Wednesday, 2018-03-21

*** AlexeyAbashkin has joined #openstack-powervm00:18
*** AlexeyAbashkin has quit IRC00:22
*** AlexeyAbashkin has joined #openstack-powervm01:18
*** AlexeyAbashkin has quit IRC01:22
*** AlexeyAbashkin has joined #openstack-powervm02:18
*** AlexeyAbashkin has quit IRC02:23
*** chhagarw has joined #openstack-powervm04:40
*** chhagarw has quit IRC04:53
*** chhagarw has joined #openstack-powervm05:19
*** AlexeyAbashkin has joined #openstack-powervm07:55
*** k0da has joined #openstack-powervm08:43
*** efried has quit IRC12:23
*** efried has joined #openstack-powervm12:24
*** edmondsw has joined #openstack-powervm12:40
*** tjakobs has joined #openstack-powervm13:27
efriededmondsw: I'm afraid this is not good news for us.13:52
edmondswtrying to do 2 things at once and doing neither well14:09
edmondswefried that should address what we need, no?14:17
efriededmondsw: Sorry, I quit following.  Catching up...14:17
edmondswnot loving that it's placement API vs. nova API, but...14:17
*** esberglu has joined #openstack-powervm14:19
efriedOkay, I'm caught up.14:21
efriedYeah, I knew we could get all the info from placement.  There were a couple reasons that wasn't ideal.14:21
efriedFirst being that I thought we weren't gonna use placement initially.14:21
efriedSecond being that all the information will be... placementified.  So pvc will need to be able to map it back into reality.14:22
efriedE.g. location codes will be slugified and need to be mapped back to real loc codes.  Descriptions will simply be absent and will need to be looked up.  Etc.14:22
edmondswlooked up how?14:23
efriedIn our yaml file or whatever.14:23
edmondswwhich negates the possibility of using the description as a fallback when there isn't an entry in yaml14:23
efriedUnless we populate the defaults in the yaml at install time based on the descriptions from the driver.14:24
efriedI mean, I'm assuming we're going through this whole song and dance because we don't want to have to ask the nvl REST API questions at runtime.14:24
efriedCause if we could do that, we'd be set.  We wouldn't need placement or anything.14:24
edmondswif by we you mean a customer, yeah, customers shouldn't be talking to NovaLink's REST API diretcly14:25
edmondswand you can consider GUI and customer to be the same thing there14:25
efriedThe GUI wants to talk to the nova API, yah?14:26
efriedand/or the pvc database?14:26
edmondswnever the latter14:26
edmondswonly APIs14:26
efriedBut nova API, not nvl API.14:27
edmondswplacement API could be done there, but not NovaLink14:27
efriedokay.  I still don't really understand why not.  Is it because the nvl API isn't (set up for) remote?14:27
edmondswthat's definitely one reason14:27
efriedpvc runs where the n-api service runs, not where n-cpu runs, kind of thing14:27
edmondswanother is that the GUI has to talk to one ip/hostname14:27
edmondswthat's the way browsers work14:28
efriedOh, it's all client side?14:28
edmondswoh, yeah... the PowerVC GUI runs in the browser14:28
efriedThe GUI comm code is driven from client side, as opposed to server talking to APIs and then presenting "static" pages.  Yeah.14:28
edmondswunlike horizon in that respect, I believe14:29
efriedI'm skeptical about the statement that a browser can't talk to multiple IPs, but won't debate it.14:29
efriedFrames, baby14:32
efriedSo we would have to write an orchestrator that proxies comm to multiple nvls.  I'm sure that's been discussed and vetoed.14:33
edmondswso even if we're not using placement for scheduling, we'll have to update allocations if we want to be able to query them14:33
edmondswI don't know that writing such an orchestrator has ever been discussed, but I expect it would be vetoed for multiple reasons14:34
efriedWe still never officially closed the books on remotable pypowervm, afaik.14:34
efriedWe have "customers" using it, tho.14:34
edmondswesberglu you probably saw that was abandoned15:20
edmondswwe have a TODO in our devstack local.conf files related to that15:20
edmondswthough I don't see a corresponding note in our ci_todo etherpad... am I overlooking it?15:21
esbergluedmondsw: There's one in the in-tree todos15:21
edmondswhmm... doesn't this only affect CI?15:23
esbergluI can move it over, makes more sense with CI15:23
edmondswI did a little digging and found that the merge failures are because they reorganized things:
edmondswif we want to repropose it would be more like
edmondswmaybe just propose that they move to the latest, which adds both ppc64 and arm6415:25
esbergluedmondsw: Yeah I can propose something this week15:25
*** efried is now known as efried_rollin16:01
esbergluedmondsw: for etcd upgrade16:13
esbergluIs devstack the same as nova where you aren't supposed to add reviewers16:13
*** AlexeyAbashkin has quit IRC16:29
*** k0da has quit IRC16:37
edmondswnot sure16:58
edmondswthat's a somewhat general thing, unless you know someone doesn't mind or will be interested16:59
edmondswtonyb you may be interested in that etc commit, which supersedes your old one that just got abandoned17:02
edmondswesberglu got a few minutes?17:34
*** toan has quit IRC17:37
*** adi_____ has quit IRC17:39
*** adi_____ has joined #openstack-powervm17:40
*** toan has joined #openstack-powervm17:41
*** openstackgerrit has joined #openstack-powervm17:45
openstackgerritChhavi Agarwal proposed openstack/nova-powervm master: Use vios_uuids property to process required vioses for iSCSI
chhagarwtjakobs,edmondsw: want to discuss on the proposed changes in review
edmondswpulling it up17:46
chhagarwthis change is to introduce vios_uuids property for processing iSCSI connects17:46
edmondswchhagarw what is the impetus behind this?17:48
chhagarwAs per the current behavior it pulls all the VIOS's and runs connect volume. In a scenario where user would like to use only the active vioses or filter this change is required, so it can process only them rather running discovery on all17:50
chhagarwWe are using the similar behavior for fileio and rbd volume driver17:50
edmondswchhagarw what qualifies as a "user" in this case?17:53
chhagarwadministrator who want to configure vioses which can be used for iSCSI connectivity17:55
edmondswI'm not seeing how this gives user a way to control anything17:55
edmondsw_connect_volume_to_vio isn't called by a user, nor do I see a way for a user to control when that is called and for what vios17:56
tjakobswhat determines if a vios is "active"?17:57
edmondswso unless I'm missing something, you're not giving a user control, you're just changing the behavior across the board17:57
tjakobsalso, is there any reason this should/shouldn't also be put in the case17:57
chhagarwvios_uuids is a property which is the list of active vioses17:57
chhagarwwhich can be used to host storage17:57
chhagarwtjakobs: ideally this should be put for vscsi as well, but I am not sure why its not there.17:59
edmondsw"Active is defined by powered on and RMC state being 'active'."18:00
edmondswchhagarw what is the downside to the current code?18:01
*** AlexeyAbashkin has joined #openstack-powervm18:03
tjakobssounds like this change was "planned". Seems to the "future changeset" of
chhagarwexisting code works, but having this change will avoid unnecessary calls for vios where we know the discoveyr will fail18:03
tjakobsI can't think of any reason not to do this.18:06
edmondswwouldn't we have race conditions with get_active_vioses?18:08
edmondswmaybe that's ok... we just log warnings if we fail to connect18:16
*** chhagarw has quit IRC18:17
edmondswchhagarw added my review. You also have a pep8 issue18:19
*** AlexeyAbashkin has quit IRC18:21
esbergluedmondsw: What's up?18:22
edmondswchhagarw added some more comments in the commit message as well18:25
edmondswesberglu hey... wanted to talk about IT work18:25
esbergluedmondsw: Sure18:25
edmondswdid you see the IRC chat we had with the nova guys yesterday about that?18:26
esbergluThe spec-less stuff?18:26
edmondswyeah... they want us to forgo a spec and use different specless bp for each feature18:26
edmondswso I abandoned the spec commit18:26
edmondswabout to mark the existing bp "obsolete"... that seems like the best option18:27
edmondswwe'll need to open, and then get approved, separate blueprints for each thing we want to get in18:27
edmondswand then update each commit to point to the corresponding new bp18:27
edmondswyou good doing that?18:28
edmondswesberglu drop the list of bps in the IT todo etherpad, and then add them to the next meeting agenda's Open Discussion section18:29
edmondswand then you or I need to make sure we're there to discuss18:29
edmondswnext mtg is tomorrow morning, so would be nice to have at least some if not all of them ready to discuss there18:30
esbergluedmondsw: Sure I can do that18:30
edmondswI should be able to make the mtg tomorrow, but would be good if you're there as well18:31
esbergluedmondsw: Yep I'll be there18:31
edmondswesberglu unrelated (and much lower priority) but I noticed that the etcd change failed zuul pretty hard18:32
esbergluedmondsw: Yeah I saw that too, haven't had a chance to debug18:33
esbergluI need to make sure it doesn't bust our CI either18:33
edmondswI think it's just that that new version isn't staged on tarballs.openstack.org18:34
esbergluedmondsw: Any idea what it would take to do that?18:35
edmondswI think someone like tonyb can help you with that18:35
edmondswdims, etc.18:35
esberglucool tx18:36
*** k0da has joined #openstack-powervm18:38
esbergluedmondsw: Does your capabilities change need a bp?18:54
edmondswI hope not...18:54
edmondswif it does, we probably just wrap it up in one of the ones you open18:55
edmondswesberglu here's some more info on specless bps:
*** k0da has quit IRC19:16
*** AlexeyAbashkin has joined #openstack-powervm19:17
*** AlexeyAbashkin has quit IRC19:21
*** k0da has joined #openstack-powervm19:29
*** efried_rollin is now known as efried19:37
*** tjakobs has quit IRC20:04
*** tjakobs has joined #openstack-powervm20:05
mdrabeefried, edmondsw: Does break OOT?20:08
edmondswmdrabe yup...20:10
edmondswI'll sling a patch20:11
*** AlexeyAbashkin has joined #openstack-powervm20:18
openstackgerritMatthew Edmonds proposed openstack/nova-powervm master: Pass user context to virt driver when detaching volume
*** AlexeyAbashkin has quit IRC20:22
efriededmondsw: no se necessite unit tests?20:23
edmondswyeah, already caught that20:23
edmondswtoo quick20:23
edmondswefried you catch that, but don't reply to the comment I directed to you in slack :)20:24
edmondsws'ok I can butt out of that convo20:25
efriedoh, I didn't think a response was warranted.  You're right.  No argument.20:25
efriedAnd it's what I'm hoping to get down to, if we ever get to the bottom of this.20:25
edmondswI just hate inconsistency20:25
edmondswmakes life hard20:25
efriedWell, in this case we've been taking advantage of a rather weird REST behavioral quirk.  Basically REST is being way nicer to us than it needs to be.  But - they can't just go RBB on us.20:26
edmondswefried yeah, I don't disagree with doing better on our end... but that doesn't excuse them doing better on their end too20:27
edmondswanyway, nuf said20:27
openstackgerritMatthew Edmonds proposed openstack/nova-powervm master: Pass user context to virt driver when detaching volume
edmondswmdrabe ^20:29
edmondswthat is failing a totally unrelated UT locally... still trying to get to the bottom of that, but I'm hoping it's an env issue20:29
esbergluedmondsw: I already did ^ for IT, I wasn't even considering OOT at the time, should have caught that20:30
*** k0da has quit IRC20:30
esbergluedmondsw: bps are up through cold migration, I kept the descriptions super basic but I think that's fine since they20:31
esbergluare pretty striaghtforward20:31
edmondswesberglu great, will look in a minute20:32
*** k0da has joined #openstack-powervm20:33
edmondswyeah, my local UT issue was env... tox recreate fixed it, UT working now20:33
edmondswwow, they actually backported that to queens!20:34
edmondswI guess that's why you noticed mdrabe :)20:34
openstackgerritMatthew Edmonds proposed openstack/nova-powervm stable/queens: Pass user context to virt driver when detaching volume
edmondswmdrabe ^ for queens20:38
tonybYeah we're trying to fix that etcd thing.  We discussed it in Denver and then I let it slide as it wasn't on anyone's critical path.20:49
tonybLet me ping infra20:49
tonybthe good news is due to that dscussion it'll be fixed properly in bionic ... but we still need the rights tarball for xenial (and releases <=queens)20:50
*** esberglu has quit IRC21:01
*** esberglu has joined #openstack-powervm21:14
*** AlexeyAbashkin has joined #openstack-powervm21:17
*** AlexeyAbashkin has quit IRC21:21
efriededmondsw: Suddenly pep8 is failing on our hacking checks?21:32
edmondswefried where?21:33
efriedand its cherry-pick21:33
edmondswhuh... worked locally...21:33
efriedand worked until really, really recently.21:34
efriedI'll propose a fix.21:34
efriedwhich we'll need to backport.21:34
efriedand then rebase these on those.21:34
edmondswif we can find what caused this, we'll know how far we need to backport21:35
efriedThe cherry-pick is actually failing two rules.21:35
efriedone of them introduced by esberglu :P21:36
efriedBoth of the hacking checks in nova-powervm are in nova's hacking checks.  We can remove them.21:36
edmondswefried I only see that in master, so why would queens be failing?21:37
edmondswoh diff failure?21:37
*** tjakobs has quit IRC21:37
efriedQueens is failing on the same one (yield must be followed by a space) as well as an additional one (redundant import aliases)21:37
edmondswno, same issue + the esberglu one21:38
edmondswso... how, if those changes aren't in queens?21:38
edmondswefried yeah, neither of those rules is in stable/queens21:40
efriedBetting we forgot to update our req when queens was cut.  So we're still pulling master nova in queens nova-powervm.21:40
efriedso, multiple fixes otw.21:42
edmondswsame problem in the other *-powervm projects21:43
edmondswwant me to fix those?21:43
openstackgerritEric Fried proposed openstack/nova-powervm master: Remove custom hacking rules
efriedI think once we're under governance, their bots will propose those changes for us.  But yeah, until then, we have to do it ourselves.21:43
efriededmondsw: Sure, you wanna fix the tox.inis?21:44
edmondswwill do21:44
efriedMeanwhile, the above will fix master.21:44
openstackgerritEric Fried proposed openstack/nova-powervm master: Pass user context to virt driver when detaching volume
efriedrebased ^ on ---^21:45
edmondswtox recreates are so slow...21:48
openstackgerritMatthew Edmonds proposed openstack/nova-powervm stable/queens: tox needs to pull from stable/queens
openstackgerritMatthew Edmonds proposed openstack/networking-powervm stable/queens: tox needs to pull from stable/queens
openstackgerritMatthew Edmonds proposed openstack/ceilometer-powervm stable/queens: tox needs to pull from stable/queens
edmondswefried esberglu ^21:57
esbergluedmondsw: One comment22:01
esbergluon the nova-powervm one22:01
openstackgerritMatthew Edmonds proposed openstack/nova-powervm stable/queens: tox needs to pull from stable/queens
edmondswesberglu ^22:02
esbergluedmondsw: My bad it's actually stable/queens22:03
edmondswoh, of course22:03
esbergluWas just amending my comment too but ya beat me to it ;)22:03
openstackgerritMatthew Edmonds proposed openstack/nova-powervm stable/queens: tox needs to pull from stable/queens
edmondswesberglu ^22:04
edmondswand I checked the url in my browser this time :)22:04
esbergluedmondsw: I'll quick approve those as soon as zuul finishes22:10
openstackgerritEric Fried proposed openstack/nova-powervm stable/queens: Pass user context to virt driver when detaching volume
efriedrebased ^ on ---^22:11
openstackgerritMerged openstack/nova-powervm master: Remove custom hacking rules
esbergluefried: edmondsw: I'm going to redeploy the CI tonight. It's struggling right now.22:24
esbergluThe zuul queue got all backed up and the rabbitmq server on the undercloud died22:25
esbergluAnecdotal, but in the past those have been happening at the same time as pok network issues22:26
efriedesberglu: Does CI rely on tox.ini?  To set up the nova requirement or whatever?22:34
efriedWondering whether we need to bother waiting for CI to merge the tox.ini fixes.22:34
efriedThe networking-powervm tox.ini fix is ill.  That policy.json thing is blocking us now.22:36
efriedesberglu: We need to backport ?22:36
esbergluefried: We're good to skip CI on the tox.ini fixes22:38
esbergluneutron didn't get added to stable/queens requirements22:42
esberglubut yeah that looks like the same issue22:42
esbergluefried: I'll try to put together a fix tonight, gotta go offline for a bit now though22:50
*** esberglu has quit IRC22:51
openstackgerritMerged openstack/nova-powervm stable/queens: tox needs to pull from stable/queens
*** edmondsw has quit IRC23:07
*** k0da has quit IRC23:10

Generated by 2.15.3 by Marius Gedminas - find it at!