Bug #73723 innodb.log_file_size fails with wrong errno 2006 ob pb2(sporadic)
Submitted: 26 Aug 2014 7:58 Modified: 28 Aug 2014 18:30
Reporter: Vinay Fisrekar Email Updates:
Status: Closed Impact on me:
None 
Category:MySQL Server: InnoDB storage engine Severity:S3 (Non-critical)
Version:5.7 OS:Any
Assigned to: CPU Architecture:Any

[26 Aug 2014 7:58] Vinay Fisrekar
Description:
innodb.log_file_size fails sporadically with wrong errno 2006: 'MySQL server has gone away', instead of 2013...
on pb2

As per mail thread in regression monitoring , its being followed up with Marko.

http://pb2.no.oracle.com/?template=mysql_show_test_failure&search=yes&push_id=7004959&test...

innodb.log_file_size                     w1 [ fail ]
        Test ended at 2014-08-26 05:30:44

CURRENT_TEST: innodb.log_file_size
mysqltest: At line 29: query 'COMMIT' failed with wrong errno 2006: 'MySQL server has gone away', instead of 2013...

The result from queries just before the failure was:
call mtr.add_suppression("InnoDB: Resizing redo log");
call mtr.add_suppression("InnoDB: Starting to delete and rewrite log files");
call mtr.add_suppression("InnoDB: New log files created");
call mtr.add_suppression("InnoDB: The log sequence numbers [0-9]+ and [0-9]+ in ibdata files do not match the log sequence number [0-9]+ in the ib_logfiles");
CREATE TABLE t1(a INT PRIMARY KEY) ENGINE=InnoDB;
BEGIN;
INSERT INTO t1 VALUES (42);

 - the logfile can be found in '/tmp/mtr-29452/var-n_mix_8k/log/innodb.log_file_size/log_file_size.log'

How to repeat:
command from pb2:
perl mysql-test-run.pl --timer --force --big-test --testcase-timeout=60 --debug-server --parallel=auto --comment=n_mix_8k_size --vardir=/tmp/mtr-29452/var-n_mix_8k --mysqld=--binlog-format=mixed --experimental=collections/default.experimental --skip-test-list=collections/disabled-daily.list --mysqld=--innodb-page-size=8k --skip-test=innodb_ignore_builtin --suite=innodb --parallel=8
[28 Aug 2014 7:45] Marko Mäkelä
Posted by developer:
 
I think that this was simply a badly written test (sorry about that).
After the server is killed, several different error codes can be returned to the client.

I eliminated almost all direct use of shutdown_server in MySQL tests, using one of the include files instead.
This should fix the innodb.log_file_size failure as well.