Bug #293 | Memory leaks? | ||
---|---|---|---|
Submitted: | 15 Apr 2003 10:56 | Modified: | 12 Jun 2003 5:47 |
Reporter: | chris loelke | Email Updates: | |
Status: | Closed | Impact on me: | |
Category: | MySQL Server | Severity: | S2 (Serious) |
Version: | 4.0.12-debug-log | OS: | Other (SCO Openserver 5.0.6) |
Assigned to: | CPU Architecture: | Any |
[15 Apr 2003 10:56]
chris loelke
[16 Apr 2003 10:42]
Indrek Siitan
Do you have an SQL script with a repeatable test case that, ran against the database server, generates those errors?
[16 Apr 2003 10:49]
chris loelke
in order to reproduce this all one has to do is run the perl script typically in usr/local/mysql/mysql-test name of the script is mysql-test-run. This is the standard script that comes with mysql distributions. It fails on the very first test (Alias).
[16 Apr 2003 10:51]
MySQL Verification Team
We test thoroughly all our builds with all tests we have. Therefore this is an error in building MySQL. Use our binary instead.
[16 Apr 2003 10:54]
chris loelke
you dont have a binary for this platform available, the one that is sometimes available has memory leaks too.
[16 Apr 2003 11:00]
MySQL Verification Team
If our binary issues memory errors that it is due to differences in patch levels between our OS and yours.
[16 May 2003 2:15]
Michael Widenius
I must first appoligize for the way we have handled this bug report in the past. It was unprofessionally done and I have talked with the developer that handled this case to ensure that this will not happen again. Back to your problem: MySQL is known to work on OpenServer 5.0.6 with your gcc version, so we need some more information to know what is wrong. I we could get a trace file from just one test that fails, then we should be able to figure out what is wrong. Try to do this: cd mysql-test ./mysql-test-run --local --debug alias if this fails with 'Freeing wrong aligned pointer' then please do the following: - Create a README file with contains a copy of the text in this ticket - Create a tar file that contains the README file and the file var/tests/master.trace - Upload this to ftp://support.mysql.com/pub/mysql/secret - Add a new comment to this bug that contains the name of the file If we can't solve it from the above, the other option is if you can give us access to a system where this happens so that we can login to it and debug this ourselves.
[21 May 2003 8:09]
chris loelke
file sco506bug.tar uploaded. the test fails with 'Freeing wrong aligned pointer' and it also never completes.
[22 May 2003 8:47]
MySQL Verification Team
Sorry for the previous comments. Can you please send us more info. We need to track a problem. One way is to run the following command which should give us the information we need: ./mysql-test-run --debug alias and then, please,upload the var/log/master.trace file. If you have any further questions, you can use this record to post them. Thanks in advance.
[22 May 2003 8:51]
MySQL Verification Team
I see that you have uploaded a file. We shall soon take a look at it and we shall advise you of our findings.
[27 May 2003 8:23]
MySQL Verification Team
Seems like the error is just in the pointer alignment. Can you please try changing line 639 in my_global.h into: #define ALIGN_SIZE(A) MY_ALIGN((A),4) and see if you still get the debug warnings
[27 May 2003 9:27]
chris loelke
Yes, that fixed the problem with the pointer alignment, very good call :) FYI, the sizeof(double) is 8 on SCO... Next issue.. (hopefully the last), when running all the tests I still get a failure as per below. constraints 0.000000 0.010000 0.198983 [ pass ] count_distinct 0.010000 0.050000 0.479109 [ pass ] count_distinct2 0.770000 1.820000 1.105322 [ pass ] create 0.020000 0.020000 1.122247 [ pass ] ERROR: /usr/mysql/mysql-4.0.12/mysql-test/var/run/master.pid was not created in 30 seconds; Aborting #
[27 May 2003 10:08]
MySQL Verification Team
This is completely different issue. Check messages in master error log. See whether it happens always in the same test.and whether it is in create or ctype-collate. Then try running that test alone, like: cd mysql-test ./mysql-test-run --local ctype-collate
[2 Jun 2003 11:03]
MySQL Verification Team
This problem was diagnosed. We need to have ALIGN_SIZE as a config variable and have to write a test for configure to determine it's value, as some OS's, like SCO align pointers at 4 and not at the sizeof(double) alignment.
[12 Jun 2003 5:47]
Michael Widenius
Thank you for your bug report. This issue has been fixed in the latest development tree for that product. You can find more information about accessing our development trees at http://www.mysql.com/doc/en/Installing_source_tree.html This will be fixed in 4.0.14 The bug was that one structure that was used by safemalloc was not 8 byte aligned with the compiler you used.