Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Cannot Reproduce
-
10.2(EOL)
-
None
Description
InnoDB tries to open file on read-only mode but some external tool (in this case ib-utils) also has opened it as read-write.
innodb.innodb-wl5522-debug-zip 'innodb' [ fail ] Found warnings/errors in server log file!
|
Test ended at 2016-12-19 10:38:46
|
line
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: Operating system error number 2 in a file operation.
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: The error means the system cannot find the path specified.
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: If you are installing InnoDB, remember that you must create directories yourself, InnoDB does not create them.
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: Cannot open datafile for read-only: './test_wl5522/t1.ibd' OS error: 71
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: Operating system error number 2 in a file operation.
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: The error means the system cannot find the path specified.
|
2016-12-19 10:38:35 139864375539584 [ERROR] InnoDB: If you are installing InnoDB, remember that you must create directories yourself, InnoDB does not create them.
|
2016-12-19 10:38:35 139864375539584 [Warning] InnoDB: Ignoring tablespace `test_wl5522/t1` because it could not be opened.
|
^ Found warnings in /dev/shm/log/mysqld.1.err
|
Attachments
Activity
The Description looks like a misunderstanding. No source code revision was stated, but based on the timestamp, this was slightly before the mariadb-10.2.3 release. The test innodb.innodb-wl5522-debug-zip had not been modified between 10.2.2 and 10.2.3, but there were some changes to the suppressions in 10.2.4. In MDEV-11927 (10.2.5) the test was partly replaced with innodb_zip.wl5522_debug_zip.
I believe that the “ib-utils” refers to sections of the test like this:
FLUSH TABLES t1 FOR EXPORT; |
|
|
perl;
|
do "$ENV{MTR_SUITE_DIR}/../innodb/include/innodb-util.pl"; |
ib_backup_tablespaces("test", "t1"); |
EOF
|
UNLOCK TABLES;
|
There is no race condition here, because the file is being accessed between FLUSH TABLES…FOR EXPORT and UNLOCK TABLES.
It looks like the problem simply was that the call mtr.add_suppression() had not been adjusted to the changed error messages:
git grep -B1 'The error means' mariadb-10.2.3 -- storage/innobase/ mysql-test/suite/innodb/t/innodb-wl5522-debug-zip.test
|
This will find several ib::error() calls with that string but no suppression in the test. (There are suppressions in the test innodb-wl5522-debug.)
I suspect that this was fixed in the 10.2.5 release.
It would be useful to see more of the server error log, to determine at which point of the test the message was issued.
I renamed the test to innodb_zip.wl5522_debug_zip and did some minor cleanup on 2017-02-24.
It is possible that my DBUG_DEBUG fix reduced the I/O workload enough so that the race condition is now harder to repeat.