| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-09-27 | |||
| 18:28:28 | mriedem | order only matters for performance | |
| 18:28:43 | Tengu | and I rebooted the controllers in order to ensure all is running at the latest config version | |
| 18:28:47 | melwitt | Tengu: no but you will want to check nova-scheduler logs to make sure some other filter isn't rejecting it | |
| 18:28:48 | Tengu | mriedem: hmm ok. | |
| 18:29:12 | melwitt | at DEBUG log level. it's possible something else is going wrong and not the key match for the metadata | |
| 18:29:16 | Tengu | melwitt: yup, but I didn't see anything. the instance "directory" was created on the right node in /var/lib/nova/instances | |
| 18:29:31 | melwitt | if you're getting NoValidHost you should see something | |
| 18:29:49 | Tengu | but after a while, paff, directory is removed, and crash, "no host found"… although it actually HAD found a host | |
| 18:30:00 | melwitt | unless a compute host rejected the request in which case you should see an error in the nova-compute logs or the nova-conductor logs | |
| 18:30:08 | Tengu | hmmm. | |
| 18:30:29 | Tengu | will check that one. | |
| 18:30:53 | melwitt | the way it works is if scheduling filters all pass, it goes to nova-compute, if something fails while it builds it, it will tear it down, log stuff, and try to reschedule to another host if you have retries configured | |
| 18:31:05 | Tengu | what would be the patter of a rejection in nova-compute.log ? | |
| 18:31:16 | Tengu | hmm ok. | |
| 18:31:21 | Tengu | I have the retryfilter | |
| 18:31:23 | melwitt | should see something logged at ERROR level I think | |
| 18:31:28 | melwitt | in nova-compute | |
| 18:31:29 | Tengu | think this one will try to re-schedule | |
| 18:31:45 | Tengu | duh | |
| 18:32:05 | Tengu | corrupted image download o_O | |
| 18:32:47 | Tengu | that might explain a bit. but that would point the glance storage | |
| 18:33:17 | openstackgerrit | Eric Berglund proposed openstack/nova master: PowerVM Driver: config drive https://review.openstack.org/409404 | |
| 18:33:18 | openstackgerrit | Eric Berglund proposed openstack/nova master: WIP(5): PowerVM driver: ovs vif https://review.openstack.org/422512 | |
| 18:33:31 | Tengu | although… hmm. timestamp doesn't really match. will dig a bit more. | |
| 18:34:31 | melwitt | Tengu: yeah, so you have other issues there. but as long as the instance is always landing on the host where the aggregate meta matches the flavor, you know at least the extra specs filtering is working correctly | |
| 18:34:44 | melwitt | (for your original concern) | |
| 18:35:11 | Tengu | melwitt: right. | |
| 18:35:37 | Tengu | so my debug steps weren't that wrong. I should have had a better look to the nova-compute.log file though. | |
| 18:36:44 | mriedem | you can also trace the request id and/or instance id through the logs if you have your logs pumped to an ELK stack | |
| 18:37:05 | mriedem | or journald like in devstack | |
| 18:38:29 | Tengu | for now we don't have an ELK (it will run on the openstack… well, yes, that might cause some issues at some point ;)). | |
| 18:38:34 | Tengu | but we want to do that, yep. | |
| 18:39:01 | dansmith | mriedem: https://imgur.com/a/FY7Oq | |
| 18:39:14 | dansmith | mriedem: over about 300 runs, my patch is consistently faster than master | |
| 18:39:37 | mriedem | oh that's w/o the policy fix :) | |
| 18:39:40 | mriedem | i was like, wtf | |
| 18:39:43 | dansmith | yes | |
| 18:39:47 | Tengu | but the image corruption is the best hint for now. Have to check why - the ceph cluster isn't a cluster for now and we have some failed disks on it, so it can explain a lot. it's not in prod for now, this also explain some issues | |
| 18:41:21 | mriedem | dansmith: throw that in https://etherpad.openstack.org/p/nova-instance-list somewhere so we don't lose it | |
| 18:42:06 | stvnoyes | hi mriedem, if you get a change to re-review https://review.openstack.org/#/c/463987/ it would be great. I am on vacation next week so there's still some time this week for me to turn the review around again if it's needed. thanks. | |
| 18:43:48 | mriedem | ok | |
| 18:44:00 | mriedem | dansmith: totally unrelated, but i'm think about throwing the ceph job in the experimental queue http://tinyurl.com/ydy3jek9 | |
| 18:44:18 | stvnoyes | johnthetubaguy: pls take a look at https://review.openstack.org/#/c/506805/ when you get a chance. it's a pretty small change, and it's needed for the cinder v3 live migrate change. thanks. | |
| 18:44:23 | dansmith | melwitt: ^ | |
| 18:44:54 | melwitt | gdi | |
| 18:46:39 | melwitt | it would take me awhile to unroll what is going on with that job | |
| 18:46:52 | mriedem | this is compared to the normal dsvm tempest job http://tinyurl.com/ydemspkl | |
| 18:47:46 | melwitt | yeah, hm. so it was tracking okay until around the 16th | |
| 18:48:14 | melwitt | I'll dig into it | |
| 18:48:19 | mriedem | well, there were also spikes in the normal job then too, just not as bad | |
| 18:48:25 | mriedem | 9/23 is where it goes nuts | |
| 18:49:10 | melwitt | yeah. last we discussed it was at the last PTG and jbernard had some TODOs but I didn't know details about what they were. I thought the first thing was something to do with a job timeout being too short | |
| 18:49:32 | mriedem | he was going to start restricting the tests | |
| 18:49:47 | melwitt | restricting in what way? | |
| 18:49:56 | mriedem | L138 https://etherpad.openstack.org/p/nova-ptg-pike | |
| 18:50:55 | melwitt | okay. so that would be my starting point, aside from the latest craziness. which might just be more timeout and OOM stuff (have to dig) | |
| 18:51:10 | efried | mriedem Re-request tough love on https://review.openstack.org/#/c/488137/ please | |
| 18:51:30 | mriedem | melwitt: https://github.com/openstack/nova/commit/980d0fcd75c2b15ccb0af857a9848031919c6c7d merged on the 22nd | |
| 18:51:43 | mriedem | cinder.tests.tempest.api.volume.test_volume_revert.VolumeRevertTests.test_volume_revert_to_snapshot_after_extended is what i see failing | |
| 18:51:51 | mriedem | so my guess is, the ceph job doesn't care for live snapshots | |
| 18:52:19 | melwitt | thanks for that info | |
| 18:54:19 | mriedem | although the test that's failing is a cinder api test | |
| 18:54:27 | mriedem | and i don't see any related errors in n-cpu | |
| 18:55:52 | openstackgerrit | Dan Smith proposed openstack/nova master: Fix policy check performance in 2.47 https://review.openstack.org/507948 | |
| 18:59:40 | dansmith | oops, didn't finish the commit message on that one | |
| 18:59:48 | dansmith | I'm just testing it in my devstack rig anyway | |
| 19:02:46 | openstackgerrit | Matt Riedemann proposed openstack/nova master: doc: make host aggregates examples more discoverable https://review.openstack.org/507950 | |
| 19:02:50 | mriedem | not even the bug link | |
| 19:02:53 | mriedem | you were so excited | |
| 19:02:59 | mriedem | about getting punched in the naughty parts | |
| 19:03:07 | mriedem | Tengu: melwitt: https://review.openstack.org/#/c/507950/ | |
| 19:03:13 | mriedem | ^ should help a bit with doc discovery | |
| 19:03:51 | Tengu | mriedem: \o/ thanks ! | |
| 19:04:07 | Tengu | for now I'm digging in glance, as apparently it's crashed. | |
| 19:07:42 | dansmith | sdague: mriedem: cfriesen_: https://imgur.com/a/IQ0Vh | |
| 19:07:58 | mriedem | dansmith: awesome | |
| 19:08:03 | mriedem | also, | |
| 19:08:30 | mriedem | i ran the 2.53 microversion, GET /servers/detail thing again w/o your patch, to see why i had just a big difference in numbers, and you're right, it's the vm | |
| 19:08:43 | mriedem | so w/o your patch, it's still closer to with your patch, | |
| 19:08:51 | mriedem | and over half of what it was the other day on the other vm | |
| 19:08:58 | mriedem | so just need to chalk that up to public cloud | |
| 19:09:09 | cdent | what happened at 2.47? | |
| 19:09:21 | mriedem | cdent: we started checking policy per instance when listing instances | |
| 19:09:21 | dansmith | mriedem: cool | |
| 19:09:29 | mriedem | which adds up when you're listing 1000 instances | |
| 19:09:30 | cdent | ouch | |
| 19:10:34 | openstackgerrit | Dan Smith proposed openstack/nova master: Fix policy check performance in 2.47+ https://review.openstack.org/507948 | |
| 19:11:48 | cfriesen_ | cdent: my bad, I didn't realize policy check was expensive | |
| 19:12:24 | mriedem | i'm sure i approved the change so don't worry about it | |
| 19:12:28 | cdent | cfriesen_: a reasonable thing to assume in a reasonable universe, but we probably left that one long ago | |
| 19:12:32 | sdague | I kind of wonder if there are other places with embedded policy checks like that are expensive | |
| 19:13:17 | sdague | cfriesen_: there is an implicit fstat because policy is live reread | |
| 19:13:53 | cdent | speaking of, that’s a potential next microoptimization in the unit tests. that file gets read over and over and over over and over and over and ... | |
| 19:14:44 | sdague | honestly, it might behoove us to change that behavior entirely, as we've got the hup handler now | |
| 19:15:16 | bauzas | dansmith: you trampled me | |
| 19:15:27 | dansmith | bauzas: I did? | |
| 19:15:35 | bauzas | dansmith: with Twitter | |
| 19:15:47 | bauzas | :p | |
| 19:16:09 | bauzas | so, maybe you should be the next US president given you use Twitter for trampling folks :p | |
| 19:16:23 | bauzas | mmm, maybe "trample" is not the right verb | |