MySQL Bugs: #31482: Cluster table with blob field fails if too many result rows are to be returned

Bug #31482	Cluster table with blob field fails if too many result rows are to be returned
Submitted:	9 Oct 2007 14:26	Modified:	5 Nov 2007 21:31
Reporter:	Hartmut Holzgraefe	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Cluster: Cluster (NDB) storage engine	Severity:	S2 (Serious)
Version:	5.1.22-ndb-6.3.2 and up	OS:	Linux
Assigned to:	Jonas Oreland	CPU Architecture:	Any

Description:
On a table with blob contents any query that would return more than a certain ammount of rows fails with 

  query 'SELECT * FROM t1' failed: 1296: Got error 4350 'Transaction already aborted' from NDBCLUSTER

in the attached test case it starts to fail somewhere between 17,000 and 18,000 result rows, the number of rows differs with the ammount of data stored in the blob field though. The smaller the blob contents the more rows are needed to trigger the problem

there is nothing in the mysqld error log, cluster log and data node error logs about this, the ndb_#_out.log files get lots of information dumped to then when this happens though

How to repeat:
see attached test case

mysqltest test case and config.ini (warning: large result file)

Attachment: bug31482.tgz (application/x-gtar, text), 394.90 KiB.

this was introduced by fix for bug#29102
(which changes read of blob table to use locks, instead of dirty read)

this has the unfortunate side effect that operation records does not get released
in tc, making it to run out...

hmmm...we could "reinvent" simple read, which reads latest committed with lock but releases lock afterwards...

however, the "send tckeyconf"-optimization from LQH does not work (which is why it was removed in the first place) as LQH has no timeout handling.

/Jonas

have patch that fixes problem...
now only need to fix testing...

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:

  http://lists.mysql.com/commits/35543

ChangeSet@1.2574, 2007-10-15 09:10:14+02:00, jonas@perch.ndb.mysql.com +17 -0
  ndb - bug#31482
    (re)impl. simple-read (read that releases lock just before LQHKEYCONF)
    use simple-read for blobs

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:

  http://lists.mysql.com/commits/35548

ChangeSet@1.2662, 2007-10-15 11:26:11+02:00, jonas@perch.ndb.mysql.com +2 -0
  ndb - 51-ndb -> 6.2 post merge bug#31482
    also fix NdbRecord code path

Documented bugfix in mysql-5.1-ndb-6.3.5 changelog; left Patch Pending status.

pushed to 51-ndb, telco-6.2, telco-6.3 & 51-telco

Pushed into 6.0.4-alpha

Pushed into 5.1.23-rc

Also documented in 5.1.23 and 6.0.4 changelogs. Closed.