Bug #18283 When InnoDB returns error 'lock table full', MySQL can write to binlog too much
Submitted: 16 Mar 2006 15:48 Modified: 22 Mar 2006 17:36
Reporter: Heikki Tuuri Email Updates:
Status: Closed Impact on me:
Category:MySQL Server: InnoDB storage engine Severity:S3 (Non-critical)
Version:4.1, 5.0, 5.1 OS:Any (All)
Assigned to: Elliot Murphy CPU Architecture:Any

[16 Mar 2006 15:48] Heikki Tuuri
InnoDB rolls back the whole transaction in HA_ERR_LOCK_TABLE_FULL, but forgot to tell that to MySQL.

Since InnoDB stores locks in very small space, it is extremely rare that a user encounters this bug.

I am assigning this to Elliot Murphy because the patch has to be applied to ha_innodb.cc.

How to repeat:
Run a VERY big SELECT ... FOR UPDATE with a 8 MB buffer pool.

Suggested fix:
This should be fixed in 4.1, 5.0, 5.1:

--- ha_innodb.cc        2006-03-16 17:42:57.305108104 +0200
+++ ha_innodb_fixed.cc  2006-03-16 17:45:37.743717728 +0200
@@ -513,6 +513,13 @@

        } else if (error == (int) DB_LOCK_TABLE_FULL) {
+               /* Since we rolled back the whole transaction, we must
+               tell it also to MySQL so that MySQL knows to empty the
+               cached binlog for this transaction */
+               if (thd) {
+                       ha_rollback(thd);
+               }

        } else {
[17 Mar 2006 9:25] Bugs System
A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

[22 Mar 2006 11:40] Elliot Murphy
Fix has been pushed for 5.1.9 and 5.0.20
[22 Mar 2006 17:36] Mike Hillyer
Noted in changelog:

          If InnoDB encountered a
          <literal>HA_ERR_LOCK_TABLE_FULL</literal> error and
          rolled-back a transaction, the transaction was still written
          to the binary log. (Bug #18283)