We believe we have resolved this by recreating some recently deleted QOS. The controller starts again OK. But the QOS have been recreated with new ID, so it's unclear how that would have fixed it.
And if the problem was an invalid QOS why wasn't this picked up rather than seg faulting? The code at https://github.com/SchedMD/slurm/blob/slurm-20.11/src/common/slurmdb_defs.c#L548: GNU gdb (GDB) Red Hat Enterprise Linux 7.6.1-120.el7 Copyright (C) 2013 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html> This is free software: you are free to change and redistribute it. There is NO WARRANTY, to the extent permitted by law. Type "show copying" and "show warranty" for details. This GDB was configured as "x86_64-redhat-linux-gnu". For bug reporting instructions, please see: <http://www.gnu.org/software/gdb/bugs/>... Reading symbols from /usr/sbin/slurmctld...done. [New LWP 147686] [New LWP 147687] [New LWP 147688] [New LWP 147690] [Thread debugging using libthread_db enabled] Using host libthread_db library "/lib64/libthread_db.so.1". Core was generated by `/usr/sbin/slurmctld -D -vvv'. Program terminated with signal 11, Segmentation fault. #0 0x00007f4f423c4671 in bit_set (b=b@entry=0x2e06c40, bit=bit@entry=-518653972) at bitstring.c:237 237 b[_bit_word(bit)] |= _bit_mask(bit); Missing separate debuginfos, use: debuginfo-install slurm-slurmctld-20.11.9-1.el7.x86_64 (gdb) bt #0 0x00007f4f423c4671 in bit_set (b=b@entry=0x2e06c40, bit=bit@entry=-518653972) at bitstring.c:237 #1 0x00007f4f4248af5e in _set_qos_bit_from_string (valid_qos=valid_qos@entry=0x2e06c40, name=<optimized out>) at slurmdb_defs.c:548 #2 0x00007f4f4248eedc in set_qos_bitstr_from_list (valid_qos=0x2e06c40, qos_list=<optimized out>) at slurmdb_defs.c:2288 #3 0x00007f4f423b896d in _set_assoc_parent_and_user (assoc=assoc@entry=0x25e3500, reset=reset@entry=0) at assoc_mgr.c:885 #4 0x00007f4f423bfae7 in _post_assoc_list () at assoc_mgr.c:1034 #5 0x00007f4f423c2093 in _get_assoc_mgr_assoc_list (enforce=27, db_conn=0x20f38d0) at assoc_mgr.c:1584 #6 assoc_mgr_init (db_conn=0x20f38d0, args=args@entry=0x7fff2bc95d10, db_conn_errno=<optimized out>) at assoc_mgr.c:2039 #7 0x000000000042bae5 in ctld_assoc_mgr_init () at controller.c:2323 #8 0x000000000042de95 in main (argc=<optimized out>, argv=<optimized out>) at controller.c:620 (gdb) The problem line is: (*(my_function))(valid_qos, bit); But 17 lines above that, the line xassert(valid_qos); Appears to check if the qos is valid - but in this case the qos existed but was deleted - should there be more checks here for a deleted QOS? Would have an assert been triggered if we had compiled slurmd with --enable-developer as per https://slurm.schedmd.com/faq.html#debug ? Many thanks, Luke -- Luke Sudbery Principal Engineer (HPC and Storage). Architecture, Infrastructure and Systems Advanced Research Computing, IT Services Room 132, Computer Centre G5, Elms Road Please note I don't work on Monday. From: slurm-users <slurm-users-boun...@lists.schedmd.com> On Behalf Of Luke Sudbery Sent: 07 July 2022 12:28 To: slurm-us...@schedmd.com Subject: [slurm-users] Slurmctl seg faulting Hello, After a restart this morning*, slurmctl is starting but then failing before it is up and running: slurmctld: debug2: user user1 default acct is account1 slurmctld: debug2: assoc 68939(account1, user2) has direct parent of 26369(account1, (null)) slurmctld: debug2: user user2 default acct is account1 slurmctld: debug2: assoc 26396(account1, user3) has direct parent of 26369(account1, (null)) slurmctld: debug2: user user3 default acct is account1 slurmctld: debug2: assoc 26390(account1, user4) has direct parent of 26369(account1, (null)) slurmctld: debug2: user user4 default acct is account1 slurmctld: debug2: assoc 26342(account2, (null)) has direct parent of 3(root, (null)) slurmctld: debug2: assoc 76299(account2, rq-worker) has direct parent of 26342(account2, (null)) Segmentation fault (core dumped) [root@bb-er-slurm01 state]# Which give us very little to go on. Earlier, there was a long list of slurmctld: error: assoc 62346 doesn't have access to it's default qos 'qos1' slurmctld: error: assoc 61271 doesn't have access to it's default qos 'qos1' slurmctld: error: assoc 61221 doesn't have access to it's default qos 'qos1' slurmctld: error: assoc 61215 doesn't have access to it's default qos 'qos1' Shortly before the segfault. We have cleared those, by setting correct default qos with sacctmgr, but the controller is still seg faulting. *The restart was initiated by our HA cluster (pacemaker/corosync) and at one point 2 copies of slurmdbd may have been running. But the second copy appeared to detect this and back off? Not sure: Jul 07 09:06:47 bb-aw-slurm01.bear.cluster systemd[1]: Started Cluster Controlled slurmdbd. Jul 07 09:06:47 bb-aw-slurm01.bear.cluster slurmdbd[212307]: debug: Log file re-opened Jul 07 09:06:47 bb-aw-slurm01.bear.cluster slurmdbd[212307]: pidfile not locked, assuming no running daemon Jul 07 09:06:47 bb-aw-slurm01.bear.cluster slurmdbd[212307]: debug: auth/munge: init: Munge authentication plugin loaded Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: accounting_storage/as_mysql: _check_mysql_concat_is_sane: MySQL server version is: 10.3.32-MariaDB Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: error: _mysql_query_internal: deadlock detected attempt 1/10: 1213 WSREP replication failed. Check your wsrep connection state and retry the query. Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: error: mysql_query failed: 1047 WSREP has not yet prepared node for application use create table if not exists table_defs_table (creation_time int unsigned not null, mod_time int unsigned default 0 not null, table_name text not null, Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: accounting_storage/as_mysql: init: Accounting storage MYSQL plugin failed Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: error: Couldn't load specified plugin name for accounting_storage/mysql: Plugin init() callback failed Jul 07 09:06:52 bb-aw-slurm01.bear.cluster slurmdbd[212307]: error: cannot create accounting_storage context for accounting_storage/mysql Jul 07 09:06:52 bb-aw-slurm01.bear.cluster systemd[1]: slurmdbd.service: main process exited, code=exited, status=1/FAILURE Jul 07 09:06:52 bb-aw-slurm01.bear.cluster systemd[1]: Unit slurmdbd.service entered failed state. Jul 07 09:06:52 bb-aw-slurm01.bear.cluster systemd[1]: slurmdbd.service failed. But either way, we suspect database corruption may be the problem, but slurmdbd logs don't give anything to go on. So questions, 1. How can we troubleshoot further? I am currently attempting to rebuild slurm with debugging symbols (https://slurm.schedmd.com/faq.html#debug) as gdb is not giving any useful information at all ATM. 2. What are the implications of restoring the database? How does slurm reconcile the database and the state directory? Many thanks, Luke -- Luke Sudbery Principal Engineer (HPC and Storage). Architecture, Infrastructure and Systems Advanced Research Computing, IT Services Room 132, Computer Centre G5, Elms Road Please note I don't work on Monday.