Refer to https://github.com/mej/nhc

Make slurmctld and slurmdbd explicitily disabled on startup (rather than just leaving it to whatever it is currently set to)

- template the public interface so NAT Masquerade command goes to right interface
- public inteface name autogenerated by looking at route to 8.8.8.8
- ACCEPT command for private interface modified to use network CIDR. THis is hardcoded to M3 if not defined, so no impact on cluster_in_a_box roles. (Monarch and others will need to pass it as a parameter)
- Not sure if ACCEPT on private interface is used, as compute nodes point to mlx0 interface for routing

Simon Michnowicz authored 7 years ago and Chris Hines committed 7 years ago

changed a role to run on "+  delegate_to: "{{ gluster_servers[0] }}"" so it is consistent within playbook
added a ping value and restarted the daemon. This was to try to fix startup bugs we see in Monarch.

collectd monitoring changes ... monitor additional things in proc, don't monitor slabinfo as its noisy, monitor gpus

Location of node, backup dir, and dummy user account are contained in defaults/main.yml
Both SQL nodes and Management node need to have this role applied, with 'server' parameter determining the different

Traceback (most recent call last):
 File "/usr/local/sbin/provision_slurm.py", line 100, in <module>
   mk_slurmuser_batch(usergroup,"default")
 File "/usr/local/sbin/provision_slurm.py", line 67, in mk_slurmuser_batch
   userdict = defaultdict(list)
NameError: global name 'defaultdict' is not defined

Update provision_slurm.py.j2 to fix logic in mk_slurmuser_batch. This previously did not add any users if one user in the list was a member of 'default'