[MDEV-6600] Some GROUP BY commands fail with -DUSER_ARIA_FOR_TMP_TABLES=OFF Created: 2014-08-18 Updated: 2015-01-19 Due: 2014-10-24 Resolved: 2015-01-17 |
|
| Status: | Closed |
| Project: | MariaDB Server |
| Component/s: | Compiling, Optimizer |
| Affects Version/s: | 5.5.37, 5.5.38, 5.5.39 |
| Fix Version/s: | 10.0.6 |
| Type: | Bug | Priority: | Minor |
| Reporter: | Weldon Whipple | Assignee: | Elena Stepanova |
| Resolution: | Fixed | Votes: | 0 |
| Labels: | None | ||
| Environment: |
CentOS 6.5 x-86-64 running MariaDB if cmake uses options -DUSER_ARIA_FOR_TMP_TABLES=OFF and -DWITH_ARIA_STORAGE_ENGINE=ON. Noticed when running Joomla application. |
||
| Description |
|
A customer's query (see mysql-test/suite/betterlinux/t/bl_aria_group.test) that consists of nested joins and a GROUP BY clause fails and displays this message: "failed: 126: Incorrect key file for table '/var/tmp/#sql_5b8d_1'; try to repair it". If both Aria options are defined when invoking cmake, the failure doesn't occur. Run test case betterlinux.bl_aria_group.test (in mariadb-bug-aria-joomla-GROUP_BY.tgz uploaded to ftp.askmonty.org) with no ARIA-related definisions, and the test case runs successfully. Repeat with cmake options -DUSER_ARIA_FOR_TMP_TABLES=OFF and -DWITH_ARIA_STORAGE_ENGINE=ON and the test case fails. |
| Comments |
| Comment by Weldon Whipple [ 2014-08-18 ] | ||||||||
|
Failure occurs in (or beneath) method ha_myisam::open (in storage/my_isam/ha_myisam.cc) in call to mi_open on line 743, which returns (MI_INFO *) 0x0 | ||||||||
| Comment by Weldon Whipple [ 2014-08-18 ] | ||||||||
|
I suggested that the customer not specify -DUSER_ARIA_FOR_TMP_TABLES=OFF -DWITH_ARIA_STORAGE_ENGINE=ON as cmake options. Might a possible "fix" be to disallow those options? | ||||||||
| Comment by Elena Stepanova [ 2014-08-22 ] | ||||||||
|
Hi, Do you have any other cnf/opt files in your betterlinux suite, generic ones that are applied to all tests? | ||||||||
| Comment by Elena Stepanova [ 2014-09-27 ] | ||||||||
|
It turned out that the test configs are not important. Here is the reduced version of the test case:
The problem is reproducible on current 5.5 tree:
But it seems to work all right on 10.0 tree. Since the problem is not critical, and it doesn't show up in the latest stable release, I don't think it's necessary to fix it in 5.5 tree. Please comment if you disagree, otherwise I'm closing it as fixed in 10.0 (I will set 10.0.6 because it's the earliest available release on the list, even though it was apparently fixed long before that). | ||||||||
| Comment by Weldon Whipple [ 2014-09-27 ] | ||||||||
|
Thank you, Elena! Weldon From my iPhone | ||||||||
| Comment by Weldon Whipple [ 2014-10-10 ] | ||||||||
|
Thank you very much for responding to JIRA I had another discussion with the ISP that was having a problem with I didn't realize when I opened the JIRA bug that a priority of Minor would Here is the ISP's problem: They have hundreds/thousands of shared servers, each with hundreds of MySQL Unfortunately, run-time options for disabling Aria on an executable that is We've tried (for example) skip-aria as hinted by the MariaDB documentation. Unfortunately, all are mysqld-killing syntax errors. (The third one is a We would love to have them use MariaDB 5.5.3x, but can't until this is I asked the ISP why they don't just completely disable Aria via CMAKE If I had more time, I might try to fix the code myself. (Perhaps just ww On Sat, Sep 27, 2014 at 4:52 AM, Elena Stepanova (JIRA) < | ||||||||
| Comment by Elena Stepanova [ 2014-10-10 ] | ||||||||
|
Hi Weldon, So, the problem happens when the ISP chooses to use Aria, but NOT for internal temporary tables.
First, choice of a storage engine for temporary tables is pretty much transparent for users, they normally don't even know which one is it. I have a feeling that the ISP gets it all wrong or can't explain properly what they want to achieve, because the description is rather confusing. Even if Aria is present, it does not mean that it's the default storage engine for user tables, applications won't be using it (except for internal tmp tables) unless the users specifically make them to. All these contradictions were the reason I decided the problem was not critical. If they have a proper explanation, the decision whether or not fix it in 5.5 can be reconsidered. | ||||||||
| Comment by roberto spadim [ 2014-10-17 ] | ||||||||
|
when i moved some apps from myisam to aria, insert delayed stoped working, i don't know if myisam or mariadb support delayed anymore, that was the only feature that aria don't have (i think), the other one is some a bug with query cache that dba must leave aria running without query cache in a production server | ||||||||
| Comment by Sergei Golubchik [ 2015-01-17 ] | ||||||||
|
Closed, as per above. |