| openstackgerrit | Matthew Edmonds proposed openstack/nova-powervm master: add lower-constraints job https://review.openstack.org/555964 | 01:16 |
|---|---|---|
| *** edmondsw has quit IRC | 01:19 | |
| *** tjakobs has joined #openstack-powervm | 02:11 | |
| *** tjakobs has quit IRC | 02:17 | |
| *** edmondsw has joined #openstack-powervm | 02:59 | |
| *** edmondsw has quit IRC | 03:04 | |
| *** prashkre has joined #openstack-powervm | 04:44 | |
| *** edmondsw has joined #openstack-powervm | 04:48 | |
| *** prashkre has quit IRC | 04:53 | |
| *** edmondsw has quit IRC | 04:53 | |
| *** chhagarw has joined #openstack-powervm | 05:24 | |
| *** prashkre has joined #openstack-powervm | 06:12 | |
| openstackgerrit | Chhavi Agarwal proposed openstack/nova-powervm master: iSCSI Live Migration Support https://review.openstack.org/567575 | 07:01 |
| *** prashkre has quit IRC | 07:12 | |
| *** prashkre has joined #openstack-powervm | 07:13 | |
| *** AlexeyAbashkin has joined #openstack-powervm | 07:47 | |
| *** prashkre has quit IRC | 07:57 | |
| *** edmondsw has joined #openstack-powervm | 08:24 | |
| *** edmondsw has quit IRC | 08:29 | |
| *** prashkre has joined #openstack-powervm | 08:51 | |
| *** AlexeyAbashkin has quit IRC | 09:53 | |
| *** edmondsw has joined #openstack-powervm | 10:13 | |
| *** edmondsw has quit IRC | 10:17 | |
| *** edmondsw has joined #openstack-powervm | 10:52 | |
| *** AlexeyAbashkin has joined #openstack-powervm | 10:54 | |
| *** prashkre has quit IRC | 11:22 | |
| *** prashkre has joined #openstack-powervm | 11:28 | |
| openstackgerrit | Chhavi Agarwal proposed openstack/nova-powervm master: iSCSI Live Migration Support https://review.openstack.org/567575 | 11:50 |
| edmondsw | efried how do you feel about https://review.openstack.org/#/c/567575/8/nova_powervm/tests/virt/powervm/volume/test_iscsi.py@563 ? | 12:03 |
| *** prashkre_ has joined #openstack-powervm | 12:04 | |
| edmondsw | I'd have preferred that mock was done via decorator | 12:04 |
| edmondsw | and others like it | 12:04 |
| edmondsw | but maybe it doesn't matter so much | 12:04 |
| edmondsw | chhagarw has this been tested in a DevStack environment? | 12:05 |
| *** prashkre has quit IRC | 12:06 | |
| *** prashkre_ has quit IRC | 12:09 | |
| chhagarw | no its not ben tested on Devstack, I have to create one using 2 powervms to try out migration | 12:28 |
| *** chhavi__ has joined #openstack-powervm | 12:40 | |
| *** chhagarw has quit IRC | 12:44 | |
| efried | edmondsw: You can't use a decorator for a mock.patch.object on a runtime-created object like self.vol_drv | 12:50 |
| efried | edmondsw: Could just mock.patch it tho. | 12:51 |
| efried | edmondsw: It's not the same thing, but it's likely fine in this case. | 12:51 |
| edmondsw | efried right, I'm not saying the mock would be identical | 12:52 |
| edmondsw | but just mock the whole class, doesn't need to be on a runtime-created object | 12:52 |
| edmondsw | by mock the whole class, I mean mock.patch | 12:53 |
| edmondsw | on the method | 12:53 |
| efried | edmondsw: Okay, then yes, I generally prefer seeing mocks as decorators if they're just used for the one test, or in setUp if they're used throughout the suite, rather than with context managers, if for no better reason than losing the extra indent. | 12:53 |
| efried | edmondsw: But it's a really soft preference. Do you have stronger reasoning? | 12:53 |
| edmondsw | yep | 12:53 |
| edmondsw | no, pretty much what you said | 12:54 |
| edmondsw | readability, indents, consistency | 12:54 |
| efried | now when there's a reason to patch something on an actual object, this is definitely the way to go. | 12:54 |
| edmondsw | agreed | 12:54 |
| efried | or to patch something for only part of the test case. | 12:54 |
| edmondsw | sure | 12:54 |
| edmondsw | I'd just prefer we only used it in those cases... I don't like doing things when they're not really needed... confusing | 12:55 |
| edmondsw | but I won't hold on that | 12:55 |
| efried | one could argue that patching the whole class rather than the specific instance is doing things that aren't needed :P | 12:57 |
| edmondsw | :P | 12:59 |
| edmondsw | chhavi__ so I'm +1 on that until you tell me this has been well tested with DevStack... not PowerVC | 13:00 |
| openstackgerrit | Matthew Edmonds proposed openstack/networking-powervm master: use master for lower constraints job https://review.openstack.org/569104 | 13:00 |
| edmondsw | efried FYI ^ ... I think this is right, but have no way of testing it locally so we'll see what zuul thinks | 13:01 |
| efried | ight | 13:01 |
| efried | edmondsw: apparently not | 13:01 |
| efried | forgot your colon again | 13:01 |
| edmondsw | oh, same : issue again :( | 13:01 |
| openstackgerrit | Matthew Edmonds proposed openstack/networking-powervm master: use master for lower constraints job https://review.openstack.org/569104 | 13:02 |
| edmondsw | one of these days I'll learn | 13:02 |
| *** esberglu has joined #openstack-powervm | 13:32 | |
| efried | esberglu: Got some comments back from mriedem on localdisk, you on that? | 13:38 |
| esberglu | efried: Not until this afternoon | 13:39 |
| efried | okay. He wanted me to wake you up. | 13:39 |
| efried | (I'm sure he was joking. It's not urgent.) | 13:39 |
| *** tjakobs has joined #openstack-powervm | 13:56 | |
| chhavi__ | esberglu: for devstack AIO setup, after cloning the devstack, do i need to manually download local.conf.aio-sea-localdisk and store it as local.conf before running stack.sh | 14:05 |
| esberglu | chhavi__: Yes. And that local.conf file will only work if you are using SEA networking and localdisk, is that what you want? | 14:06 |
| chhavi__ | yes i am using SEA | 14:07 |
| chhavi__ | want to create one for iSCSI, | 14:07 |
| chhavi__ | localdisk ? | 14:08 |
| esberglu | chhavi__: I've never done anything with iSCSI, afraid I won't be much help there | 14:17 |
| edmondsw | chhavi__ you have two options for the root disk... our localdisk driver or our SSP driver | 14:28 |
| edmondsw | well, you could try boot-from-volume | 14:28 |
| edmondsw | esberglu what you're doing for vSCSI may have some similarities. I sent chhavi__ a link to your WIP CI changes | 14:31 |
| edmondsw | if you know of anything that's not there that she may need... | 14:32 |
| esberglu | edmondsw: chhavi__: Yeah that is probably a good place to start. Take a look at the [[post-config|$CINDER_CONF]] section | 14:36 |
| esberglu | I was given all of that config by thorst and it just worked for vSCSI, not sure what would be needed for iSCSI | 14:36 |
| efried | edmondsw: Wanna throw some relevant -infra reviewers on https://review.openstack.org/#/c/569104/ ? | 14:47 |
| efried | edmondsw: Given that the other ones still fail, it's not clear whether it's actually doing anything. | 14:47 |
| edmondsw | efried yeah | 14:48 |
| edmondsw | I wanted to do that after zuul passed | 14:48 |
| *** AlexeyAbashkin has quit IRC | 15:12 | |
| *** AlexeyAbashkin has joined #openstack-powervm | 15:15 | |
| esberglu | edmondsw: Why can't you use rootvg for the volume group? | 16:05 |
| efried | It's not that you can't. | 16:14 |
| efried | It's that you shouldn't. | 16:14 |
| efried | It allows deployers to fill up your rootvg. | 16:15 |
| efried | which can make everything break. | 16:15 |
| efried | esberglu: ^ | 16:16 |
| edmondsw | yeah, that | 16:18 |
| esberglu | edmondsw: efried: New localdisk is up | 16:49 |
| *** AlexeyAbashkin has quit IRC | 17:04 | |
| *** chhavi__ has quit IRC | 17:18 | |
| esberglu | edmondsw: efried: For vSCSI CI I have a big list of volume tests that are testing the volume API | 17:26 |
| esberglu | Initially I was thinking we would run these to make sure that our cinder config is valid | 17:26 |
| esberglu | But now I'm leaning the other way. They do a lot of volume create/deletes and don't touch our driver at all | 17:27 |
| efried | can you pick out the ones that do touch the driver? | 17:27 |
| esberglu | Without them the vSCSI CI will run 1 (or maybe up to 4, still testing out a few) scenario | 17:27 |
| esberglu | tests | 17:27 |
| esberglu | Those are the ones we actually care about | 17:28 |
| esberglu | I think we just do the scenario tests | 17:28 |
| esberglu | Not sure if either of you have an opinion | 17:28 |
| efried | I'm in favor of skipping tests that don't touch the driver. If that leaves us with only one (or four), so be it. I feel like the nova PTBs will accept that explanation, and we can include comments in whatever config file you're going to have to touch to make this happen. | 17:29 |
| edmondsw | esberglu agreed | 17:30 |
| esberglu | sounds good. I'm planning on setting up the CI pipeline to run only on "powervm:volume-check" comments or something like that | 17:31 |
| esberglu | We could also potentially set it up to run on any changes to powervm/volume | 17:32 |
| esberglu | I'm hoping to have a working (although not optimized) vSCSI pipeilne for OOT master by the end of the week | 17:33 |
| esberglu | Then leave it to mujahid to port it to IT, stable branches, and clean it up a little bit | 17:34 |
| esberglu | efried: edmondsw: https://review.openstack.org/#/c/568982/ | 17:35 |
| esberglu | ^ What's the best way to do handle that? | 17:35 |
| esberglu | I think that we should be able to check the RMC status in that code block, but that will never get merged so we would have to carry a patch long term | 17:36 |
| esberglu | Otherwise I could add config option to add a delay before attaching volume | 17:36 |
| esberglu | Which also would probably never merge tbh but has a better chance | 17:37 |
| edmondsw | esberglu rather than a sleep, could you setup a ping loop? | 17:37 |
| edmondsw | might have a better chance of merging | 17:37 |
| efried | I doubt any of that will have a chance of merging, tbh. | 17:37 |
| edmondsw | oh, this isn't network, it's RMC | 17:37 |
| esberglu | edmondsw: Ping and active RMC aren't the same though. It will ping way before active RMC | 17:38 |
| efried | because needing to have RMC active to attach a volume is a PowerVM thing. | 17:38 |
| edmondsw | yep | 17:38 |
| efried | I think you're going to have to carry the patch. In which case polling for RMC status is probably the most efficient thing. | 17:38 |
| esberglu | efried: Yeah, that's why I thought the delay *might have a chance (default to 0, doesn't affect anyone else) | 17:38 |
| esberglu | But I doubt it as well | 17:38 |
| efried | esberglu: Possible. Worth a try, I suppose. | 17:38 |
| esberglu | Checking the RMC status is so much cleaner though. And it shouldn't be a hard patch to maintain going forward | 17:39 |
| esberglu | I say we do that | 17:39 |
| efried | ++ | 17:39 |
| edmondsw | wfm | 17:39 |
| esberglu | Thoughts on triggering the job for all powervm/volume changes? I'm actually leaning against that | 17:40 |
| edmondsw | I would not at this point | 17:40 |
| esberglu | Something that we could potentially add later | 17:40 |
| esberglu | The other thing that I haven't considered at all is how to do cleanup | 17:41 |
| esberglu | Any thoughts on that? | 17:41 |
| esberglu | The tests will try to delete the volumes even if they fail. But if the volume delete fails there's nothing | 17:42 |
| efried | esberglu: [compute]ready_wait may already be that delay you were talking about. | 17:43 |
| efried | still back on the RMC thing | 17:43 |
| esberglu | efried: Ah I think you're right. I'm spinning up a test instance now to confirm | 17:48 |
| efried | esberglu: There also seems to be something called OS-EXT-STS which may do exactly what you want. If my guess is correct, it's a callback hook that you can use to make tempest ask a function of your choice for things like vm state - which you would override in this case with "lpar active *and* rmc active". | 17:49 |
| efried | yeah no | 17:56 |
| efried | that's not what it is at all. | 17:56 |
| esberglu | efried: I don't see how I would override that? | 17:56 |
| esberglu | Oh nvm | 17:56 |
| esberglu | Oh yeah that doesn't have any idea about RMC | 17:58 |
| efried | No, but if you could somehow manage to set a OS-EXT-STS:task_state (to any value) until RMC comes up, that would do the trick. | 18:00 |
| esberglu | efried: ready_wait works, I'm just gonna do that for now | 18:33 |
| efried | esberglu: roger that. | 18:33 |
| efried | esberglu: tbc, this is going to be a separate job that runs on demand only? | 18:33 |
| esberglu | efried: Yes. And it only runs a whitelist of vSCSI specific scenario tests | 18:34 |
| efried | esberglu: And let's say RMC comes up in one minute: the extra wasted four minutes will be felt once per test we actually run, divided by however many we can run in parallel? | 18:34 |
| esberglu | efried: Right. But I'm not too concerned with runtimes for an on demand job | 18:35 |
| efried | esberglu: Well, yeah, unless it's like hours. | 18:35 |
| efried | esberglu: Just wondering whether long-term it'd be worth implementing and carrying that wait-for-RMC patch instead. | 18:36 |
| efried | esberglu: And thus adding it to the to-do list (edmondsw) | 18:36 |
| edmondsw | yep | 18:38 |
| esberglu | efried: We're only running 1 (or up to 4, in which case all will be parallel) so we would only fell that extra 4 minutes one time per run | 18:39 |
| efried | esberglu: Any sense for how long the overall test takes? | 18:40 |
| esberglu | efried: Well right now I'm using a 10 min wait for RMC to be active which can probably be reduced. With that I think the runtime for the single test was ~20 minutes | 18:42 |
| esberglu | I haven't been paying a ton of attention to runtime though, just trying to get it passing atm | 18:42 |
| efried | esberglu: yeah, I dig it. Just trying to assess the value. Let's go ahead and add it to the to-do list, but it'll be a thing to happen after you're gone :) | 18:42 |
| edmondsw | esberglu we also need https://review.openstack.org/#/c/567599 backported to pike, right? | 20:56 |
| esberglu | edmondsw: Yeah, I thought proper protocol was to wait for the initial patch to merge, is that incorrect? | 20:57 |
| edmondsw | I don't think that's necessary for something like this | 20:57 |
| edmondsw | if ^ was on master, maybe, but it's already on stable/queens | 20:58 |
| edmondsw | probably easier to get someone to review and approve both pike and queens at the same time | 20:58 |
| esberglu | edmondsw: ack will do | 21:00 |
| edmondsw | tx | 21:01 |
| *** esberglu has quit IRC | 21:16 | |
| *** edmondsw has quit IRC | 21:46 | |
| *** tjakobs has quit IRC | 22:20 | |
| *** edmondsw has joined #openstack-powervm | 23:03 | |
| *** edmondsw has quit IRC | 23:08 | |
| *** edmondsw has joined #openstack-powervm | 23:46 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!