Bug#54583 InnoDB: Assertion failure sync/sync0sync.c:1226

      
Silence the UNIV_SYNC_DEBUG assertion failure while upgrading old data files
to multiple rollback segments during server startup. Because the upgrade
takes place while InnoDB is running a single thread, we can safely ignore the
latching order checks without fearing deadlocks.
      
innobase_start_or_create_for_mysql(): Set srv_is_being_started = FALSE
only after trx_sys_create_rsegs() has completed.
      
sync_thread_add_level(): If srv_is_being_started, ignore latching order
violations for SYNC_TRX_SYS_HEADER and SYNC_IBUF_BITMAP.

Create all the non-IO threads after creating the extra rollback segments.

Patch originally from Marko with some additions by Sunny.

1 files changed, 21 insertions, 14 deletions

diff --git a/storage/innobase/srv/srv0start.c b/storage/innobase/srv/srv0start.c
index 3d4aaa9d5d4..4da836672ec 100644
--- a/storage/innobase/srv/srv0start.c
+++ b/storage/innobase/srv/srv0start.c
@@ -1703,20 +1703,6 @@ innobase_start_or_create_for_mysql(void)
 	/* fprintf(stderr, "Max allowed record size %lu\n",
 	page_get_free_space_of_empty() / 2); */
 
-	/* Create the thread which watches the timeouts for lock waits */
-	os_thread_create(&srv_lock_timeout_thread, NULL,
-			 thread_ids + 2 + SRV_MAX_N_IO_THREADS);
-
-	/* Create the thread which warns of long semaphore waits */
-	os_thread_create(&srv_error_monitor_thread, NULL,
-			 thread_ids + 3 + SRV_MAX_N_IO_THREADS);
-
-	/* Create the thread which prints InnoDB monitor info */
-	os_thread_create(&srv_monitor_thread, NULL,
-			 thread_ids + 4 + SRV_MAX_N_IO_THREADS);
-
-	srv_is_being_started = FALSE;
-
 	if (trx_doublewrite == NULL) {
 		/* Create the doublewrite buffer to a new tablespace */
 
@@ -1729,8 +1715,29 @@ innobase_start_or_create_for_mysql(void)
 	We create the new segments only if it's a new database or
 	the database was shutdown cleanly. */
 
+	/* Note: When creating the extra rollback segments during an upgrade
+	we violate the latching order, even if the change buffer is empty.
+	We make an exception in sync0sync.c and check srv_is_being_started
+	for that violation. It cannot create a deadlock because we are still
+	running in single threaded mode essentially. Only the IO threads
+	should be running at this stage. */
+
 	trx_sys_create_rsegs(TRX_SYS_N_RSEGS - 1);
 
+	/* Create the thread which watches the timeouts for lock waits */
+	os_thread_create(&srv_lock_timeout_thread, NULL,
+			 thread_ids + 2 + SRV_MAX_N_IO_THREADS);
+
+	/* Create the thread which warns of long semaphore waits */
+	os_thread_create(&srv_error_monitor_thread, NULL,
+			 thread_ids + 3 + SRV_MAX_N_IO_THREADS);
+
+	/* Create the thread which prints InnoDB monitor info */
+	os_thread_create(&srv_monitor_thread, NULL,
+			 thread_ids + 4 + SRV_MAX_N_IO_THREADS);
+
+	srv_is_being_started = FALSE;
+
 	err = dict_create_or_check_foreign_constraint_tables();
 
 	if (err != DB_SUCCESS) {