lock0lock.c:

  Eliminate a spurious deadlock when an insert waits for an update and a second update comes to wait after the insert


innobase/lock/lock0lock.c:
  Eliminate a spurious deadlock when an insert waits for an update and a second update comes to wait after the insert

innobase/lock/lock0lock.c

@@ -53,10 +53,9 @@ bump into an x-lock set there.
 
 Different transaction can have conflicting locks set on the gap at the
 same time. The locks on the gap are purely inhibitive: an insert cannot
-be made, or a select cursor may have to wait, if a different transaction
+be made, or a select cursor may have to wait if a different transaction
 has a conflicting lock on the gap. An x-lock on the gap does not give
-the right to insert into the gap if there are conflicting locks granted
-on the gap at the same time.
+the right to insert into the gap.
 
 An explicit lock can be placed on a user record or the supremum record of
 a page. The locks on the supremum record are always thought to be of the gap
@@ -88,7 +87,7 @@ locks.
 RULE 4: If a there is a waiting lock request in a queue, no lock request,
 -------
 gap or not, can be inserted ahead of it in the queue. In record deletes
-and page splits, new gap type locks can be created by the database manager
+and page splits new gap type locks can be created by the database manager
 for a transaction, and without rule 4, the waits-for graph of transactions
 might become cyclic without the database noticing it, as the deadlock check
 is only performed when a transaction itself requests a lock!
@@ -839,40 +838,93 @@ lock_mode_compatible(
 }
 
 /*************************************************************************
-Returns LOCK_X if mode is LOCK_S, and vice versa. */
+Checks if a lock request for a new lock has to wait for request lock2. */
 UNIV_INLINE
-ulint
-lock_get_confl_mode(
-/*================*/
-			/* out: conflicting basic lock mode */
-	ulint	mode)	/* in: LOCK_S or LOCK_X */
+ibool
+lock_rec_has_to_wait(
+/*=================*/
+			/* out: TRUE if new lock has to wait for lock2 to be
+			removed */
+	trx_t*	trx,	/* in: trx of new lock */
+	ulint	mode,	/* in: LOCK_S or LOCK_X */
+	ulint	gap,	/* in: LOCK_GAP or 0 */
+	ulint	insert_intention,
+			/* in: LOCK_INSERT_INTENTION or 0 */
+	lock_t*	lock2)	/* in: another record lock; NOTE that it is assumed
+			that this has a lock bit set on the same record as
+			in lock1 */
 {
-	ut_ad(mode == LOCK_X || mode == LOCK_S);
+	ut_ad(trx && lock2);
+	ut_ad(lock_get_type(lock2) == LOCK_REC);
+	ut_ad(mode == LOCK_S || mode == LOCK_X);
+	ut_ad(gap == LOCK_GAP || gap == 0);
+	ut_ad(insert_intention == LOCK_INSERT_INTENTION
+	      				|| insert_intention == 0);
+
+	if (trx != lock2->trx && !lock_mode_compatible(mode,
+				     		lock_get_mode(lock2))) {
+
+		/* We have somewhat complex rules when gap type
+		record locks cause waits */
+
+		if (!gap && lock_rec_get_insert_intention(lock2)) {
 
-	if (mode == LOCK_S) {
+			/* Request of a full next-key record does not
+			need to wait for an insert intention lock to be
+			removed. This is ok since our rules allow conflicting
+			locks on gaps. This eliminates a spurious deadlock
+			caused by a next-key lock waiting for an insert
+			intention lock; when the insert intention lock was
+			granted, the insert deadlocked on the waiting
+			next-key lock. */
+				
+			return(FALSE);
+		}
+
+		if (insert_intention && lock_rec_get_insert_intention(lock2)) {
+
+			/* An insert intention is not disturbed by another
+			insert intention; this removes a spurious deadlock
+			caused by inserts which had to wait for a next-key
+			lock to be removed */
+
+			return(FALSE);
+		}
 
-		return(LOCK_X);
+		return(TRUE);
 	}
 
-	return(LOCK_S);
+	return(FALSE);
 }
 
 /*************************************************************************
-Checks if a lock request lock1 has to wait for request lock2. NOTE that we,
-for simplicity, ignore the gap bits in locks, and treat gap type lock
-requests like non-gap lock requests. */
-UNIV_INLINE
+Checks if a lock request lock1 has to wait for request lock2. */
+static
 ibool
 lock_has_to_wait(
 /*=============*/
-			/* out: TRUE if lock1 has to wait lock2 to be removed */
-	lock_t*	lock1,	/* in: waiting record lock */
+			/* out: TRUE if lock1 has to wait for lock2 to be
+			removed */
+	lock_t*	lock1,	/* in: waiting lock */
 	lock_t*	lock2)	/* in: another lock; NOTE that it is assumed that this
-			has a lock bit set on the same record as in lock1 */
+			has a lock bit set on the same record as in lock1 if
+			the locks are record locks */
 {
+	ut_ad(lock1 && lock2);
+
 	if (lock1->trx != lock2->trx
 			&& !lock_mode_compatible(lock_get_mode(lock1),
 				     		lock_get_mode(lock2))) {
+		if (lock_get_type(lock1) == LOCK_REC) {
+			ut_ad(lock_get_type(lock2) == LOCK_REC);
+				
+			return(lock_rec_has_to_wait(lock1->trx,
+				lock_get_mode(lock1),
+				lock_rec_get_gap(lock1),
+				lock_rec_get_insert_intention(lock1),
+				lock2));
+		}
+
 		return(TRUE);
 	}
 
@@ -1021,6 +1073,7 @@ lock_rec_get_next_on_page(
 	ulint	page_no;
 
 	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type(lock) == LOCK_REC);
 
 	space = lock->un_member.rec_lock.space;
 	page_no = lock->un_member.rec_lock.page_no;
@@ -1147,6 +1200,7 @@ lock_rec_get_next(
 	lock_t*	lock)	/* in: lock */
 {
 	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type(lock) == LOCK_REC);
 
 	for (;;) {
 		lock = lock_rec_get_next_on_page(lock);
@@ -1204,6 +1258,8 @@ lock_rec_bitmap_reset(
 	ulint	n_bytes;
 	ulint	i;
 
+	ut_ad(lock_get_type(lock) == LOCK_REC);
+
 	/* Reset to zero the bitmap which resides immediately after the lock
 	struct */
 
@@ -1233,6 +1289,8 @@ lock_rec_copy(
 	lock_t*	dupl_lock;
 	ulint	size;
 
+	ut_ad(lock_get_type(lock) == LOCK_REC);
+
 	size = sizeof(lock_t) + lock_rec_get_n_bits(lock) / 8;	
 
 	dupl_lock = mem_heap_alloc(heap, size);
@@ -1348,6 +1406,7 @@ lock_rec_has_expl(
 		if (lock->trx == trx
 		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)
 		    && !lock_get_wait(lock)
+		    && !lock_rec_get_insert_intention(lock) /* we play safe */
 		    && !(lock_rec_get_gap(lock)
 					|| page_rec_is_supremum(rec))) {
 		    	return(lock);
@@ -1360,9 +1419,8 @@ lock_rec_has_expl(
 }
 			
 /*************************************************************************
-Checks if some other transaction has an explicit lock request stronger or
-equal to mode on rec or gap, waiting or granted, in the lock queue. */
-UNIV_INLINE
+Checks if some other transaction has a lock request in the queue. */
+static
 lock_t*
 lock_rec_other_has_expl_req(
 /*========================*/
@@ -1372,15 +1430,9 @@ lock_rec_other_has_expl_req(
 			into account, or 0 if not */
 	ulint	wait,	/* in: LOCK_WAIT if also waiting locks are
 			taken into account, or 0 if not */
-	ibool	ignore_insert_intention,/* in: FALSE if also locks where the
-			LOCK_INSERT_INTENTION
-			flag is set should be taken into account; we can
-			use this option to prevent unnecessary deadlocks
-			between inserts which have needed to set a wait
-			lock to wait for some next-key lock to be released */
 	rec_t*	rec,	/* in: record to look at */	
-	trx_t*	trx)	/* in: transaction, or NULL if requests
-			by any transaction are wanted */
+	trx_t*	trx)	/* in: transaction, or NULL if requests by all
+			transactions are taken into account */
 {
 	lock_t*	lock;
 	
@@ -1388,8 +1440,7 @@ lock_rec_other_has_expl_req(
 	ut_ad(mode == LOCK_X || mode == LOCK_S);
 	ut_ad(gap == 0 || gap == LOCK_GAP);
 	ut_ad(wait == 0 || wait == LOCK_WAIT);
-	ut_ad(ignore_insert_intention == TRUE
-					|| ignore_insert_intention == FALSE);
+
 	lock = lock_rec_get_first(rec);
 
 	while (lock) {
@@ -1399,14 +1450,7 @@ lock_rec_other_has_expl_req(
 		    && (wait || !lock_get_wait(lock))
 		    && lock_mode_stronger_or_eq(lock_get_mode(lock), mode)) {
 
-		    	if (ignore_insert_intention
-				&& lock_rec_get_insert_intention(lock)) {
-				
-				/* We can ignore this lock */
-			} else {
-
-		    		return(lock);
-		    	}
+		    	return(lock);
 		}
 
 		lock = lock_rec_get_next(rec, lock);
@@ -1415,6 +1459,44 @@ lock_rec_other_has_expl_req(
 	return(NULL);
 }
 
+/*************************************************************************
+Checks if some other transaction has a conflicting explicit lock request
+in the queue, so that we have to wait. */
+static
+lock_t*
+lock_rec_other_has_conflicting(
+/*===========================*/
+			/* out: lock or NULL */
+	ulint	mode,	/* in: lock mode of the lock we are going to reserve */
+	ulint	gap,	/* in: LOCK_GAP if we are going to reserve a gap type
+			lock, else 0 */
+	ulint	insert_intention,
+			/* in: LOCK_INSERT_INTENTION if we are going to
+			reserve an insert intention lock */
+	rec_t*	rec,	/* in: record to look at */	
+	trx_t*	trx)	/* in: our transaction */
+{
+	lock_t*	lock;
+	
+	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(mode == LOCK_X || mode == LOCK_S);
+	ut_ad(gap == 0 || gap == LOCK_GAP);
+	ut_ad(insert_intention == LOCK_INSERT_INTENTION
+						|| insert_intention == 0);
+	lock = lock_rec_get_first(rec);
+
+	while (lock) {
+		if (lock_rec_has_to_wait(trx, mode, gap, insert_intention,
+								lock)) {
+			return(lock);
+		}
+		
+		lock = lock_rec_get_next(rec, lock);
+	}
+
+	return(NULL);
+}
+
 /*************************************************************************
 Looks for a suitable type record lock struct by the same trx on the same page.
 This can be used to save space when a new record lock should be set on a page:
@@ -1584,7 +1666,7 @@ lock_rec_enqueue_waiting(
 	ulint		type_mode,/* in: lock mode this transaction is
 				requesting: LOCK_S or LOCK_X, ORed with
 				LOCK_GAP if a gap lock is requested, ORed
-				with LOCK_INSERT_INTENTION if the waiting
+				with LOCK_INSERT_INTENTION if this waiting
 				lock request is set when performing an
 				insert of an index record */
 	rec_t*		rec,	/* in: record */
@@ -1674,11 +1756,11 @@ lock_rec_add_to_queue(
 	ut_ad((type_mode & (LOCK_WAIT | LOCK_GAP))
 	      || ((type_mode & LOCK_MODE_MASK) != LOCK_S)
 	      || !lock_rec_other_has_expl_req(LOCK_X, 0, LOCK_WAIT,
-						FALSE, rec, trx));
+						rec, trx));
 	ut_ad((type_mode & (LOCK_WAIT | LOCK_GAP))
 	      || ((type_mode & LOCK_MODE_MASK) != LOCK_X)
 	      || !lock_rec_other_has_expl_req(LOCK_S, 0, LOCK_WAIT,
-						FALSE,rec, trx));
+						rec, trx));
 	type_mode = type_mode | LOCK_REC;
 
 	page = buf_frame_align(rec);
@@ -1794,7 +1876,6 @@ lock_rec_lock_slow(
 	dict_index_t*	index,	/* in: index of record */
 	que_thr_t* 	thr)	/* in: query thread */
 {
-	ulint	confl_mode;
 	trx_t*	trx;
 	ulint	err;
 
@@ -1802,7 +1883,6 @@ lock_rec_lock_slow(
 	ut_ad((mode == LOCK_X) || (mode == LOCK_S));
 
 	trx = thr_get_trx(thr);
-	confl_mode = lock_get_confl_mode(mode);
 
 	ut_ad((mode != LOCK_S) || lock_table_has(trx, index->table,
 								LOCK_IS));
@@ -1813,8 +1893,8 @@ lock_rec_lock_slow(
 		nothing */
 
 		err = DB_SUCCESS;
-	} else if (lock_rec_other_has_expl_req(confl_mode, 0, LOCK_WAIT,
-							FALSE, rec, trx)) {
+	} else if (lock_rec_other_has_conflicting(mode, 0, 0, rec, trx)) {
+
 		/* If another transaction has a non-gap conflicting request in
 		the queue, as this transaction does not have a lock strong
 		enough already granted on the record, we have to wait. */
@@ -1875,9 +1955,7 @@ lock_rec_lock(
 }
 
 /*************************************************************************
-Checks if a waiting record lock request still has to wait in a queue.
-NOTE that we, for simplicity, ignore the gap bits in locks, and treat
-gap type lock requests like non-gap lock requests. */
+Checks if a waiting record lock request still has to wait in a queue. */
 static
 ibool
 lock_rec_has_to_wait_in_queue(
@@ -1892,6 +1970,7 @@ lock_rec_has_to_wait_in_queue(
 
 	ut_ad(mutex_own(&kernel_mutex));
  	ut_ad(lock_get_wait(wait_lock));
+	ut_ad(lock_get_type(wait_lock) == LOCK_REC);
  	
 	space = wait_lock->un_member.rec_lock.space;
 	page_no = wait_lock->un_member.rec_lock.page_no;
@@ -1901,8 +1980,8 @@ lock_rec_has_to_wait_in_queue(
 
 	while (lock != wait_lock) {
 
-		if (lock_has_to_wait(wait_lock, lock)
-				&& lock_rec_get_nth_bit(lock, heap_no)) {
+		if (lock_rec_get_nth_bit(lock, heap_no)
+		    && lock_has_to_wait(wait_lock, lock)) {
 
 		    	return(TRUE);
 		}
@@ -1945,6 +2024,7 @@ lock_rec_cancel(
 	lock_t*	lock)	/* in: waiting record lock request */
 {
 	ut_ad(mutex_own(&kernel_mutex));
+	ut_ad(lock_get_type(lock) == LOCK_REC);
 
 	/* Reset the bit (there can be only one set bit) in the lock bitmap */
 	lock_rec_reset_nth_bit(lock, lock_rec_find_set_bit(lock));
@@ -3944,7 +4024,7 @@ lock_rec_queue_validate(
 		impl_trx = lock_clust_rec_some_has_impl(rec, index);
 
 		if (impl_trx && lock_rec_other_has_expl_req(LOCK_S, 0,
-				LOCK_WAIT, FALSE, rec, impl_trx)) {
+				LOCK_WAIT, rec, impl_trx)) {
 
 			ut_a(lock_rec_has_expl(LOCK_X, rec, impl_trx));
 		}
@@ -3959,7 +4039,7 @@ lock_rec_queue_validate(
 		impl_trx = lock_sec_rec_some_has_impl_off_kernel(rec, index);
 
 		if (impl_trx && lock_rec_other_has_expl_req(LOCK_S, 0,
-				LOCK_WAIT, FALSE, rec, impl_trx)) {
+				LOCK_WAIT, rec, impl_trx)) {
 
 			ut_a(lock_rec_has_expl(LOCK_X, rec, impl_trx));
 		}
@@ -3984,10 +4064,10 @@ lock_rec_queue_validate(
 		
 			if (lock_get_mode(lock) == LOCK_S) {
 				ut_a(!lock_rec_other_has_expl_req(LOCK_X,
-						0, 0, FALSE, rec, lock->trx));
+						0, 0, rec, lock->trx));
 			} else {
 				ut_a(!lock_rec_other_has_expl_req(LOCK_S,
-						0, 0, FALSE, rec, lock->trx));
+						0, 0, rec, lock->trx));
 			}
 
 		} else if (lock_get_wait(lock) && !lock_rec_get_gap(lock)) {
@@ -4235,11 +4315,11 @@ lock_rec_insert_check_and_lock(
 	had to wait for their insert. Both had waiting gap type lock requests
 	on the successor, which produced an unnecessary deadlock. */
 
-	if (lock_rec_other_has_expl_req(LOCK_S, LOCK_GAP, LOCK_WAIT,
-						TRUE, next_rec, trx)) {
+	if (lock_rec_other_has_conflicting(LOCK_X, LOCK_GAP,
+				LOCK_INSERT_INTENTION, next_rec, trx)) {
 		err = lock_rec_enqueue_waiting(LOCK_X | LOCK_GAP
-					| LOCK_INSERT_INTENTION, next_rec,
-								index, thr);
+						| LOCK_INSERT_INTENTION,
+						next_rec, index, thr);
 	} else {
 		err = DB_SUCCESS;
 	}