| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-04-21 | |||
| 08:43:11 | lyarwood | in v2 | |
| 08:44:28 | mdbooth | johnthetubaguy: The problem with having differing scope of locks is that it's hard to represent that across an api boundary | |
| 08:45:17 | mdbooth | I mean, if backend A requires X to be serialised only with itself, backend B requires X and Y to be serialised, and backend C has no such restrictions | |
| 08:45:39 | mdbooth | How do you abstract that for the caller? | |
| 08:46:03 | mdbooth | I think the locking would need to live in the backend | |
| 08:46:41 | asettle | cdent: you around? :) | |
| 08:46:52 | cdent | asettle: I just showed up, what's up? | |
| 08:46:58 | asettle | Magic@ | |
| 08:47:03 | asettle | Placement API :( | |
| 08:47:28 | openstack | Launchpad bug 1683585 in openstack-manuals "Placement endpoint configuration incorrect as written" [Undecided,New] - Assigned to Brian Moss (bmoss) | |
| 08:47:28 | asettle | Any chance you could take a look at this bug to help out Brian and myself? https://bugs.launchpad.net/openstack-manuals/+bug/1683585 | |
| 08:48:11 | lyarwood | mdbooth: so just serialise all volume operations (both local n-cpu and remote via c-api) for all backends? | |
| 08:48:31 | mdbooth | lyarwood: You could do that, but that's a big hammer | |
| 08:48:48 | mdbooth | Akin to a GIL ;) | |
| 08:49:01 | lyarwood | GBDML | |
| 08:49:11 | cdent | asettle: ah, that. Yeah, so the issue is that there hasn't been consensus between nova, devstack, deb packagers, rpm packages on "where" placement is supposed to live | |
| 08:49:19 | mdbooth | hehe | |
| 08:49:37 | lyarwood | but yeah, massive hammer, slows everything down | |
| 08:49:38 | asettle | cdent: yep. Which is causing issues, it's our second bug of the same nature. We're going to have to chat about reversing it, I think. IT's just not working. | |
| 08:49:41 | cdent | asettle: I would guess that what needs to happen is find out from the packages where they are putting it (it can go wherever they like), and make the docs reflect that | |
| 08:49:56 | cdent | s/packages/packagers/ | |
| 08:50:17 | mdbooth | Suspect the problem would mostly be that 1 slow/malfunctioning thing would kill everything | |
| 08:50:34 | cdent | asettle: I can reflect this on the bug report too | |
| 08:50:37 | asettle | cdent: who can we get to sort out the packaging then? | |
| 08:51:22 | cdent | asettle: I don't have the names at the top of my head. johnthetubaguy and mdbooth you know who the contacts are for ubuntu and rdo packages? | |
| 08:51:59 | mdbooth | lyarwood: ^^^ ? | |
| 08:53:55 | mdbooth | cdent: You looking for a default port number? | |
| 08:54:15 | cdent | mdbooth: no | |
| 08:54:41 | cdent | looking for agreement between the various parties on where it will be placed, or at least authoritative statements on how things are current configured | |
| 08:55:09 | asettle | cdent: thanks for looking into it. Contextually, could you please explain to me why this was implemented without consensus? Am I msising something? :/ | |
| 08:55:21 | cdent | hmmm | |
| 08:55:45 | cdent | I'm pretty sure that the rpm and debs use the same port. | |
| 08:55:45 | mdbooth | cdent: Being dim, just trying to understand what 'where' means in this context. | |
| 08:55:51 | mdbooth | where == port number? | |
| 08:55:58 | asettle | Okay, well, that's good? | |
| 08:56:15 | alex_xu | johnthetubaguy: yea, agree with that. begin to do some cleanup https://review.openstack.org/458738 https://review.openstack.org/458739 | |
| 08:56:50 | bauzas | asettle: mmm I do remember a problem with RDO packaging for the above bug, sec | |
| 08:57:06 | asettle | bauzas: The same issue we were fiddling with a few weeks back? | |
| 08:57:06 | cdent | The issue entered into the discussion because there is a movement to stop using ports for hosting each of the individual services and instead put them on default http (80) or https (443) ports on the same controller host, distinguishing between services by a path prefix | |
| 08:57:16 | bauzas | not using ports means you have to have your virtualhost able to accept on this location | |
| 08:57:24 | openstackgerrit | jichenjc proposed openstack/nova master: [placement] Add doc for resource provider usages https://review.openstack.org/450105 | |
| 08:57:49 | cdent | asettle, bauzas: there have been a few different bugs related to the configuration, for example, needing <Directory> | |
| 08:58:10 | bauzas | cdent: yup, AFAIR, it was related to that | |
| 08:58:16 | asettle | Okay. | |
| 08:58:23 | bauzas | I just have a shit ton of BZs to look at | |
| 08:58:43 | cdent | asettle: I commented about those prefixes on the reviews, but it wasn't meant as a way of saying "do this", but rather as a way of saying "we should be moving in this direction, eventually" | |
| 08:58:58 | bauzas | a-ha! think I found it | |
| 08:59:00 | cdent | (like a warning of future change) | |
| 08:59:13 | bauzas | cdent: asettle: Red Hat Bugzilla – Bug 1434944 | |
| 08:59:15 | bauzas | oops | |
| 08:59:23 | openstack | bugzilla.redhat.com bug 1434944 in openstack-nova "[Ocata] openstack-nova-placement-api-15.0.0-1.el7.noarch missing <Directory> parameter" [Medium,Assigned] - Assigned to sferdjao | |
| 08:59:23 | bauzas | https://bugzilla.redhat.com/show_bug.cgi?id=1434944 | |
| 08:59:43 | asettle | I always liked the name 'bugzilla' | |
| 08:59:51 | asettle | cdent: okay, thanks, that's a start :) | |
| 09:00:27 | asettle | Ah, this was after I pushed sgordon with a stick | |
| 09:00:34 | bauzas | asettle: indeed :) | |
| 09:00:47 | bauzas | that came back to my ears a week or so ago | |
| 09:00:54 | asettle | Thanks for finding this, bauzas | |
| 09:01:15 | bauzas | so I'd say the problem is on track for RDO, fix needs to go on its way | |
| 09:01:15 | asettle | Okay, looks like Petr is on this. | |
| 09:01:52 | bauzas | asettle: FWIW, Red Hat product-based OpenStack Platform is not impacted because we use TripleO | |
| 09:01:52 | cdent | bauzas: this issue is different: it is making sure that the docs are aligned with the rdo config | |
| 09:02:15 | cdent | the directory thing causes a different issue | |
| 09:02:19 | bauzas | asettle: where Puppet nukes that file and replaces by a correct config | |
| 09:02:26 | asettle | Okay | |
| 09:02:38 | bauzas | cdent: probably, have very few context | |
| 09:02:48 | bauzas | cdent: just saying that RDO .rpms aren't working anyway | |
| 09:02:52 | cdent | here's another semi-related thing the other day that I made: https://bugzilla.redhat.com/show_bug.cgi?id=1441200 | |
| 09:02:53 | openstack | bugzilla.redhat.com bug 1441200 in openstack-nova "openstack-placement-api.conf not configured properly by the openstack-nova-placement-api rpm" [Unspecified,Post] - Assigned to nlevinki | |
| 09:03:13 | openstack | bugzilla.redhat.com bug 1441208 in openstack-nova "placement api apache configuration does not align with documentation, ensure system does not work" [High,Closed: duplicate] - Assigned to nlevinki | |
| 09:03:13 | cdent | sorry, I mean this one: https://bugzilla.redhat.com/show_bug.cgi?id=1441208 | |
| 09:03:49 | bauzas | cdent: BZ1441200 points to the same fix for RDO than 1434944 for resolution | |
| 09:03:55 | cdent | the first one is a duplicate of the one I created, mine has more context | |
| 09:04:17 | cdent | bauzas: I'm providing these so asettle has context, which is what was asked for, not to find the right bug | |
| 09:04:28 | asettle | Yep, context me ahoy. | |
| 09:04:31 | asettle | Thanks cdent | |
| 09:04:40 | bauzas | cdent: np, just trying to understand the missing bits and the resolution path | |
| 09:05:21 | cdent | asettle: in my bug report there are a lot of lurking "warnings" | |
| 09:06:40 | bauzas | anyway, seems like the RDO fix needs to be aligned with docs, or the other way | |
| 09:06:52 | bauzas | since both are pulling on two different directions | |
| 09:07:07 | asettle | I completely agree. | |
| 09:07:22 | bauzas | but FWIW, whatever lands in RDO needs to properly amend the upstream docs, I do agree | |
| 09:07:35 | bauzas | if those docs become wrong of course | |
| 09:07:36 | asettle | Provided we're not going to have substantially broken docs for another month, I'm happy to do what we need to do. | |
| 09:10:56 | bauzas | asettle: cdent: we have an internal bug triaging meeting today, will raise it up | |
| 09:11:05 | cdent | bauzas++ | |
| 09:11:22 | asettle | Thank you bauzas :) | |
| 09:11:32 | asettle | Could you keep us up to date on the launchpad bug if I'm not around? | |
| 09:11:39 | asettle | I mean, I'm sure I will be :( | |
| 09:11:39 | bauzas | we = Nova team in Red Hat, not RDO team | |
| 09:12:30 | bauzas | asettle: that's at 1400UTC | |
| 09:12:37 | asettle | Lovely. Just in time for tea. | |
| 09:12:59 | cdent | asettle: I think it is pretty clear that the communication for "how to do placement" went awry to many parties over the past... six months? So we can probably do some kind of retrospective or something to figure out what to do better in the future. From my standpoint I didn't have enough insight into the process and made too many assumptions about it being under control | |
| 09:13:42 | bauzas | cdent: well, mostly because that's deployment choices | |
| 09:14:09 | bauzas | cdent: so that's hard to find a common denominator for all | |
| 09:14:37 | bauzas | and we all say that Service Catalog serves the purpose to enable services the way you want | |
| 09:14:49 | cdent | bauzas: sure, but I think there's been a lot of confusion that we could have made less of a problem. The <Directory> thing, for example, seems like something that never should have happened because when that's wrong things simply don't work. | |
| 09:15:10 | bauzas | so, you, as a deployer, just needs to make consistent what you have in your service catalog entry and what you configure server-side | |
| 09:15:41 | asettle | cdent: thanks for your honesty. What can I do to help? | |
| 09:16:00 | bauzas | cdent: the port discussion clearly showed up that a lot people were having wrong expectations | |