Percona Server with XtraDB

federated.federated_archive, federated.federated fails in valgrind 32bit

Reported by Stewart Smith on 2011-06-09
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Percona Server
High
Unassigned
5.1
High
Unassigned
5.5
High
Unassigned

Bug Description

http://jenkins.percona.com/job/percona-server-5.5-valgrind/BUILD_TYPE=valgrind,Host=ubuntu-natty-32bit/3/consoleFull

federated.federated_archive w2 [ fail ] Found warnings/errors in server log file!
        Test ended at 2011-06-09 04:21:24
line
==9772== Uninitialised byte(s) found during client check request
==9772== at 0x8525ABB: buf_page_hash_get_low (buf0buf.ic:1041)
==9772== by 0x852E065: buf_page_get_gen (buf0buf.c:3494)
==9772== by 0x84EC5DA: trx_sys_file_format_max_read (trx0sys.c:1418)
==9772== by 0x84EF5B8: trx_sys_file_format_max_check (trx0sys.c:1466)
==9772== by 0x84D7786: innobase_start_or_create_for_mysql (srv0start.c:1834)
==9772== by 0x848FCDA: innobase_init(void*) (ha_innodb.cc:2858)
==9772== by 0x82D859D: ha_initialize_handlerton(st_plugin_int*) (handler.cc:462)
==9772== by 0x81BD3C7: plugin_initialize(st_plugin_int*) (sql_plugin.cc:1063)
==9772== by 0x81C2C5A: plugin_init(int*, char**, int) (sql_plugin.cc:1349)
==9772== by 0x8132B15: init_server_components() (mysqld.cc:3962)
==9772== by 0x8136781: mysqld_main(int, char**) (mysqld.cc:4634)
==9772== by 0x812B6B6: main (main.cc:24)
==9772== Address 0x92b5e38 is not stack'd, malloc'd or (recently) free'd
==9772== Uninitialised byte(s) found during client check request
==9772== at 0x852E93D: buf_page_get_gen (buf0buf.c:3774)
==9772== by 0x84EC5DA: trx_sys_file_format_max_read (trx0sys.c:1418)
==9772== by 0x84EF5B8: trx_sys_file_format_max_check (trx0sys.c:1466)
==9772== by 0x84D7786: innobase_start_or_create_for_mysql (srv0start.c:1834)
==9772== by 0x848FCDA: innobase_init(void*) (ha_innodb.cc:2858)
==9772== by 0x82D859D: ha_initialize_handlerton(st_plugin_int*) (handler.cc:462)
==9772== by 0x81BD3C7: plugin_initialize(st_plugin_int*) (sql_plugin.cc:1063)
==9772== by 0x81C2C5A: plugin_init(int*, char**, int) (sql_plugin.cc:1349)
==9772== by 0x8132B15: init_server_components() (mysqld.cc:3962)
==9772== by 0x8136781: mysqld_main(int, char**) (mysqld.cc:4634)
==9772== by 0x812B6B6: main (main.cc:24)
==9772== Address 0x92b5e38 is not stack'd, malloc'd or (recently) free'd

Stewart Smith (stewart) wrote :

federated.federated looks to fail the same way.

summary: - federated.federated_archive fails in valgrind 32bit
+ federated.federated_archive, federated.federated fails in valgrind 32bit

Like for bug 794837, we need "mysql-5.5-trunk-valgrind" setup in Jenkins to find out if this is ours or upstream.

The bug is repeatable with -DHAVE_purify too.

The unitialized address is of bpage.flush_list in buf_page_hash_get_low. I have verified that this is not due struct padding.

bpage.list of InnoDB-Plugin is split into each dedicated list (UT_LIST_NODE_T)
bpage.free , bpage.flush_list and bpage.zip_list for more optimization.

So, at the case of "bpage.list", free list operation at buf_chunk_init() initialize the "bpage.list".
But, at split case, only "bpage.free" is initialized.

-----------------------
block->page.free.prev = NULL;
block->page.free.next = NULL;
block->page.zip_list.prev = NULL;
block->page.zip_list.next = NULL;
------------------------
I think, the lines should be added to buf_block_init() which called from buf_chunk_init()
to avoid complaining from valgrind/purify

I will fix soon

Oops
block->page.flush_list.prev = NULL;
block->page.flush_list.next = NULL;
block->page.zip_list.prev = NULL;
block->page.zip_list.next = NULL;

But I never think the bug is Critical.
The uninitialized values are never used.

If the bug fixed by the branch, I should fix also for 5.1
please inform the result

On Thu, 23 Jun 2011 07:45:22 -0000, Yasufumi Kinoshita <email address hidden> wrote:
> If the bug fixed by the branch, I should fix also for 5.1
> please inform the result

I've just set a valgrind run going in Jenkins on it:

http://jenkins.percona.com/view/Percona%20Server%205.5/job/percona-server-5.5-valgrind-param/2/

--
Stewart Smith

Stewart Smith (stewart) wrote :

It appears to be fixed for 5.5 with this branch, so I'll merge it for 5.5. Have also targeted to 5.1 so we can track 5.1 there too.

Thanks!

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers