Tuesday, 2018-01-02

*** edmondsw has joined #openstack-powervm02:45
*** edmondsw has quit IRC03:03
*** edmondsw has joined #openstack-powervm03:05
*** edmondsw has quit IRC03:06
*** edmondsw has joined #openstack-powervm04:35
*** edmondsw has quit IRC04:40
*** edmondsw has joined #openstack-powervm06:23
*** edmondsw has quit IRC06:28
*** edmondsw has joined #openstack-powervm08:12
*** edmondsw has quit IRC08:16
*** edmondsw has joined #openstack-powervm10:00
*** edmondsw has quit IRC10:04
*** edmondsw has joined #openstack-powervm12:19
*** edmondsw has quit IRC12:20
*** chhavi has joined #openstack-powervm12:56
*** chhavi has quit IRC13:01
*** chhavi has joined #openstack-powervm13:02
*** edmondsw has joined #openstack-powervm13:39
*** edmondsw has quit IRC13:44
*** efried has joined #openstack-powervm13:46
*** edmondsw has joined #openstack-powervm14:14
*** esberglu has joined #openstack-powervm14:37
*** mdrabe_ is now known as mdrabe14:38
esbergluefried: edmondsw: You guys in today?14:53
efriedesberglu I'm here.14:53
esbergluefried: I've got a few changes out preparing for IT SEA CI14:54
efriedYup, on my radar.14:54
efriedClearing out some of the email backlog first.14:54
esbergluefried: Cool. The local.conf cleanup and change from whitelist to blacklist are ready14:55
esbergluWe need a new pypowervm to be cut before we can merge the actual SEA IT CI change14:55
esbergluI don't think there's anything we are waiting to get in before we do so?14:56
*** chhavi has quit IRC14:57
*** chhavi has joined #openstack-powervm14:58
*** efried has quit IRC15:02
edmondswesberglu I'm here as well15:10
*** efried has joined #openstack-powervm15:13
esbergluedmondsw: efried: OVS IT is working as expected for spawns and deletes. I'm still working on getting a full tempest run passing on that15:18
esbergluBut networking specific tempest tests are looking good15:18
esbergluIt just ran out of ips last try, so I'm adding more and will let you guys know how that looks15:19
edmondswsounds good, thanks15:19
edmondswesberglu I've got a couple things I have to knock out and then I'll try to catch up with your change proposals15:20
esbergluedmondsw: thx15:21
catmandohey chaps17:17
catmandojust noticed the activity; don't know if you're able to see my questions above17:18
catmandodon't want to repaste if you can17:18
*** csky has joined #openstack-powervm17:33
efriedcatmando I don't have 'em; please re-paste.17:40
catmandoi think i found a bug in powervc 1.417:42
catmandobut i don't know if it's really a bug or just missing documentation17:42
catmandoi cannot create a server using the POST to nova's /servers unless i use an imageRef17:42
catmandobut https://developer.openstack.org/api-ref/compute/ states that imageref is optional if i set block_device_mapping_v217:42
catmandoany attempt to not use an imageref fails at line 981 in powervc-nova/compute/manager.py17:42
catmandocommenting out that block just leads to more failures later on17:42
catmandothe error that eventually bubbles through to powervc is "failed prepping block device"17:42
catmandothis is a real issue for us as powervc's naming scheme when it comes to creating instances from images makes management very difficult (especially on the Brocade switches)17:42
catmandowe were hoping to resolve this by doing our own flash copies (backups) on the SVC and then creating an instance from that nicely named copy17:42
catmandothere also seems to be some custom undocumented method for marking volumes as bootable17:42
catmandospecifically, marking a volume as bootable in cinder does nothing, it has to be marked as "is_boot_volume" in the volume_metadata table17:42
efriededmondsw mdrabe ^^ Your wheelhouse.  (Or maybe gfm needs to brought in?)17:44
edmondswreading back17:44
edmondswyeah, let me try to pull in gfm17:46
catmandofor the time being I have worked around this by changing the configs and some places where the instance and volume templates are hardcoded17:47
edmondswcsky are you around?17:48
*** chhavi has quit IRC17:51
mdrabecatmando: Are you wanting the PowerVC instance name, hypervisor LPAR name, and zone name on the fabric to all be similar?17:53
catmandomdrabe: yes, and without cutting off the instance name / adding UUIDS17:54
catmandoit makes reading zones on the SAN switches very painful and we have a pattern in place for naming17:54
catmandoit also makes looking at the HMC painful17:54
catmandoas I say, my workaround does work, but the above feels like a bug17:55
mdrabeYea you can change the name templates in conf to make it more eyeball friendly, I would just proceed with caution17:55
catmandothere's no reason for the block_device_mapping_v2 to fail with that error. i can clearly see that the code ALWAYS expectes an imageRef, which in that case is wrong17:56
mdrabeI believe the reason we don't mention those conf options at all is because we want to avoid naming collisions17:56
catmandomdrabe you can, but there are also hardcoded lines17:56
catmandothe conf files won't allow you to drop all the UUIDs17:56
catmandoI can provide a patch set17:56
catmando(but it feels like an edge case)17:57
catmandoas i say, my concern is regarding the imageRef and the bootable flag17:57
mdrabeAre you looking at nova?17:57
catmandonova and cinder17:57
catmandoin fact, nova is actually setting  a portion of the volume name that cinder then picks up and modifies further17:58
mdrabeCan you link the line #s from github?17:58
mdrabeer this is pvc specific isn't it17:59
catmandoI can, but have left the office now. Will leave it here17:59
catmandoYes, it is17:59
mdrabeWhat Python module is it? I can find it from there17:59
catmandoHave a search for .37s18:00
catmandoIt's the string format that's hard coded in there somewhere18:01
catmandoI forget where exactly18:01
catmandoThe patches are small, maybe 4 or 5 lines in total18:01
mdrabeHmm not finding ".37s"18:03
edmondswmdrabe powervc_cinder/db/sqlalchemy/models.py18:05
mdrabeah thnx18:05
mdrabecatmando are you using the pvc_volume_name_template option in cinder conf?18:07
catmandoYes, I did try18:08
catmandoThere are a number of places where uuids are appended18:08
mdrabeLookin through it18:11
*** csky has quit IRC18:31
mdrabecatmando: From what I can tell each volume driver defines it's own naming conventions based on the limitations of the storage provider.18:44
mdrabeThe pvc_volume_name_template conf option looks like it's used only for existing volumes though18:45
mdrabeThe community cinder conf option volume_name_template is used for setting the volume name out on the storage provider18:46
edmondswmdrabe I think his primary question is why the code is expecting imageRef when he specified block_device_mapping_v219:01
edmondswput another why... why isn't boot-from-volume (in the nova sense) working?19:03
edmondswI suspect that's a limitation more than a bug, but I don't know if it's something we can address19:05
edmondswso if we can find another solution to his case where he doesn't need to use boot-from-volume, great19:05
mdrabeYea I'm pretty sure pvc has to boot from an image19:05
edmondswyeah, I just don't know the ins and outs well enough to know if that was more of a test statement and a couple code tweaks or an intentional "that's not gonna work" thing19:06
mdrabeYea me neither19:07
edmondswcatmando ^19:08
edmondswNone of the folks I've looked for that should be able to give a more complete answer seem to be around today19:08
*** zerick has joined #openstack-powervm19:22
catmando@edmondsw: no worries, maybe something will crop up over the next couple of days19:33
esbergluefried: edmondsw: Got a working tempest run for OVS IT20:10
esbergluI started with the OOT SEA skip list, so there are probably some more tests I can enable20:10
esbergluBut I'm satisfied with the results for now. Just need your reviews and then we can bug the cores20:11
edmondswesberglu excellent20:23
efriedesberglu So right now your powervm-ci patch is setting us up for SEA - is the plan to switch that over to OVS at some point soon?20:27
efriedOh, I had forgotten where we were with in-tree patches.  Looks like OVS isn't even proposed yet.  So this is The Right Thing for now.20:29
esbergluefried: I thought that the plan was to support SEA IT CI for queens? Then do further work to get both SEA and OVS CI pipelines next release20:30
efriedyeah yeah, that sounds right.20:30
efriedoh, I wasn't seeing the ovs patch because it's still under wangqwsh's name.20:31
edmondswefried I was just about to ask what you meant there :)20:31
edmondswabout not being proposed20:31
efriedWell, so that brings up the question: why does it need to be patched in for the CI?20:31
edmondswesberglu have you looked at why the IT CI run failed for the OVS change? Does that need to be rechecked again?20:31
efriedOkay, so yeah, I'm confused.20:32
esbergluedmondsw: We need a new pypowervm release for that20:32
efriedIf we're not putting OVS in tree right now, it shouldn't be in the series ahead of SEA.20:32
esbergluefried: We are putting OVS in tree. Just not CI20:32
efriedoy vay.  Okay.20:33
efriedIf only there was a regular meeting for us to sync up on this stuff.20:33
edmondswbecause we were testing SEA in the OOT CI, and not OVS, so this was easier20:33
efriedBut if you've got OVS working now...20:33
edmondswefried if only we weren't all hazy after our time off :)20:33
edmondswwe had OVS working OOT for a long time, but we weren't testing it in CI20:33
esbergluedmondsw: OVS is working for a IT stack. But nothing has been done for an OVS CI pipeline yet20:34
edmondswagreed that we need to look at that... but nova didn't seem to care when we talked about this at the PTG20:34
edmondswesberglu right, I'm trying to explain why we're not doing OVS in CI yet20:35
edmondswit's on the TODO list, though, right?20:35
esbergluedmondsw: Yeah, just not something we were targeting for queens20:35
efriedSo when you say, "Got a working tempest run for OVS IT" -- that's not the same as having OVS CI?20:35
edmondswhe was doing some manual testing20:36
efriedI guess it's the first & hardest part; then we need to do all the crazy yaml stuff to make it actually occur in the CI?20:36
esbergluefried: Exactly20:36
efriedOkay, I'm on board now.20:36
esberglu^ per the comment on L168 we weren't aiming for OVS CI this release20:36
edmondswesberglu have you looked at why the IT CI run failed for the OVS change? Does that need to be rechecked?20:37
efriedesberglu Remind me how we patch in change sets in the CI.  Does including a change set number also pull in its dependencies?20:37
esbergluefried: Right now we list each change individually. The dependencies are not pulled in20:38
esbergluedmondsw: We need a new pypowervm cut20:38
esberglu^ we need that change20:38
edmondswso when you asked about cutting pypowervm this morning and said it was needed for the SEA IT CI change... it's really for this as well?20:40
esbergluedmondsw: Yeah both need it20:40
edmondswwas this done because of an earlier comment?20:41
esbergluedmondsw: We're trying to remove host_uuid from pypowervm. This was added as part of that effort20:48
esbergluSome of the rest calls still need a parent_uuid so we cache the sys_uuid to use for that20:48
edmondswyeah, I see that... and I guess it makes sense to do that from the get go instead of switching it IT later20:48
edmondswefried do we need to include _cleanup_orphan_adapters in the OVS/SEA patches?21:33
edmondswmdrabe any thoughts on that?21:36
esbergluedmondsw: iirc that's only needed for migration21:53
efriededmondsw Sorry, context?21:57
edmondswOOT powervm/driver.py's method21:58
efriedNo, orphan adapters should only apply to vscsi/fc21:58
edmondswoh, I thought that was networks... guess I didn't look closely enough21:58
efriedSo OOT yes.  IT not until https://review.openstack.org/#/c/526094/21:58
edmondswdon't we have some cleanup related to networks as well somewhere?21:59
efriedAnd it does apply to migration most of the time, but that's not the only way you can end up with an orphan.21:59
efriedUm, I don't think so.21:59
edmondswesberglu there's a hack at the top of vif.py in the OVS patch because we need a newer version of pypowervm... If we need that anyway for the CI now, we should go ahead and address that22:02
edmondswefried do you want to get 6121 in before we rel pypowervm or it doesn't matter?22:03
edmondswI'll admit I haven't looked at it yet22:03
edmondswsince vacay anyway22:04
efriededmondsw It's not critical.  We had a customer issue around that.  I don't remember whether they confirmed that wasn't their problem, or just haven't gotten back to me yet, but either way it's clearly not urgent.22:05
edmondswefried anything else you're aware of needing to get in before release?22:06
efriedNothing else is close to ready.22:07
edmondswesberglu couple quick things to address on the OVS patch, but nearly there... comments added22:28
esbergluedmondsw: ack22:39
edmondswesberglu also just left a question in the SEA patch that may affect OVS patch22:40
edmondswefried  I left you a couple comments on 612123:18
edmondswgotta run now23:18
efriededmondsw Ack, thx23:18
*** edmondsw has quit IRC23:19
*** edmondsw has joined #openstack-powervm23:19
*** edmondsw has quit IRC23:24
*** openstackgerrit has joined #openstack-powervm23:33
openstackgerritTaylor Jakobson proposed openstack/nova-powervm master: Refactor volume drivers  https://review.openstack.org/53081623:33

Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!