So, you downgraded both servers where the problem was observed, and you are not getting the crash anymore?
If so, it's going to be hard to reproduce it.

If you still have a server where it happens, please enable the general log for a while (until the next crash). After the server crashes, store both the general log and error log, this way we'll be able to find the guilty query.

I tried reproducing it on a similar setup running CentOS 6, but no dice. I can't reproduce it on any other setup so far. I'm afraid the affected servers are blocked for the next 1-2 months and I can't touch them during that period. That's why I was hoping to be able to reproduce it elsewhere :/

I can say however that the query itself didn't really matter. Any kind of SELECT query on the oqgraph table could cause the crash, from a simple "SELECT * FROM table" to a dijkstra shortest path query, just not every time.

Closing for now as not reproducible. Please comment to re-open if you encounter the problem again.

Managed to finally reproduce this again, error and general log attached.
This now happened on a single server setup running 10.0.21 on CentOS 6.

Got a debug version of mariadb installed on the affected server and managed to reproduce the crash. General log and error log attached as well, should contain more useful information now.

pprkut (on Freenode IRC) has still this issue and think he can reproduce it)

pprkut, thanks.
Can you paste the structure of 'version_history' (and underlying tables if there are any)? Even better if you can provide the data dump.
Please also attach your cnf file.
Thanks.

mysqld: /opt/mariadb-10.0.21/mariadb-10.0.21/sql/handler.cc:2473: THD* handler::ha_thd() const: Assertion `!table || !table->in_use || table->in_use == _current_thd()' failed.

150925 15:07:07 [ERROR] mysqld got signal 6 ;

sql/handler.cc:2474(handler::ha_thd() const)[0x861bcc]

handler/ha_innodb.cc:13024(ha_innobase::extra(ha_extra_function))[0xa1b01b]

/opt/mariadb-10.0.21/mariadb-10.0.21/sql/lib/plugin/ha_oqgraph.so(_ZN10ha_oqgraph5extraE17ha_extra_function+0x3b)[0x7ff5b193675d]

sql/table_cache.cc:342(tc_acquire_table)[0x7ead7c]

sql/table_cache.cc:765(tdc_acquire_share(THD*, char const*, char const*, char const*, unsigned int, unsigned int, unsigned int, TABLE**))[0x7ebc6b]

sql/sql_base.cc:2371(open_table(THD*, TABLE_LIST*, st_mem_root*, Open_table_context*))[0x603792]

sql/sql_base.cc:4032(open_and_process_table)[0x606441]

sql/sql_base.cc:4566(open_tables(THD*, TABLE_LIST**, unsigned int*, unsigned int, Prelocking_strategy*))[0x607552]

sql/sql_base.cc:5209(open_and_lock_tables(THD*, TABLE_LIST*, bool, unsigned int, Prelocking_strategy*))[0x608640]

sql/sql_base.h:490(open_and_lock_tables(THD*, TABLE_LIST*, bool, unsigned int))[0x5fc65e]

sql/sql_parse.cc:5219(execute_sqlcom_select)[0x66d748]

sql/sql_parse.cc:2562(mysql_execute_command(THD*))[0x6659b1]

sql/sql_parse.cc:6531(mysql_parse(THD*, char*, unsigned int, Parser_state*))[0x6705d6]

sql/sql_parse.cc:1310(dispatch_command(enum_server_command, THD*, char*, unsigned int))[0x662b43]

sql/sql_parse.cc:999(do_command(THD*))[0x661dc9]

sql/sql_connect.cc:1378(do_handle_one_connection(THD*))[0x784376]

sql/sql_connect.cc:1294(handle_one_connection)[0x7840d5]

/lib64/libpthread.so.0(+0x7a51)[0x7ff5b3475a51]

/lib64/libc.so.6(clone+0x6d)[0x7ff5b1e5893d]

Trying to get some variables.

Some pointers may be invalid and cause the dump to abort.

Query (0x7ff584823088): SELECT COUNT(*) FROM version_history

Connection ID (thread ID): 5

Status: NOT_KILLED

Uploaded the files as requested. Please let me know if you need anything else.

Just a punt, but the following patch might work

diff --git a/storage/oqgraph/ha_oqgraph.cc b/storage/oqgraph/ha_oqgraph.cc
index 89bc912..b4cf03c 100644
— a/storage/oqgraph/ha_oqgraph.cc
+++ b/storage/oqgraph/ha_oqgraph.cc
@@ -1099,6 +1099,10 @@ int ha_oqgraph::info(uint flag)

int ha_oqgraph::extra(enum ha_extra_function operation)
{
+ if (graph->get_thd() != current_thd)

return edges->file->extra(operation);
}

If it does I'll submit a pull request

I tried the patch with 10.0.22 and it looks to be working. With my testcase I get a bunch of crashes without the patch, whereas it runs smoothly and stable with the patch

andymc73, see above – please do submit the pull request if you haven't done it yet.

Hi, elenst I just made the pull requests

https://github.com/MariaDB/server/pull/152 against 10.1 branch
https://github.com/MariaDB/server/pull/153 against 10.0 branch

Note I also have an open pull request with additional test cases for previous issues:
https://github.com/MariaDB/server/pull/18
which I also checked just and it still seems to merge OK

andymc73, shall I merge now? I see there are some comments about the patch on github. I'll wait until you confirm that it's ok to merge. If it'll be this week, the patch will make it into 10.0.24.

Sorry this got lost ... I just pushed a commit that address the comment against https://github.com/MariaDB/server/pull/153 so that should be good to go to finally resolve this off...

https://github.com/MariaDB/server/pull/152/commits/7c968ccc635a13b30210f09b7d57dc1852e26801

https://github.com/MariaDB/server/pull/153/commits/19b132327e6220176b562a4e1533345b275dc409

I noticed other uses of current_thd but perhaps that can be made a separate issue if there is a performance problem, so this crash bug can be marked resolved