Bug #73928 JET replicationRowDataTypeConversionTest fails in 5.6.21 and trunk

• Submitted: 16 Sep 2014 10:22
• Modified: 9 Jan 2015 7:38
• Reporter: Jorge Jordao
• Status: Closed
• Category: MySQL Server: Replication
• Severity: S2 (Serious)
• Version: 5.6.21
• OS: Any
• CPU Architecture: Any

[16 Sep 2014 10:22] Jorge Jordao

Description:
The replicationRowDataTypeConversionTest runs in a standard master->slave topology with binlog-format=row.

Iterating over a list of increasingly large data types, it 
- Creates tables in the master with the datatype (ex: VARCHAR(4096))
- Syncs slave with master
- Recreates tables in slave with equal or larger datatype (ex: VARCHAR(4096), TEXT, MEDIUMTEXT, LONGTEXT)
- In master, inserts rows with the length 1, maximum length and various in-between random lengths
- Syncs slave with master and verifies that the converted values are equal in both servers

This test works fine up to 5.6.20. However, from 5.6.21 the slave server dies with mysqld exit code 139, presumably due to "too big row".

How to repeat:
Run JET replicationRowDataTypeConversionTest.xml with 5.6.21 or later 5.6 build.

This test is executed daily as part of the jet-rpl-daily-5.6-5.6 Hudson project: http://fimafeng11.no.oracle.com:8080/view/jet-rpl/job/jet-rpl-daily-5.6-5.6/

[9 Jan 2015 7:38] Erlend Dahl

Fixed in 5.6.22 and 5.7.6 under the heading of

Bug#19704825 TEMPORARY SLAVE TYPE CONVERSION TABLES RELEASED TO EARLY