Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-03-28
14:16:57 ildikov mdbooth: I also think it would be a good discussion to the meeting we have on Thursday, so other folks can chime in too
14:17:27 ildikov I think we want to lock on the host
14:17:33 mdbooth ildikov: Ok, I'll bring it up there.
14:18:05 mdbooth Probably the os-brick part doesn't need to directly concern itself with locking: that could be done by Nova.
14:18:19 mdbooth However, I'm convinced it does need to live on the host.
14:18:56 ildikov mdbooth: I think we are in an agreement on that
14:19:08 ildikov but I might miss something now if that's not the case :)
14:19:38 pjm6 hi there
14:19:57 mdbooth ildikov: Discussion of cinder passing lists of other users of a shared connection initially sounded like a good idea, but now they make me nervous.
14:20:10 mdbooth ildikov: That information can never be safely used, so I'd prefer it didn't exist.
14:20:43 pjm6 there's a way to reset quota of openstack projects?
14:20:58 pjm6 in stable/ocata there was a bug that when deleting some VMs, the quota were not updated
14:20:58 ildikov mdbooth: we had multiple ideas on that, like sharing the info with Nova whether the driver shares the connection in general or not
14:21:00 mdbooth mriedem: Incidentally, I made a thing: https://review.openstack.org/#/c/450383/2/nova/test.py
14:21:07 mdbooth It's a poor man's compiler substitute
14:21:12 ildikov mdbooth: the other was the list of remaining attachments or smth like
14:21:33 ildikov in the new API attachment-delete does not return anything right now
14:21:38 mdbooth ildikov: Yeah, that's the dangerous info
14:21:56 mdbooth The host needs to work that out itself.
14:22:20 mdbooth Even if cinder passes it, Nova can never safely use it
14:22:30 mdbooth However, somebody might be tempted
14:23:15 openstack Launchpad bug 1670627 in OpenStack Compute (nova) ocata "duplicate for #1668267 quota is always in-use after delete the ERROR instances " [Critical,In progress] - Assigned to Matt Riedemann (mriedem)
14:23:15 pjm6 https://bugs.launchpad.net/nova/+bug/1668267
14:23:30 pjm6 this bug, theres a way to reset quota or i must do it manually?
14:23:48 ildikov mdbooth: I think the info on the driver's behavior is what's needed, the rest Nova can figure out after locking
14:23:54 ildikov mdbooth: if this makes sense
14:24:51 mriedem mdbooth: ok
14:25:06 mdbooth ildikov: With a little thought I think we can define an interface between os-brick and Nova which will allow os-brick to do just the attach/detach, and Nova to ensure correctness
14:25:27 mdbooth I suspect we'll need an is_already_connected() call
14:26:02 ildikov does your idea require states in os-brick?
14:26:06 mdbooth With that, attach, and detach, we're probably good.
14:26:29 mdbooth ildikov: Not if we implement it like this. It does require state on the host, though.
14:26:40 mdbooth But the state doesn't need to be managed by os-brick
14:29:05 ildikov mdbooth: if we can do a lock and figure out the state when we do detach that would be the best
14:29:27 alex_xu mriedem: ok, i will check that
14:29:43 ildikov mdbooth: you want to track the connections on the host and store the info?
14:30:01 mdbooth ildikov: My patch stores state, but not persistently.
14:30:29 mdbooth It inspects the host state on startup, then tracks attach/detach while running.
14:30:49 mdbooth And it has a bunch of additional sanity checking in there so it'll recover if things get out of whack anyway.
14:31:16 mdbooth The model is entirely applicable to any shared connection.
14:31:48 mdbooth All I require of the lowest level is: attach, detach, is_connected
14:37:00 ildikov mdbooth: got it, I need to look into the patch a bit
14:37:55 mdbooth ildikov: The patch as-is is specifically for drivers which use mount. Don't get hung up on that, though. The state tracking is the interesting bit.
14:38:12 ildikov yeah, that part what I meant more
14:38:32 ildikov I'm personally always get scared when we start to do state tracking... :)
14:38:44 ildikov mdbooth: :)
14:40:18 mdbooth ildikov: Amongst other things, it fixes a bug we had in the old model. We'd just try to unmount and see if it succeeded or not. Problem is, it *would* succeed if one of the instances using it was stopped. You can scan the entire host state for that every time, or you can keep track of it.
14:41:12 mdbooth Also, it meant we never knew when a failure to unmount was an error.
14:41:13 ildikov I guess scanning every time would be overkill
14:41:39 ildikov although re-scan every once in a while would still be good to make a sanity check
14:42:00 ildikov as we have issues identified with the old model, we should be able to fix that with the new
14:42:08 mdbooth My patch just has sanity checking built in to connect/disconnect
14:43:33 openstackgerrit Balazs Gibizer proposed openstack/nova master: Remove **kwargs passing in payload __init__ https://review.openstack.org/418489
14:44:23 ildikov mdbooth: oh ok, I thought it's only host startup
14:44:56 mdbooth It does a full scan at host startup, but it also has enough context to do more sensible recovery during connect/disconnect
14:45:33 ildikov ok, got it
14:45:40 mdbooth Incidentally, if you re-do the full host scan you'd need to block all connect/disconnect globally while you did it
14:46:10 ildikov that does not sound ideal
14:46:18 openstackgerrit Balazs Gibizer proposed openstack/nova master: Remove **kwargs passing in payload __init__ https://review.openstack.org/418489
14:46:24 mdbooth It's a fundamental requirement, though.
14:47:27 ildikov I just meant that if we would re-do the full scan at all connect/disconnect that does not sound ideal, I get the block
14:48:44 mriedem johnthetubaguy: i'm about to pull a +W on https://review.openstack.org/#/c/373203/ unless you think we need at least one more specs core to review it first?
14:50:01 johnthetubaguy mriedem: I mean, we have quite a few other folks +1s we could upgrade for this one
14:53:18 jaypipes so, gibi, it seems to me that since you've been doing the "top 5 notifications patches" email each week, that we've made steady progress getting patches merged. would you agree?
14:53:45 mriedem johnthetubaguy: works for me, i'll just tell everyone you told me sum(+1) == +2 :)
14:54:05 gibi jaypipes: yes. I expressed exactly this to mriedem last week on the subteam meeting
14:54:14 gibi jaypipes: so I'm happy :)
14:54:16 jaypipes gibi: cool.
14:54:20 jaypipes :)
14:54:33 gibi jaypipes: and thankful for all the reviews
14:55:08 jaypipes gibi: hey, no worries. I'm excited to get to searchlight.
14:56:03 openstackgerrit dane-fichter proposed openstack/nova master: Use cursive for signature verification https://review.openstack.org/351232
14:56:48 ildikov mriedem: :)
14:57:07 ildikov johnthetubaguy: mriedem: I rebased jgriffith's detach PoC on lyarwood's refactor patch already
14:57:27 mriedem ok
14:57:49 mriedem i'll try to review the refactor change this week between orientation stuff
14:58:14 ildikov cool
14:58:19 ildikov thank you!
15:00:10 sfinucan vladikr: comment on https://review.openstack.org/#/c/430026/3
15:01:17 mriedem alex_xu: were you working on removing api extensions in pike? i know you had a poc for removing the stevedore entries. did you want to start tracking that under https://blueprints.launchpad.net/nova/+spec/api-no-more-extensions-ocata ? we'd need to rename that to https://blueprints.launchpad.net/nova/+spec/api-no-more-extensions-pike which is fine since we did not make any progress there in ocata.
15:01:51 alex_xu mriedem: yea, I'm working on that, I will reuse that bp
15:05:21 mriedem alex_xu: ok i'll rename it quick
15:05:28 alex_xu mriedem: thanks
15:06:22 mriedem done https://blueprints.launchpad.net/nova/+spec/api-no-more-extensions-pike
15:06:23 mriedem np
15:11:14 openstackgerrit Alex Xu proposed openstack/nova master: Use plain routes list for '/servers' endpoint instead of stevedore https://review.openstack.org/445864
15:11:15 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opts in multi-create https://review.openstack.org/450831
15:11:15 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opts in BDM tests https://review.openstack.org/450830
15:11:16 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opt in config drive https://review.openstack.org/450832
15:11:17 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opt in user data https://review.openstack.org/450834
15:11:17 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opt in AZ https://review.openstack.org/450833
15:11:18 openstackgerrit Alex Xu proposed openstack/nova master: Remove the extension check for os-networks in servers API https://review.openstack.org/450837
15:11:18 openstackgerrit Alex Xu proposed openstack/nova master: Remove config opts for extension black/white list https://review.openstack.org/450836
15:11:18 openstackgerrit Alex Xu proposed openstack/nova master: Remove the usage of extension black/white list opt in scheduler hints https://review.openstack.org/450835
15:23:39 mdbooth Does anybody happen to know if we use SELinux labels when connecting an NFS volume using libvirt?
15:23:43 mdbooth kashyap: ^^^ ?
15:24:03 kashyap mdbooth: I guess we do, I recall triaging a bug in that area
15:24:08 mdbooth If we do, then it occurs to me this is another complication for multi-attach.
15:24:11 mdbooth johnthetubaguy: ^^^
15:24:19 kashyap Let me see if I can find it

Earlier   Later