08:05:34 <aspiers> #startmeeting ha
08:05:35 <openstack> Meeting started Wed Mar 29 08:05:34 2017 UTC and is due to finish in 60 minutes.  The chair is aspiers. Information about MeetBot at http://wiki.debian.org/MeetBot.
08:05:36 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
08:05:38 <openstack> The meeting name has been set to 'ha'
08:05:39 <aspiers> let's have a quick chat
08:05:47 <aspiers> maybe about the architecture diagram
08:06:01 <beekhof> sounds like a good place to start
08:06:37 <beekhof> so the workflows are all done?
08:06:47 <aspiers> no
08:06:59 <beekhof> oh
08:07:16 <aspiers> but does it make sense to you?
08:07:21 <beekhof> are mistral workflows HA yet?
08:07:21 <aspiers> as a way forward
08:07:35 <aspiers> I don't think so, let me check
08:07:55 <aspiers> #topic next generation instance HA architecture
08:08:17 <aspiers> #link https://drive.google.com/file/d/0B8tqeaAn45VOWEh4RmgzVEJ1Ums/view?usp=sharing
08:08:34 <beekhof> what was fence_evacuate?
08:08:36 <aspiers> #info aspiers drew this diagram, make sure you open in draw.io to view properly
08:08:53 <aspiers> that was ddeja's old fencing agent
08:08:59 <beekhof> or at least in a size that doesn't require a magnifying glass :)
08:09:11 <aspiers> https://github.com/gryf/mistral-evacuate/blob/master/fence_evacuate.py
08:09:41 <beekhof> nova compute wasn't talking to fence agents in our deployments
08:09:48 <aspiers> it's set to A3 paper size, I have no idea why it displays so small
08:10:39 <aspiers> you sure about that? :) https://github.com/openstack/openstack-resource-agents/blob/master/ocf/nova-compute-wait#L279
08:11:03 <aspiers> this was one of the decouplings which you and I agreed in one of these meetings a month or two
08:11:03 <beekhof> of course not, i always talk out of my butt :)
08:11:08 <aspiers> lol
08:11:42 <beekhof> thats compute _wait_ though. different agent
08:11:51 <aspiers> it was the same in NovaCompute
08:12:02 <aspiers> you just split NovaCompute into systemd + nova-compute-wait
08:12:05 <beekhof> we dont use that agent :)
08:12:10 <beekhof> right
08:12:26 <aspiers> nothing else about this aspect of the architecture is different
08:12:41 <beekhof> anyhooo... devil's advocate... if we get masakari in there, why both with fence_compute and attrd?
08:12:52 <beekhof> s/both/bother/
08:13:00 <aspiers> #info previous architecture discussions in http://eavesdrop.openstack.org/meetings/ha/2017/ha.2017-02-01-09.20.log.html
08:13:15 <aspiers> because we still need to queue evacuation work somewhere reliable
08:13:20 <beekhof> dude, stop using my own words against me :)
08:13:39 <beekhof> didnt they already have the problem solved?
08:14:25 <aspiers> the idea is to decouple the failure monitor/notifier from the failure recovery controller
08:14:47 <aspiers> which would allow mix'n'match architectures
08:15:26 <aspiers> e.g. mistral without masakari, or masakari without mistral, or masakari+mistral, or something completely different (who knows, maybe Senlin can recover stuff)
08:15:38 <beekhof> so fence_compute can loose all the talking to nova functionality
08:16:03 <beekhof> and nova evacuate becomes multi-evac and talks to masakari
08:16:05 <beekhof> y?
08:16:15 <aspiers> yes
08:16:26 <aspiers> or to whatever else
08:17:04 <beekhof> is masakari monitoring VMs in this scenario?
08:17:15 <beekhof> or are we still at compute node failures only
08:17:23 <aspiers> this diagram is just for the latter
08:17:54 <aspiers> but yes I think the way masakari monitors VMs is good
08:18:08 <beekhof> if we're going to switch, i need to be able to sell it. VM monitoring might be the key to that
08:18:26 <aspiers> sure, exactly the same for me
08:19:03 <aspiers> I think masakari has lots of really good aspects
08:19:10 <aspiers> it's much more OpenStacky
08:19:11 <beekhof> do those requests need to be stored in attrd?
08:19:46 <aspiers> well they're not requests, just failures which still need to be delegated to a recovery workflow controller
08:19:54 <beekhof> gotta run. i'll be back in ~30 though
08:19:57 <aspiers> the controller decides what to do with them
08:20:01 <beekhof> hmmm
08:20:10 <beekhof> lets continue this soon
08:20:15 <aspiers> that gives us way more flexibility for different policies
08:20:20 <aspiers> ok, I'll close this meeting for now then
08:20:27 <aspiers> thanks - see you in #openstack-ha in 30
08:22:13 <aspiers> BTW for the record, https://blueprints.launchpad.net/mistral/+spec/mistral-ha was closed as obsolete in January
08:22:32 <aspiers> IIRC this was discussed in one of our weeklies around that time
08:22:36 <aspiers> #endmeeting