Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-06-28
16:05:41 jaypipes :)
16:16:36 openstackgerrit Sean Dague proposed openstack/nova master: Clarify that vlan feature means nova-network support https://review.openstack.org/478551
16:23:52 cfriesen I'd welcome some more reviews of https://review.openstack.org/#/c/461456 It appears that Sahid and I are at a bit of an impasse as to what the semantics of "hw:cpu_realtime_mask" should be. I think it's confusing to have an implied "all-cpus-are-realtime" and that it should be treated as indicating which vCPUs are realtime rather than an exclusion mask indicating which vCPUs are not realtime
16:56:24 openstackgerrit Ed Leafe proposed openstack/nova master: Add a reset for traits DB sync https://review.openstack.org/478564
16:56:41 edleafe jaypipes: ^^ Made it a separate patch
16:57:25 edleafe jaypipes: I can rebase the series on that, or if you think it can merge fast, you can rebase the series on master later
17:01:21 jaypipes edleafe: ah, cool.
17:02:34 jaypipes edleafe: +2 from me.
17:02:42 jaypipes bauzas: feel free to fast-track above patch from edleafe
17:03:41 jaypipes in other news, someone just needs to boot the damn EMC ScaleIO CI system. it's completely unreliable and just noise.
17:04:40 edleafe jaypipes: bauzas is on PTO today
17:05:02 jaypipes alex_xu, melwitt: see edleafe's patch above. simple and sweet and would be good to merge.
17:05:11 jaypipes dansmith: ^^ you too.
17:05:18 dansmith yeah I'm looking
17:05:25 dansmith it seems like a weird way to accomplish that
17:05:25 jaypipes danke
17:05:41 dansmith being in the database fixture I mean
17:05:46 jaypipes dansmith: you mean as opposed to just resetting the _TRAITS_SYNC variable?
17:05:54 dansmith we do all the rest of these global resets in the test base
17:06:07 dansmith lemme get you a link
17:06:32 dansmith https://github.com/openstack/nova/blob/master/nova/test.py#L259
17:06:33 dansmith for example
17:06:47 dansmith and https://github.com/openstack/nova/blob/master/nova/test.py#L295
17:06:48 dansmith and https://github.com/openstack/nova/blob/master/nova/test.py#L306
17:07:02 dansmith and https://github.com/openstack/nova/blob/master/nova/test.py#L267
17:07:08 openstackgerrit Ed Leafe proposed openstack/nova master: placement: support GET /allocation_candidates https://review.openstack.org/475448
17:07:21 dansmith and https://github.com/openstack/nova/blob/master/nova/test.py#L266
17:08:08 jaypipes dansmith: so in short, you'd prefer to call that in the base test setup, yes?
17:08:53 dansmith jaypipes: that and I don't think it should be an object method
17:09:02 melwitt I also don't like the arbitrary putting it on the DatabaseFixture. it's not related to that
17:09:17 edleafe dansmith: so maybe around line 276 in that file?
17:09:22 dansmith jaypipes: we have a bunch of other things in the test base that just import a module and nuke the module-level variable
17:09:57 dansmith edleafe: not in that if, no.. just put it down like 296 or so
17:10:05 dansmith no need to make it conditional, just always nuke
17:11:40 edleafe dansmith: ok, so no worries about it being _PRIVATE? I see the others are referenced directly
17:11:50 dansmith edleafe: no
17:11:54 edleafe k
17:12:12 dansmith all the tests do all day is butt into the private business of other stuff.. :)
17:18:28 openstackgerrit Ed Leafe proposed openstack/nova master: Add a reset for traits DB sync https://review.openstack.org/478564
17:18:36 edleafe dansmith: jaypipes: amended^^
17:20:19 jaypipes edleafe: approved.
17:42:29 openstackgerrit melanie witt proposed openstack/nova master: Make security_group_rules use check_deltas() for quota https://review.openstack.org/477700
17:42:30 openstackgerrit melanie witt proposed openstack/nova master: Add scatter gather utilities for cells https://review.openstack.org/475957
17:42:31 openstackgerrit melanie witt proposed openstack/nova master: Remove 'reserved' count from used limits https://review.openstack.org/446242
17:42:32 openstackgerrit melanie witt proposed openstack/nova master: Make key_pairs use check_deltas() for quota https://review.openstack.org/477699
17:42:34 openstackgerrit melanie witt proposed openstack/nova master: Remove useless quota_usage_refresh from nova-manage https://review.openstack.org/446243
17:42:37 openstackgerrit melanie witt proposed openstack/nova master: Count instances to check quota https://review.openstack.org/416521
17:42:42 openstackgerrit melanie witt proposed openstack/nova master: Make Quotas object favor the API database https://review.openstack.org/410945
17:43:02 openstackgerrit melanie witt proposed openstack/nova master: Add online migration to move quotas to API database https://review.openstack.org/410946
17:44:27 openstackgerrit Ed Leafe proposed openstack/nova master: Add traits to the ResourceProviders filters https://review.openstack.org/474602
18:02:46 dansmith melwitt: what's your thinking on the 300 second timeout?
18:02:58 dansmith I'm thinking anything waiting for us would have given up before that
18:03:35 melwitt dansmith: I copied it from the wait_for_instance_event() in compute/manager as a starting point
18:04:07 melwitt yeah, I wasn't sure what to make it
18:05:21 dansmith okay, that is waiting for neutron do do a thing, but nothing else is blocked on us at that point
18:05:46 melwitt okay. maybe it should be related to the rpc timeout then. I can't remember if that defaults to 120 or so. checking
18:05:54 dansmith but here, theres an http or rpc client waiting on us in most cases, which would be 60s max I think
18:05:58 dansmith 60
18:06:27 dansmith I think if we have a hard-coded timeout, 60 is a good choice personally
18:07:00 dansmith I also was thinking the inner method would take the timeout as a param.. because we might want instance list to timeout quickly, but other things we might want to be more tolerant
18:07:13 dansmith either way, we don't have to have the timeout perfect to land
18:08:02 melwitt okay. do you think maybe timeout should be a kwarg to each utility method? I thought about it but wasn't sure if it was over optimizing
18:08:47 dansmith melwitt: the inner one I think.. the helpers can use a hard coded value I think
18:11:44 melwitt dansmith: okay. I was thinking the helpers don't know what they are going to do (instance list vs something else) but I think you're thinking people can use the inner method for more control so it can be only there. makes sense to me
18:12:27 openstackgerrit melanie witt proposed openstack/nova master: Detach volume after deleting instance with no host https://review.openstack.org/340614
18:12:28 dansmith yeah
18:13:24 melwitt dansmith: adding the timeout thing now. let me know if there are any other things to roll into the change
18:13:32 dansmith I just committed a few other comments
18:13:56 dansmith smack me if I'm wrong
18:14:07 melwitt heh
18:16:48 melwitt dansmith: I wondered about the stalling thing, but I don't see how else I could wait for all the threads in parallel? that's why I went with the linking thing before
18:20:47 dansmith melwitt: using a condition or event, or a queue
18:20:53 dansmith also there's a pool thing, which lets you wait for all of them to complete
18:21:02 dansmith but a queue might be good
18:21:11 melwitt I did look at the pool thing yesterday
18:21:27 dansmith you'll consume off the queue in the waiting thread, processing results as they come, stashing them into the thing to return,
18:21:37 dansmith if you get interrupted by the timeout, you return what you have
18:22:24 dansmith queue.get() also lets you specify a timeout for just that one get,
18:22:56 dansmith so you could manage the timeout yourself, subtracting time you've waited from the total, to avoid having to swallow the timeout exception and move on
18:23:11 dansmith either works I think, but one probably looks cleaner
18:24:56 dansmith eventlet.LightQueue looks to be the most applicable to your case It hink
18:26:14 melwitt dansmith: cool, thanks. I knew I must have been missing something :)
18:46:09 jaypipes edleafe: are you pushing another revision on the GET /allocation_candidates patch?
18:46:18 jaypipes edleafe: if not, lemme know and I will hop on it.
18:53:50 edleafe jaypipes: you mean the two more things efried pointed out?
19:00:46 jaypipes efried: yes. I just didn't want to interfere if you were already doing it.
19:01:00 efried edleafe ^
19:01:09 jaypipes doink, sorry :)
19:01:18 edleafe jaypipes: will fix right after this meeting
19:01:20 efried It's cool, I like hearing my name.
19:01:32 jaypipes edleafe: k, sounds good. thanks again.
19:12:22 openstackgerrit Ed Leafe proposed openstack/nova master: placement: support GET /allocation_candidates https://review.openstack.org/475448
19:12:43 edleafe jaypipes: efried: ^^ knock yourselves out
19:12:51 jaypipes edleafe: thx
19:13:19 efried oooo
19:13:41 edleafe efried: I cleaned up a few more
19:13:50 efried Yuh, see that. Verranize.
19:13:52 edleafe efried: I knew you were planning yet another -1
19:14:16 efried Nope, I was gonna leave those other ones alone, since they involved curlies and/or squares.
19:14:25 efried ...and I didn't want to be TOO much of a dick

Earlier   Later