		ut_a(error == DB_OUT_OF_FILE_SPACE
+		     || error == DB_TOO_MANY_CONCURRENT_TRXS);
 
 		fprintf(stderr, "InnoDB: creation failed\n");
 		fprintf(stderr, "InnoDB: tablespace is full\n");

innobase/include/db0err.h

@@ -70,6 +70,11 @@ Created 5/24/1996 Heikki Tuuri
 					work with e.g., FT indexes created by
 					a later version of the engine. */
 
+#define DB_TOO_MANY_CONCURRENT_TRXS 47	/* when InnoDB runs out of the
+					preconfigured undo slots, this can
+					only happen when there are too many
+					concurrent transactions */
+
 /* The following are partial failure codes */
 #define DB_FAIL 		1000
 #define DB_OVERFLOW 		1001

innobase/include/trx0undo.h

@@ -222,11 +222,14 @@ trx_undo_lists_init(
 Assigns an undo log for a transaction. A new undo log is created or a cached
 undo log reused. */
 
-trx_undo_t*
+ulint
 trx_undo_assign_undo(
 /*=================*/
-			/* out: the undo log, NULL if did not succeed: out of
-			space */
+				/* out: DB_SUCCESS if undo log assign
+				 * successful, possible error codes are:
+				 * ER_TOO_MANY_CONCURRENT_TRXS
+				 * DB_OUT_OF_FILE_SPAC
+				 * DB_OUT_OF_MEMORY */
 	trx_t*		trx,	/* in: transaction */
 	ulint		type);	/* in: TRX_UNDO_INSERT or TRX_UNDO_UPDATE */
 /**********************************************************************

innobase/row/row0mysql.c

@@ -494,7 +494,8 @@ handle_new_error:
 		/* MySQL will roll back the latest SQL statement */
 	} else if (err == DB_ROW_IS_REFERENCED
 		   || err == DB_NO_REFERENCED_ROW
-		   || err == DB_CANNOT_ADD_CONSTRAINT) {
+		   || err == DB_CANNOT_ADD_CONSTRAINT
+		   || err == DB_TOO_MANY_CONCURRENT_TRXS) {
            	if (savept) {
 			/* Roll back the latest, possibly incomplete
 			insertion or update */

innobase/trx/trx0rec.c

@@ -1013,6 +1013,7 @@ trx_undo_report_row_operation(
 	ibool		is_insert;
 	trx_rseg_t*	rseg;
 	mtr_t		mtr;
+	ulint		err		= DB_SUCCESS;
 	mem_heap_t*	heap		= NULL;
 	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 	ulint*		offsets		= offsets_;
@@ -1024,7 +1025,7 @@ trx_undo_report_row_operation(
 
 		*roll_ptr = ut_dulint_zero;
 
-		return(DB_SUCCESS);
+		return(err);
 	}
 		
 	ut_ad(thr);
@@ -1042,7 +1043,7 @@ trx_undo_report_row_operation(
 
 		if (trx->insert_undo == NULL) {
 
-			trx_undo_assign_undo(trx, TRX_UNDO_INSERT);
+			err = trx_undo_assign_undo(trx, TRX_UNDO_INSERT);
 		}
 
 		undo = trx->insert_undo;
@@ -1052,7 +1053,7 @@ trx_undo_report_row_operation(
 
 		if (trx->update_undo == NULL) {
 
-			trx_undo_assign_undo(trx, TRX_UNDO_UPDATE);
+			err = trx_undo_assign_undo(trx, TRX_UNDO_UPDATE);
 
 		}
 
@@ -1060,11 +1061,11 @@ trx_undo_report_row_operation(
 		is_insert = FALSE;
 	}
 
-	if (undo == NULL) {
-		/* Did not succeed: out of space */
+	if (err != DB_SUCCESS) {
+		/* Did not succeed: return the error encountered */
 		mutex_exit(&(trx->undo_mutex));
 
-		return(DB_OUT_OF_FILE_SPACE);
+		return(err);
 	}
 
 	page_no = undo->last_page_no;
@@ -1154,7 +1155,7 @@ trx_undo_report_row_operation(
 	if (UNIV_LIKELY_NULL(heap)) {
 		mem_heap_free(heap);
 	}
-	return(DB_SUCCESS);
+	return(err);
 }
 
 /*============== BUILDING PREVIOUS VERSION OF A RECORD ===============*/

innobase/trx/trx0undo.c

@@ -374,26 +374,31 @@ trx_undo_page_init(
 /*******************************************************************
 Creates a new undo log segment in file. */
 static
-page_t*
+ulint
 trx_undo_seg_create(
 /*================*/
-				/* out: segment header page x-latched, NULL
-				if no space left */
+				/* out: DB_SUCCESS if page creation OK
+				possible error codes are:
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE */
 	trx_rseg_t*	rseg __attribute__((unused)),/* in: rollback segment */
 	trx_rsegf_t*	rseg_hdr,/* in: rollback segment header, page
 				x-latched */
 	ulint		type,	/* in: type of the segment: TRX_UNDO_INSERT or
 				TRX_UNDO_UPDATE */
 	ulint*		id,	/* out: slot index within rseg header */
+	page_t**	undo_page,
+				/* out: segment header page x-latched, NULL
+				if there was an error */
 	mtr_t*		mtr)	/* in: mtr */
 {
 	ulint		slot_no;
 	ulint		space;
-	page_t* 	undo_page;
 	trx_upagef_t*	page_hdr;
 	trx_usegf_t*	seg_hdr;
 	ulint		n_reserved;
 	ibool		success;
+	ulint		err = DB_SUCCESS;
 
 	ut_ad(mtr && id && rseg_hdr);
 #ifdef UNIV_SYNC_DEBUG
@@ -411,7 +416,7 @@ trx_undo_seg_create(
 "InnoDB: Warning: cannot find a free slot for an undo log. Do you have too\n"
 "InnoDB: many active transactions running concurrently?\n");
 
-		return(NULL);
+		return(DB_TOO_MANY_CONCURRENT_TRXS);
 	}
 
 	space = buf_frame_get_space_id(rseg_hdr);
@@ -420,29 +425,29 @@ trx_undo_seg_create(
 									mtr);
 	if (!success) {
 
-		return(NULL);
+		return(DB_OUT_OF_FILE_SPACE);
 	}
 
 	/* Allocate a new file segment for the undo log */
-	undo_page = fseg_create_general(space, 0,
+	*undo_page = fseg_create_general(space, 0,
 			TRX_UNDO_SEG_HDR + TRX_UNDO_FSEG_HEADER, TRUE, mtr);
 
 	fil_space_release_free_extents(space, n_reserved);
 			
-	if (undo_page == NULL) {
+	if (*undo_page == NULL) {
 		/* No space left */
 
-		return(NULL);
+		return(DB_OUT_OF_FILE_SPACE);
 	}
 
 #ifdef UNIV_SYNC_DEBUG
-	buf_page_dbg_add_level(undo_page, SYNC_TRX_UNDO_PAGE);
+	buf_page_dbg_add_level(*undo_page, SYNC_TRX_UNDO_PAGE);
 #endif /* UNIV_SYNC_DEBUG */
 
-	page_hdr = undo_page + TRX_UNDO_PAGE_HDR;
-	seg_hdr = undo_page + TRX_UNDO_SEG_HDR;
+	page_hdr = *undo_page + TRX_UNDO_PAGE_HDR;
+	seg_hdr = *undo_page + TRX_UNDO_SEG_HDR;
 
-	trx_undo_page_init(undo_page, type, mtr);
+	trx_undo_page_init(*undo_page, type, mtr);
 
 	mlog_write_ulint(page_hdr + TRX_UNDO_PAGE_FREE,
 				TRX_UNDO_SEG_HDR + TRX_UNDO_SEG_HDR_SIZE,
@@ -456,10 +461,11 @@ trx_undo_seg_create(
 					page_hdr + TRX_UNDO_PAGE_NODE, mtr);
 
 	trx_rsegf_set_nth_undo(rseg_hdr, slot_no,
-				buf_frame_get_page_no(undo_page), mtr);
+				buf_frame_get_page_no(*undo_page), mtr);
+
 	*id = slot_no;
 	
-	return(undo_page);
+	return(err);
 }
 
 /**************************************************************************
@@ -1400,6 +1406,11 @@ trx_undo_mem_create(
 
 	undo = mem_alloc(sizeof(trx_undo_t));
 
+	if (undo == NULL) {
+
+		return NULL;
+	}
+
 	undo->id = id;
 	undo->type = type;
 	undo->state = TRX_UNDO_ACTIVE;
@@ -1479,11 +1490,15 @@ trx_undo_mem_free(
 /**************************************************************************
 Creates a new undo log. */
 static
-trx_undo_t*
+ulint
 trx_undo_create(
 /*============*/
-				/* out: undo log object, NULL if did not
-				succeed: out of space */
+				/* out: DB_SUCCESS if successful in creating
+				the new undo lob object, possible error
+				codes are: 
+				DB_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE 
+				DB_OUT_OF_MEMORY*/
 	trx_t*		trx,	/* in: transaction */
 	trx_rseg_t*	rseg,	/* in: rollback segment memory copy */
 	ulint		type,	/* in: type of the log: TRX_UNDO_INSERT or
@@ -1491,14 +1506,16 @@ trx_undo_create(
 	dulint		trx_id,	/* in: id of the trx for which the undo log
 				is created */ 
 	XID*		xid,	/* in: X/Open transaction identification*/
+	trx_undo_t**	undo,	/* out: the new undo log object, undefined
+				 * if did not succeed */
 	mtr_t*		mtr)	/* in: mtr */
 {
 	trx_rsegf_t*	rseg_header;
 	ulint		page_no;
 	ulint		offset;
 	ulint		id;
-	trx_undo_t*	undo;
 	page_t*		undo_page;
+	ulint		err;
 
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(rseg->mutex)));
@@ -1506,21 +1523,22 @@ trx_undo_create(
 
 	if (rseg->curr_size == rseg->max_size) {
 
-		return(NULL);
+		return(DB_OUT_OF_FILE_SPACE);
 	}
 
 	rseg->curr_size++;
 
 	rseg_header = trx_rsegf_get(rseg->space, rseg->page_no, mtr);
 
-	undo_page = trx_undo_seg_create(rseg, rseg_header, type, &id, mtr);
+	err = trx_undo_seg_create(rseg, rseg_header, type, &id,
+							&undo_page, mtr);
 	
-	if (undo_page == NULL) {
+	if (err != DB_SUCCESS) {
 		/* Did not succeed */
 
 		rseg->curr_size--;
 
-		return(NULL);
+		return(err);
 	}
 
 	page_no = buf_frame_get_page_no(undo_page);
@@ -1532,9 +1550,14 @@ trx_undo_create(
 					undo_page + offset, mtr);
 	}
 
-	undo = trx_undo_mem_create(rseg, id, type, trx_id, xid,
+	*undo = trx_undo_mem_create(rseg, id, type, trx_id, xid,
 							page_no, offset);
-	return(undo);
+	if (*undo == NULL) {
+
+		err = DB_OUT_OF_MEMORY;
+	}
+
+	return(err);
 }
 
 /*================ UNDO LOG ASSIGNMENT AND CLEANUP =====================*/
@@ -1653,17 +1676,20 @@ trx_undo_mark_as_dict_operation(
 Assigns an undo log for a transaction. A new undo log is created or a cached
 undo log reused. */
 
-trx_undo_t*
+ulint
 trx_undo_assign_undo(
 /*=================*/
-			/* out: the undo log, NULL if did not succeed: out of
-			space */
+				/* out: DB_SUCCESS if undo log assign
+				successful, possible error codes are:
+				DD_TOO_MANY_CONCURRENT_TRXS
+				DB_OUT_OF_FILE_SPACE DB_OUT_OF_MEMORY*/
 	trx_t*		trx,	/* in: transaction */
 	ulint		type)	/* in: TRX_UNDO_INSERT or TRX_UNDO_UPDATE */
 {
 	trx_rseg_t*	rseg;
 	trx_undo_t*	undo;
 	mtr_t		mtr;
+	ulint		err = DB_SUCCESS;
 
 	ut_ad(trx);
 	ut_ad(trx->rseg);
@@ -1684,15 +1710,11 @@ trx_undo_assign_undo(
 	undo = trx_undo_reuse_cached(trx, rseg, type, trx->id, &trx->xid,
 									&mtr);
 	if (undo == NULL) {
-		undo = trx_undo_create(trx, rseg, type, trx->id, &trx->xid,
-									&mtr);
-		if (undo == NULL) {
-			/* Did not succeed */
-
-			mutex_exit(&(rseg->mutex));
-			mtr_commit(&mtr);
+		err = trx_undo_create(trx, rseg, type, trx->id, &trx->xid,
+								&undo, &mtr);
+		if (err != DB_SUCCESS) {
 
-			return(NULL);
+			goto func_exit;
 		}
 	}
 
@@ -1710,10 +1732,11 @@ trx_undo_assign_undo(
 		trx_undo_mark_as_dict_operation(trx, undo, &mtr);
 	}
 
+func_exit:
 	mutex_exit(&(rseg->mutex));
 	mtr_commit(&mtr);
 
-	return(undo);
+	return err;
 }
 
 /**********************************************************************

sql/ha_innodb.cc

@@ -524,6 +524,20 @@ convert_error_code_to_mysql(
                 mark_transaction_to_rollback(thd, TRUE);
 
     		return(HA_ERR_LOCK_TABLE_FULL);
+	} else if (error == DB_TOO_MANY_CONCURRENT_TRXS) {
+
+		/* Once MySQL add the appropriate code to errmsg.txt then
+		we can get rid of this #ifdef. NOTE: The code checked by
+		the #ifdef is the suggested name for the error condition
+		and the actual error code name could very well be different.
+		This will require some monitoring, ie. the status
+		of this request on our part.*/
+#ifdef ER_TOO_MANY_CONCURRENT_TRXS
+		return(ER_TOO_MANY_CONCURRENT_TRXS);
+#else
+		return(HA_ERR_RECORD_FILE_FULL);
+#endif
+
 	} else if (error == DB_UNSUPPORTED) {
 
 		return(HA_ERR_UNSUPPORTED);