| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-09-09 | |||
| 18:28:20 | sean-k-mooney | i have it configured | |
| 18:28:34 | sean-k-mooney | and set to different values on each host | |
| 18:28:50 | sean-k-mooney | oh wait | |
| 18:29:12 | artom | Oh, made me realize a thing | |
| 18:29:18 | artom | If the source has it, but dest doesn | |
| 18:29:22 | artom | doesn't | |
| 18:29:28 | artom | We need to remove cpuset from <vcpus> | |
| 18:29:33 | sean-k-mooney | i still have the the versions pinned | |
| 18:29:35 | artom | Which I don't think I'm doing | |
| 18:29:37 | sean-k-mooney | for the rpc | |
| 18:29:42 | sean-k-mooney | that is why its disabled | |
| 18:30:24 | sean-k-mooney | ill unping the rpc and test again | |
| 18:31:49 | sean-k-mooney | shoudl i resterat all nova service for it to pick up the new rpc version | |
| 18:31:57 | sean-k-mooney | oh both hosts | |
| 18:31:58 | mriedem | artom: dansmith: comments in the big patch https://review.opendev.org/#/c/634606/80 | |
| 18:32:01 | sean-k-mooney | *on | |
| 18:32:18 | mriedem | -1 since changes need to be made, but we can decide if they can just be in a follow up after dansmith goes through it | |
| 18:32:38 | artom | mriedem, dansmith, I need to fix https://review.opendev.org/#/c/635229/61 anyways | |
| 18:32:43 | mriedem | i have to run to an appointment, back in a couple of hours | |
| 18:33:17 | dansmith | I think we can just integrate the fixes into the main patch, if artom is okay with it | |
| 18:33:18 | dansmith | and still have time to get it all landed tomorrow | |
| 18:33:19 | mriedem | artom: ok, well try to coordinate that with sean's local testing and if dan is actively doing a review | |
| 18:33:57 | dansmith | I'm still mulling this pcpu series, so I'm cool waiting until artom has put all your comments into the latest version | |
| 18:34:15 | artom | dansmith, you're suggesting we start +W'ing the bottom patches and fix my own -1 later? | |
| 18:34:19 | dansmith | I can plan to hit that one first thing tomorrow and hopefully be all good for mriedem to be happy with it | |
| 18:34:37 | dansmith | artom: no I'm saying go ahead and fix those things | |
| 18:34:58 | artom | OK rihgt, for a second there I didn't recognize you ;) | |
| 18:35:28 | dansmith | glad we cleared that up :) | |
| 18:36:04 | dansmith | I mean, we could land the bottom patch since its just setup, but I don't think we're down to that level of panic just yet | |
| 18:36:21 | sean-k-mooney | dansmith: artom ok it does work correctly. and i guess i also confim that if you pin the rpc i also gets disabled as we expect | |
| 18:36:28 | sean-k-mooney | sorry about that :) | |
| 18:36:57 | dansmith | heh, good | |
| 18:38:20 | sean-k-mooney | artom: ill try removing the vcpu_pin_set form one of the nodes and test that next. then ill go back to reviewing | |
| 18:38:31 | sean-k-mooney | xml before and after http://paste.openstack.org/show/774347/ | |
| 18:38:42 | artom | sean-k-mooney, if it's the dest I expect the XML to not be updated and still have cpuset | |
| 18:39:06 | artom | (In <vcpu>) | |
| 18:39:36 | dansmith | artom: isn't fixing LM for pinned non-numa instances kindof a happy side effect of this code? | |
| 18:39:42 | dansmith | but not really the primary purpose? | |
| 18:39:48 | sean-k-mooney | dansmith: yes it is | |
| 18:39:59 | sean-k-mooney | so i dont think its a blocker for this | |
| 18:40:02 | artom | dansmith, it's technically a tangent, correct | |
| 18:40:09 | sean-k-mooney | its a very very latent bug/feature | |
| 18:40:12 | dansmith | yeah, so I would be fine with that being a follow-on patch *if* it makes things easier | |
| 18:40:28 | sean-k-mooney | ya i think a follow up patch makes sense | |
| 18:40:38 | sean-k-mooney | provided this does not explode | |
| 18:40:39 | dansmith | yeah, kinda like "now we have the infra in place to be able to fix this long-standing bug" | |
| 18:40:54 | dansmith | artom: make sense? | |
| 18:40:56 | artom | dansmith, hrmm. What I wanted to avoid was an objects version bump, but I suppose we could add allocation_cpu_tuning now in the object, and just not use it until the fup | |
| 18:41:17 | dansmith | oh does that require extra fields? | |
| 18:41:24 | artom | Yeah : | |
| 18:41:25 | artom | :/ | |
| 18:41:56 | dansmith | hmm, well, then maybe keeping that separate is also good just from the perspective of not increasing the scope of this already-large patch | |
| 18:42:02 | sean-k-mooney | artom: you dont need extra fileds | |
| 18:42:07 | sean-k-mooney | the ones in the current patch are enough | |
| 18:42:23 | sean-k-mooney | you jsut wont populate it if its not set on the souce or dest | |
| 18:42:42 | artom | sean-k-mooney, I mean adding allocations_cpu_pins | |
| 18:42:52 | artom | We can't get around that, reusing cpu_pins isn't an option | |
| 18:42:55 | sean-k-mooney | oh well you added that already | |
| 18:43:04 | sean-k-mooney | but yes that was an extra filed that was needed | |
| 18:43:11 | artom | sean-k-mooney, right, we're debating taking it back out to simplify things | |
| 18:43:29 | dansmith | artom: this must have been added since the last time I looked? | |
| 18:43:47 | artom | dansmith, yeah, over the weekend I think, or Friday | |
| 18:44:09 | dansmith | okay I see it now | |
| 18:44:49 | dansmith | well, I dunno.. another object bump isn't very impactful, tbh, so putting that in the follow-on patch to actually do the right thing for that case doesn't seem too bad, | |
| 18:44:57 | dansmith | but also more churn doesn't make re-review easier | |
| 18:45:05 | sean-k-mooney | if its not set then we shoudl ensure the cpuset attibute is not in the final xml othersize set it to its value | |
| 18:45:06 | dansmith | so use your judgment I think | |
| 18:45:27 | dansmith | but I think it's okay to exclude fixing that longstanding bug in your series and follow up with the actual fix, however that looks | |
| 18:46:53 | sean-k-mooney | ok so if the souce does not have vcpu_pin_set and the dest has we go form <vcpu placement='static'>1</vcpu> to <vcpu placement='static' cpuset='4-7,12-15'>1</vcpu> | |
| 18:46:54 | artom | dansmith, I know sunk costs is a fallacy, but I think at this point it'd be as much work (for me, anyways) to take it out as it would be to fix it | |
| 18:46:58 | sean-k-mooney | which is what we expect | |
| 18:47:04 | sean-k-mooney | now to check the other way around | |
| 18:47:12 | dansmith | artom: copy that | |
| 18:47:15 | artom | sean-k-mooney, I expect that to be broken | |
| 18:47:24 | sean-k-mooney | well it works correctly | |
| 18:47:45 | sean-k-mooney | source no vcpu_pin_set dest vcpu_pin_set works | |
| 18:47:54 | sean-k-mooney | i need to test the other way now | |
| 18:48:02 | artom | Yes *that's* the broken bit :) | |
| 18:48:54 | sean-k-mooney | kind of | |
| 18:49:04 | sean-k-mooney | it migrated but did not update | |
| 18:49:10 | sean-k-mooney | which is the existing master behavior | |
| 18:49:28 | artom | Yeah, it's what I mean by broken | |
| 18:49:35 | artom | Technically I could it as is :D | |
| 18:49:37 | sean-k-mooney | and a hard reboot fixes it | |
| 18:49:39 | artom | Talk about half-assed | |
| 18:49:52 | artom | "I started fixing it, but then I saw something shiny" | |
| 18:50:18 | sean-k-mooney | well you fix 3 or the 4 combindations | |
| 18:50:32 | sean-k-mooney | it works when the dest has it set or both have it set | |
| 18:50:40 | sean-k-mooney | just not when the souce only has it set | |
| 18:51:12 | sean-k-mooney | so that is better then without your code by a large margin and in the final case it matains the old behavior | |
| 18:51:58 | artom | Eh, I'd not be comfortable leaving it as is | |
| 18:51:59 | sean-k-mooney | so its better or equal to the old behavior so i woudl leave it in and maybe add a todo which we can fix and backport later when we have time | |
| 18:52:02 | artom | Take out completely, sure | |
| 18:52:06 | artom | Fix it, yeah | |
| 18:52:17 | artom | But not this half-assedness | |
| 18:52:57 | sean-k-mooney | ok well its up to you. either fix it or revert it out. im going to get stuff for dinner then contiue reviewing | |
| 18:53:15 | sean-k-mooney | ill be afk for about an hour then ill be back | |
| 18:53:48 | artom | sean-k-mooney, cheers! | |
| 19:09:08 | openstackgerrit | Merged openstack/nova master: Add min service level check for migrate with bandwidth https://review.opendev.org/680394 | |
| 19:10:09 | openstackgerrit | Merged openstack/nova master: migrate: Add bw min service level check of source compute https://review.opendev.org/680395 | |