17:01:55 #startmeeting 2016-07-01 discussion 17:01:55 Meeting started Wed Jun 1 17:01:55 2016 UTC. The chair is vbatts. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:01:55 Useful Commands: #action #agreed #help #info #idea #link #topic. 17:01:55 The meeting name has been set to '2016_07_01_discussion' 17:02:16 vbatts: chair me? 17:02:36 #chair wking 17:02:36 Current chairs: vbatts wking 17:02:38 thanks 17:03:04 #topic 1.0 issues 17:03:15 #topic runtime 1.0 issues 17:03:26 mrunalp: is crosbymichael on? 17:03:37 mrunalp: maybe image spec first 17:03:44 #topic image 1.0 issues 17:03:56 philips: plan is to cut 0.2 stuff today 17:04:08 https://github.com/opencontainers/image-spec/milestones/v1.0.0-rc 17:04:18 actually, vbatts cut 0.2 already 17:04:28 https://github.com/opencontainers/image-spec/issues/11 17:04:45 philips: we probably want to punt this^ until after 1.0 17:05:00 remaining 0.2 issues: https://github.com/opencontainers/image-spec/milestones/v0.2.0 17:05:16 https://github.com/opencontainers/image-spec/releases/tag/v0.2.0 17:05:18 :-) 17:05:20 philips: doing 1.0 before naming is sorted is still valuable 17:05:49 #topic runtime 1.0 issues 17:05:58 mrunalp: create/start landed last Thursday :) 17:06:08 https://github.com/opencontainers/runtime-spec/milestones/1.0.0 17:06:22 https://github.com/opencontainers/runtime-spec/pull/449 17:06:23 we're looking over these PRs now^ 17:08:01 mrunalp: any comments on #449? 17:08:05 vbatts: moved the two remaining issues from v0.2.0 to v1.0.0-rc https://github.com/opencontainers/image-spec/milestones/v1.0.0-rc 17:08:12 nothing that I haven't already said on the PR ;) 17:08:23 dqminh_: #449 looks ok 17:08:42 crosbymichael: I think PullApprove is making it look like it needs a rebase 17:08:56 mrunalp: Is this a PullApprove issue or a GitHub issue 17:09:07 crosbymichael: the merge button still goes up. I'm not sure what the issue is 17:09:23 crosbymichael: did we change a setting? 17:09:29 mrunalp: I didn't, can we check? 17:09:50 crosbymichael: ah, master is a protected branch, and there's a checkbox that says PR branches have to be up-to-date before merging 17:10:03 vbatts: It's still going to check for conflicts though, right? 17:10:05 crosbymichael: yeah 17:10:11 crosbymichael: I don't think we need it 17:10:16 vbatts: I don't think we need it either 17:10:19 mrunalp: I agree 17:10:31 crosbymichael: I unchecked the up-to-date requirement 17:10:47 https://github.com/opencontainers/runtime-spec/pull/412 17:11:19 I haven't had new ideas on this 17:11:31 mrunalp: Maybe we can go back to the original, sparser namespace docs 17:11:38 https://github.com/opencontainers/runtime-spec/pull/397 17:11:46 #action wking to roll back #412 namespace description 17:12:08 #topic add support for cgroup namespace (#397) 17:12:26 #link https://github.com/opencontainers/runtime-spec/pull/397 add support for cgroup namespace 17:13:13 trevor: We need some way to know if your kernel is going to be compabible 17:13:27 Mruanl: The runtime can detect if a particular namespace is supported 17:13:59 ???: I agree, seems like having the tool validate is plenty 17:14:02 tianon: I agree that having the tool validate should be plenty 17:14:16 ??? = tianon 17:14:42 https://github.com/opencontainers/runtime-spec/issues/299 17:14:43 RobDolinMS, do you want to take over for notes, or should I keep going? 17:15:04 #topic Separate container sandbox lifecycle from that of the processes inside it #299 17:15:18 mrunalp: what are we doig about #299 after create/start. Is there anything else? 17:15:32 vbatts or wking - could one of you #chair me so I can update topic too? 17:15:35 vishh: the second part of #299 was post-stop hooks, and we still don't know what we're doing there 17:15:38 #chair RobDolinMS 17:15:38 Current chairs: RobDolinMS vbatts wking 17:15:58 #link https://github.com/opencontainers/runtime-spec/issues/299 Separate container sandbox lifecycle from that of the processes inside it #299 17:16:13 vishh: maybe we can punt on post-stop for 1.0 17:16:32 vishh: for now, folks who need that sort of thing will need a deeper understanding 17:16:44 vishh: user may need to reverse-engineer to get post-stop hooks to work 17:16:55 mrunal: I agree we should punt from 1.0 17:17:07 https://github.com/opencontainers/runtime-spec/issues/20 17:17:17 RobDolinMS: I'm just going to let you take notes ;) 17:17:20 #agreed Move #299 out of 1.0 17:17:43 wking - let's trade-off. I'll tag-in for now :) 17:17:59 wking: runc now has a run command that does create and start together 17:18:28 wking: We talked about having the more unified command in #384 but it got pulled 17:18:36 mrunal: Michael, what are your thoughts? 17:18:44 #link https://github.com/opencontainers/runtime-spec/issues/473 17:18:49 Michael: It's a convenience command, it may not need to be in the spec 17:18:50 ^ me suggesting we drop hooks 17:19:23 wking: One of the other arguments for pre-start hooks was mrunal's pre-pivot mount scenario 17:19:36 wking: would like to better understand use case there 17:19:56 mrunal: question is what do we do for 1.0 17:21:03 wking: spec gets much simpler if we just get rid of hooks for v1.0 17:21:15 #action wking to file a hook-dropping PR 17:21:42 vishh: Would it be OK to use post-stop hooks in runc? 17:21:52 Michael: how would you persist the namespaces? 17:23:11 vishh: It's more about usability that not actually being able to do it 17:23:35 Mrunal: Are we going to continue supporting hooks in runc or are we going to get rid of them entirely? 17:24:06 crosbymichael: how do I use ‘runc create’ and ‘runc start’ and then be attached to the container? I can’t seem to figure out the right combo of options 17:24:39 Michael: If we're removing it, remove and everyone can orchestrate through create, start, delete 17:25:29 Julz: If they're still in config.json, it's odd for one runtime to support and another not to support 17:26:17 Julz: Maybe have in a seperate hooks.json file 17:26:24 Mrunal: Annotations are another way to support 17:26:41 wking: There is not a spec for the orchestration layer 17:27:23 wking: there is some lack of clarity of what should be put into an image 17:27:31 Julz: Most people use a hook for networking 17:27:52 wking: Do we have a networking spec? 17:27:59 Julz: There are already several 17:29:07 wking: so there will be a config file for networking 17:29:18 Julz: The networking should be de-coupled 17:30:34 Mrunal: OK, so what is the decision? 17:30:59 Michael: Either have it and support it, or don't. It's not good to be in a middle ground. 17:32:07 Sorry Question: So Pre-Post Hooks are being taken out of the spec, but the functionality will be there ? 17:32:18 I'm interested in this for user space networking. 17:32:37 Julz: I would prefer for it not to be in config.json 17:32:43 Julz: Fine as a flag for runc 17:34:16 Vish: If you're going to have some hooks, is it better to support of have people work through namespaces? 17:34:39 Michael: As far as the spec is concerned, run doesn't exist so everyone should be using the create-start-delete flow 17:35:34 Brandon: We need to have implementers notes on this stuff 17:35:42 Vish: We could solve this with tooling too 17:35:57 Vish: It's not that we're preventing people from doing this, we're just not making this obvious 17:38:13 +1 to punt for now 17:38:24 Vish: I hope adding features to the spec wouldn't break existing users. 17:38:38 Julz: For now, it's undefined behavior right now. 17:39:44 Vish: I prefer us supporting post-stop as a first class thing 17:40:04 wking: You could have sub-containers as a possible work-around 17:40:15 wking: At least on Linux 17:41:18 Vish: If OCI Tools can gurantee a post-stop hook can run, that's fine 17:42:52 Vish: Does anyone have a proposal for how to handle this 17:44:59 Julz: What about exec'ing another container? 17:47:01 for 1.0 I think we really need these usecases written up in the docs - e.g. exec 17:47:16 so people know how to do them with v1.0 17:47:19 Vish: We might want to add language into the spec in the future and that should not be a blocker. 17:48:32 #link https://github.com/opencontainers/runtime-spec/pull/391 17:48:34 dug: these seem like pretty common use cases. We should describe how to do this with 1.0 of the spec 17:48:40 dug: It should be there for 1.0 17:48:42 previous cookbook discussion in #391 17:49:30 duglin: spec should include the high-level notion of how to do something like exec. 17:49:59 duglin: could include an example of how to do this with runc 17:50:40 duglin: will help with documentation, but may have questions 17:51:15 Mrunal: what do other maintainers think? 17:51:24 Vish: Having clear language would be helpful 17:52:25 wking: where should I re-submit #391? 17:52:45 Vish: Would be good to have 2-3 recepies that could be compared 17:53:44 #action wking to reroll runtime-spec#391 to include multiple examples and add it to a runtime-spec subdir 17:54:18 Michael: I'm not sure about location, but have no objection 17:56:02 Mrunal: What about a use cases subdirectory with one file per use case? 17:56:16 RobDolinMS: Examples / use cases should be in runtime-spec 17:56:37 Mrunal: should we include this for 1.0? 17:56:48 Mrunal: We shouldn't rush 1.0 17:57:03 Michael: It shouldn't block 17:57:29 Duglin: I agree that examples shouldn't block the release, but we don't have a guarantee that what we claim you can do with the spec, you can actually do 17:57:45 Duglin ... and that the workflow makes sense 17:58:24 Duglin: This is sort of like regression testing 17:59:02 #topic OCITools 17:59:12 #link https://github.com/opencontainers/ocitools/issues/83 18:00:49 RobDolinMS: Do we want tooling to be up-to-date with spec or have tooling follow/offset from spec? 18:00:56 I need to drop for another call 18:01:36 Brandon: releasing the spec without tooling doesn't make a ton of sense. 18:02:08 wking: what about Solaris? 18:02:33 wking: it's a bigger change to the spec if you have to PR the test suite 18:03:38 #action more discussion in ocitools#83 18:03:56 #link https://github.com/opencontainers/ocitools/issues/83 18:04:42 #topic When to cut next runtime-spec release 18:05:27 Michael: Most of the stuff outstanding is relatively small 18:06:56 RobDolinMS: either v0.6 or v1.0-RC0; RC0 indicating there may be further RC's 18:07:40 #topic v1.1 and v2.0 18:08:04 #info Michael: We can create milestones and start moving some issues to those 18:12:55 #endmeeting