MySQL Bugs: #29102: ndb_dd_sql_features fails in pushbuild

Bug #29102	ndb_dd_sql_features fails in pushbuild
Submitted:	14 Jun 2007 10:37	Modified:	17 Sep 2007 10:34
Reporter:	Mads Martin Joergensen	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Cluster: Cluster (NDB) storage engine	Severity:	S1 (Critical)
Version:	5.1	OS:	Any
Assigned to:	Guangbao Ni	CPU Architecture:	Any
Tags:	pbfail

Description:
Hi!  I notice an intermittent test failure of ndb_dd_sql_features, which
seems to have first shown up with this push in December (by serge into
5.1-ndb):

https://intranet.mysql.com/secure/pushbuild/showpush.pl?dir=mysql-5.1-new-ndb&order=231

That push shouldn't have caused it, though, it just changes another test
case file.

The failure looks like:

mysqltest: At line 544: query 'DROP TABLESPACE ts2 ENGINE=NDB' failed:
1517: Failed to drop TABLESPACE
(the last lines may be the most important ones)

It's shown up a couple of times in the current 5.1-maint tree, which is
why I'm looking at it.

https://intranet.mysql.com/secure/pushbuild/xref.pl?startdate=&enddate=&dir=&plat=&testtyp...

How to repeat:
https://intranet.mysql.com/secure/pushbuild/xref.pl?testname=ndb_dd_sql_features

Pushed into 5.1.20-beta

ndb_dd_sql_features            [ fail ]

Errors are (from /dev/shm/var-n_stm-130/log/mysqltest-time) :
mysqltest: Result content mismatch
(the last lines may be the most important ones)
Below are the diffs between actual and expected results:
-------------------------------------------------------
*** r/ndb_dd_sql_features.result	Wed Aug  1 18:06:15 2007
--- r/ndb_dd_sql_features.reject	Wed Aug  1 19:08:18 2007
***************
*** 157,163 ****
  SELECT a1,length(a2),substr(a2,4998,2),length(a3),substr(a3,49997,3)
  FROM test.t1 WHERE a1=1;
  a1	length(a2)	substr(a2,4998,2)	length(a3)	substr(a3,49997,3)
! 1	5000	dd	50000	555
  UPDATE test.t1 SET a2=@vc4, a3=@bb4 WHERE a1=2;
  SELECT a1,length(a2),substr(a2,3998,2),length(a3),substr(a3,39997,3)
  FROM test.t1 WHERE a1=2;
--- 157,163 ----
  SELECT a1,length(a2),substr(a2,4998,2),length(a3),substr(a3,49997,3)
  FROM test.t1 WHERE a1=1;
  a1	length(a2)	substr(a2,4998,2)	length(a3)	substr(a3,49997,3)
! 1	5000	dd	50000	XXX
  UPDATE test.t1 SET a2=@vc4, a3=@bb4 WHERE a1=2;
  SELECT a1,length(a2),substr(a2,3998,2),length(a3),substr(a3,39997,3)
  FROM test.t1 WHERE a1=2;
-------------------------------------------------------

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/33380

ChangeSet@1.2476, 2007-08-29 21:45:53+03:00, pekka@sama.ndb.mysql.com +1 -0
  ndb - bug#29102 : use locked read even for blob parts

Documented bugfix in mysql-5.1.22-ndb-6.2.5 changelog as:

<literal>DROP TABLESPACE</literal> failed when a table that had been using the tablespace contained a <literal>BLOB</literal> column.

Left bug report in PP status pending mainline merges.

merge into mysql-5.1.22-ndb-6.2.5

Pushed into 5.0.50

Pushed into 5.1.23-beta

Documented fix in 5.1.23 changelog.

Why was a Disk Data bugfix pushed to 5.0?

Hi Jon,

  No, It isn't only for Disk Data. The real problem is that unlock reading can get the real data of Blob part commited last time.
  So, this patch is to use locked read even for blob parts.

    race condition : s/committedRead/readTuple/ when reading parts since TUP commits tuples separately in NdbBlob.cpp

/Guangbao

Set back to Documenting status so I won't forget to fix this... :) Thanks!

Thank you for your bug report. This issue has been committed to our source repository of that product and will be incorporated into the next release.

If necessary, you can access the source repository and build the latest available version, including the bug fix. More information about accessing the source trees is available at

    http://dev.mysql.com/doc/en/installing-source.html

Documented with new description in mysql-5.1.22-ndb-6.2.5, 5.1.23, and 5.0.50 changelogs:

  Reads on BLOB columns were not locked when they needed to be, in order to 
  guarantee consistency.

Also reverted Category from ClusterDD to Cluster per note from GNi.