[slurm-users] slurmdbd crashes with segmentation fault following DBD_GET_ASSOCS

[Steven Dick]

Latest releases of slurm (17-20) don't work with mysql 5.7.30,
Latest version of mariadb works fine.


On Tue, May 5, 2020 at 3:41 PM Dustin Lang <dstndstn at gmail.com> wrote:
>
> I tried upgrading Slurm to 18.08.9 and I am still getting this Segmentation Fault!
>
>
>
> On Tue, May 5, 2020 at 2:39 PM Dustin Lang <dstndstn at gmail.com> wrote:
>>
>> Hi,
>>
>> Apparently my colleague upgraded the mysql client and server, but, as far as I can tell, this was only 5.7.29 to 5.7.30, and checking the mysql release notes I  don't see anything that looks suspicious there...
>>
>> cheers,
>> --dustin
>>
>>
>> On Tue, May 5, 2020 at 1:37 PM Dustin Lang <dstndstn at gmail.com> wrote:
>>>
>>> Hi,
>>>
>>> We're running Slurm 17.11.12.  Everything has been working fine, and then suddenly slurmctld is crashing and slurmdbd is crashing.
>>>
>>> We use fair-share as part of the queuing policy, and previously set up accounts with sacctmgr; that has been working fine for months.
>>>
>>> If I run slurmdbd in debug mode,
>>>
>>>  slurmdbd -D -v -v -v -v -v
>>>
>>> it eventually (after being contacted by slurmctld) segfaults with:
>>>
>>> ...
>>> slurmdbd: debug2: DBD_NODE_STATE: NODE:cn049 STATE:UP REASON:(null) TIME:1588695584
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug2: DBD_NODE_STATE: NODE:cn050 STATE:UP REASON:(null) TIME:1588695584
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug2: DBD_GET_TRES: called
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug2: DBD_GET_QOS: called
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug2: DBD_GET_USERS: called
>>> slurmdbd: debug4: got 0 commits
>>> slurmdbd: debug2: DBD_GET_ASSOCS: called
>>> slurmdbd: debug4: 10(as_mysql_assoc.c:2033) query
>>> call get_parent_limits('assoc_table', 'root', 'slurm_cluster', 0); select @par_id, @mj, @msj, @mwpj, @mtpj, @mtpn, @mtmpj, @mtrm, @def_qos_id, @qos, @delta_qos;
>>> Segmentation fault (core dumped)
>>>
>>>
>>> It looks (running slurmdbd in gdb) like that segfault is coming from
>>>
>>> https://github.com/SchedMD/slurm/blob/slurm-17-11-12-1/src/plugins/accounting_storage/mysql/as_mysql_assoc.c#L2073
>>>
>>> and If I connect to the mysql database directly and call that stored procedure, I get
>>>
>>> mysql> call get_parent_limits('assoc_table', 'root', 'slurm_cluster', 0);
>>> +---------------------+-----------------+-------------------------+----------------------+---------------------------+-------------+-----------------------------------------------------------------+-------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------+--------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------+-----------------------------+
>>> | @par_id := id_assoc | @mj := max_jobs | @msj := max_submit_jobs | @mwpj := max_wall_pj | @def_qos_id := def_qos_id | @qos := qos | @delta_qos := REPLACE(CONCAT(delta_qos, @delta_qos), ',,', ',') | @mtpj := CONCAT(@mtpj, if (@mtpj != '' && max_tres_pj != '', ',', ''), max_tres_pj) | @mtpn := CONCAT(@mtpn, if (@mtpn != '' && max_tres_pn != '', ',', ''), max_tres_pn) | @mtmpj := CONCAT(@mtmpj, if (@mtmpj != '' && max_tres_mins_pj != '', ',', ''), max_tres_mins_pj) | @mtrm := CONCAT(@mtrm, if (@mtrm != '' && max_tres_run_mins != '', ',', ''), max_tres_run_mins) | @my_acct_new := parent_acct |
>>> +---------------------+-----------------+-------------------------+----------------------+---------------------------+-------------+-----------------------------------------------------------------+-------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------+--------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------+-----------------------------+
>>> |                   1 |            NULL |                    NULL |                 NULL |                      NULL | ,1,         | NULL                                                            | NULL                                                                                | NULL                                                                                | NULL                                                                                             | NULL                                                                                            |                             |
>>> +---------------------+-----------------+-------------------------+----------------------+---------------------------+-------------+-----------------------------------------------------------------+-------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------+--------------------------------------------------------------------------------------------------+-------------------------------------------------------------------------------------------------+-----------------------------+
>>>
>>> and if I run
>>>
>>> mysql> call get_parent_limits('assoc_table', 'root', 'slurm_cluster', 0); select @par_id, @mj, @msj, @mwpj, @mtpj, @mtpn, @mtmpj, @mtrm, @def_qos_id, @qos, @delta_qos;
>>>
>>> I get
>>>
>>> +---------+------+------+-------+-------+-------+--------+-------+-------------+------+------------+
>>> | @par_id | @mj  | @msj | @mwpj | @mtpj | @mtpn | @mtmpj | @mtrm | @def_qos_id | @qos | @delta_qos |
>>> +---------+------+------+-------+-------+-------+--------+-------+-------------+------+------------+
>>> |       1 | NULL | NULL |  NULL | NULL  | NULL  | NULL   | NULL  |        NULL | ,1,  | NULL       |
>>> +---------+------+------+-------+-------+-------+--------+-------+-------------+------+------------+
>>>
>>> but I don't know what to do about this.
>>>
>>> We use another product ("Bright Cluster Manager") to manage some aspects of the cluster and Slurm installation, so we are hesitant to just upgrade Slurm.
>>>
>>> I would appreciate any tips.
>>>
>>> Thanks,
>>> --dustin
>>>
>>>