Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-01-27
22:13:10 superdan diana_clarke: yeah
22:13:21 diana_clarke So perhaps that should read: "Returns 0 when the instance is successfully mapped to a cell, 1 if the instance is not mapped to a cell, and 2 if the instance is in the process of being mapped to a cell."
22:13:23 superdan it's a really small window
22:13:31 superdan sure
22:13:47 diana_clarke so return code 1 is an error case, right?
22:13:50 superdan I mean if you like less awesome comments
22:14:08 diana_clarke superdan: not changing your comment, trying to write the man page
22:14:10 diana_clarke ;)
22:14:20 superdan they both mean that the instance isn't functional,
22:14:48 superdan and 1 could be a transient case as well I think, let me think about it
22:14:50 superdan I know, I meant "awesome" in the snarky way
22:15:35 openstackgerrit Dan Smith proposed openstack/nova: Scheduler calling the Placement API https://review.openstack.org/417961
22:15:36 openstackgerrit Dan Smith proposed openstack/nova: Block starting compute unless placement conf is provided https://review.openstack.org/425806
22:16:33 diana_clarke Or asking differently (like johnthetubaguy), what action should be taken for return codes 1 & 2?
22:16:33 superdan diana_clarke: okay, so from here forward, 1 is always an error, but not one we expect to happen, because:
22:17:15 superdan in the past, we didn't have instance mappings, so if you have an instance in the database and no mapping, it was created on an older release and you didn't do your homework to patch up all the mappings yet
22:17:31 superdan so the point of that tool is really to verify that you created instance mappings for all of them
22:17:52 superdan however, if you run that while things are going on, you could get retval=2 due to a transient luck-based event
22:18:12 superdan but doing the online migrations should fix both cases up for any non-booting instances
22:18:29 superdan so the answer to "what to do" is run the map command
22:19:13 superdan map_instances for 1 and map_cell_and_hosts for 2 (modulo transientness)
22:19:17 superdan clear as mud?
22:21:13 diana_clarke superdan: I going to re-read that 10 times in a row, and then hopefully it will be ;) much appreciated!
22:21:45 superdan let me say it again linearly now that I have it in my head...
22:22:06 superdan If you get retval=1, that means you need to run map_instances as some instances (which were created with a previous version) don't have instance mappings yet
22:22:48 superdan if you get retval=2, that may be because you're racing with instances that are booting, or it may mean that you are missing cell mappings, which means you need to run map_cell_and_hosts
22:23:24 superdan someone could modify that code to not return 2 if the instance has a building-related task state such that we know it's the former case for retval=2
22:24:49 diana_clarke nice, thx!
22:43:57 openstackgerrit Ed Leafe proposed openstack/nova: placement: RT now adds proper Ironic inventory https://review.openstack.org/404472
22:44:29 figleaf johnthetubaguy: cdent: ^^ adds the workarounds for the problem we identified earlier.
22:48:26 mriedem_beach hola
22:48:55 mriedem donde esta el progresso?
22:49:12 superdan where is the progress?
22:49:21 mriedem si
22:49:25 mriedem grenade patch is +W
22:49:32 superdan is it?
22:49:38 mriedem i thought you were going to +W it
22:49:51 superdan I will when I see the nova one yeah
22:49:57 mriedem oh
22:50:04 superdan I think the nova patch was running against a stale version of the grenade one,
22:50:08 superdan so it's going again now
22:50:16 superdan john fixed up a bunch of the unit and pep8 issues
22:50:34 mriedem yeah i see the series must have been rebased
22:50:38 mriedem i'll review that all again
22:51:34 superdan still getting NoValidHost on 11 tests on regular grenade ...
22:51:56 superdan jeez
22:52:05 superdan compute is still not starting because of detecting the placement config
22:52:11 mriedem i thought things were peachy this morning?
22:52:18 mriedem or was that not valid b/c of bad deps?
22:52:48 superdan yeah, I dunno now, I thought we were past that
22:53:36 mriedem maybe changing the grenade patch messed something up?
22:53:55 superdan PS3 worked. but that was the last time
22:54:11 openstackgerrit Pushkar Umaranikar proposed openstack/nova: Adopts keystoneauth with glance client. https://review.openstack.org/412634
22:54:21 superdan but I see no reason why, there's been almost no change since
22:56:53 superdan yeah no placement config in nova.conf
22:56:55 superdan grr
22:58:00 mriedem hmm,
22:58:13 mriedem we're now deploying placement before nova in grenade right?
22:58:18 superdan yeah
22:58:24 mriedem which should configure nova.conf on the primary node,
22:58:33 mriedem which would actually be newton at that point
22:58:36 mriedem since it's pre-upgrade
22:58:44 mriedem 55 is updating a newton compute i mean
22:58:57 superdan but we still call the configure bit
22:59:03 mriedem sure,
22:59:13 mriedem and that shouldn't get wiped out or anything by the 60_nova part
22:59:19 openstackgerrit Pushkar Umaranikar proposed openstack/nova: Adopts keystoneauth with glance client. https://review.openstack.org/412634
23:00:46 mriedem huh, so now both of the nova changes depend on the grenade change
23:00:50 mriedem doesn't hurt anything
23:01:02 superdan I did that because it ran against a stale grenade patch for some reason
23:01:06 superdan so when I rebased I just added it on
23:01:11 superdan didn't know why
23:01:32 superdan ahhh
23:01:47 superdan I wonder if we still need to do configure_placement_nova_compute in the 60_nova bit
23:02:02 superdan not sure why it wouldn't survive, but that's the part that gets us placemetn config
23:02:35 superdan also, what about this? http://logs.openstack.org/06/425806/6/check/gate-grenade-dsvm-neutron-ubuntu-xenial/fb281f1/logs/grenade.sh.txt.gz#_2017-01-27_20_02_44_142
23:02:38 mriedem https://review.openstack.org/#/c/424730/15/projects/55_placement/from-newton/upgrade-placement@17
23:02:47 mriedem ^ should configure nova-compute on the primary node for placement
23:02:57 superdan I know
23:03:02 superdan but
23:03:09 superdan after the run it's not set
23:03:18 superdan oh wait
23:03:26 superdan it doesn't have $NOVA_CONF in the run
23:04:16 mriedem because we didn't source the nova stuff yet?
23:04:22 superdan yeah
23:04:25 mriedem teehee
23:04:30 mriedem i believe i called this ~6 hours ago
23:04:35 superdan yes you did
23:05:18 superdan it's definitely weird doing this out of body but still part of nova
23:05:53 mriedem superdan: so do we no longer need the thing you added where it saves off the number of computes before and checks for that after?
23:06:09 superdan mriedem: that is being challenged, so I split it out
23:06:49 superdan with this set we're introducing a very specific race because the scheduler checks the min vers at startup
23:06:50 superdan which is why I think we need it
23:06:51 superdan but we haven't hit it yet, so ...
23:07:49 superdan mriedem: this is the standalone patch: https://review.openstack.org/#/c/426310/
23:08:09 superdan which I see I have tabs in.. gdi
23:17:56 mriedem ok the 55_placement thing looks good now
23:19:18 mriedem now we play the waiting game
23:21:29 superdan cool
23:21:34 superdan personally,

Earlier   Later