22:01:23 #startmeeting 2016-12-14 discussion 22:01:23 Meeting started Wed Dec 14 22:01:23 2016 UTC. The chair is wking. Information about MeetBot at http://wiki.debian.org/MeetBot. 22:01:23 Useful Commands: #action #agreed #help #info #idea #link #topic. 22:01:23 The meeting name has been set to '2016_12_14_discussion' 22:01:33 #chair vbatts mrunalp 22:01:33 Current chairs: mrunalp vbatts wking 22:01:55 philips, stevvooe Joining? 22:02:01 yea 22:02:03 one sec 22:02:12 mrunalp: I am on the line 22:02:18 cyphar, Are you on? 22:04:08 #topic runtime-spec anchor patches 22:04:25 https://github.com/opencontainers/runtime-spec/pull/635 22:04:52 mrunalp: we want this to unblock some work in runtime-tools 22:05:01 mrunalp: they need maintainer review 22:05:18 #link https://github.com/opencontainers/runtime-spec/issues/634 22:05:21 ^ meta issue 22:05:41 https://github.com/opencontainers/runtime-spec/pulls/RobDolinMS 22:05:58 Although that tracking issue is stale 22:06:07 vbatts: no real objections... 22:06:16 vbatts: HTML works, but whatever 22:06:28 #topic Command-line interface 22:06:41 mrunalp: still waiting on runC and then wking will update runtime-spec#513 22:06:45 ^yup 22:07:20 mrunalp: cyphar or crosbymichael was going to open an issue about passing down the console socket... 22:07:42 mrunalp: that's the last major change before 1.0 22:07:49 mrunalp: any other runtime topics? 22:07:58 vbatts: we cut runtime-spec rc3 today 22:08:13 Congratulations! 22:08:17 vbatts: we should start populating an rc4 and/or 1.0 milestone 22:08:34 #topic next runtime-spec milestone 22:08:45 mrunalp: we can put anchors and the CLI in the next milestone 22:08:57 mikebrow: so an rc4 with those before 1.0? 22:09:07 #topic image config conversion 22:09:13 https://github.com/opencontainers/image-spec/pull/492 22:09:36 stevvooe: what about it? 22:09:42 mrunalp: is cyphar around? 22:10:00 stevvooe: I haven't voiced my concerns here yet, but we need to be careful about wording 22:10:11 stevvooe: we need to support intermediate formats unknown to OCI 22:10:27 stevvooe: imagine I have an on-disk image -> local format -> runtime config format 22:10:58 stevvooe: if we tightly bind them, even by saying "you shall convert the image config to a runtime spec", it will be impossible to implement this with intermediates like we're doing now 22:11:12 stevvooe: there's another issue about CPU shares and memory fields 22:11:26 #link https://github.com/opencontainers/image-spec/pull/495 22:11:39 stevvooe: these probably don't make a lot of sense in an image format 22:11:52 philips: what is the significance of writing it to disk? 22:12:06 stevvooe: does it end up in the config whose hash names the image? 22:12:20 philips: you're saying that if it's written to disk it effects the config digest? 22:12:32 stevvooe: yeah, and that config hash is used from the manifest 22:13:22 that sounds more like runtime-spec config -> image config, and cyphar's PR is currently focused on image config -> runtime-spec config 22:13:26 stevvooe: It's really not 22:13:34 mrunalp: anything else for image? 22:14:08 #topic project workflow 22:14:49 #link https://github.com/opencontainers/image-spec/pull/486 22:15:12 stevvooe: I float this establishe code from Docker, and wking posts lots of distracting feedback, but maintainers are silent 22:15:19 stevvooe: what should we do about this? 22:15:43 vbatts: I hear the frustration. The volume of comments is stifling (and I've heard this from many people) 22:15:58 vbatts: it feels like too much conversation that's not going anywhere 22:16:21 vbatts: this spec has been open from the beginning, without pre-planning to gather maintainer direction 22:16:35 vbatts: to exclude anyone at this point would be contrary to the past year 22:16:42 vbatts: I don't have any recommendations... 22:16:51 stevvooe: I'm happy to stay open, but this is rediculous 22:17:24 #link https://github.com/opencontainers/image-spec/pull/487 22:17:39 stevvooe: push-back on bumping Go to 1.7 for test-suite support isn't helpful 22:17:48 philips: I agree (and just merged the Go bump) 22:18:20 philips: humans operate on context, and PRs are fairly targetted. But the conversation pulls in too much context, which causes confusion and frustration 22:18:50 stevvooe: so what are we going to do? 22:19:41 stevvooe: we've discussed this in the past, and you keep trying to go to a specific issue 22:19:50 stevvooe: we need to focus on the thematic elements 22:21:19 It's hard for me to find a way to be clear with less context 22:22:07 I can communicate via PRs (and not comment on other's PRs), since that's actionable 22:22:36 vbatts: all the references are impressive, but also a big drag on a lot of conversations 22:23:05 vbatts: folks expect one line or no response at all, and then they get a few paragraphs with lots of footnotes 22:23:32 vbatts: it's difficult. 22:24:05 vbatts: merge/reject PRs are still not always helpful for quick iteration 22:24:28 stevvooe: I'm not saying, don't comment on PRs. I'm saying, try rate limiting 22:25:59 I think we need a good place for long-term planning (vs. small-step iterations) 22:26:20 stevvooe: I agree in principle, but we need a stable base building on existing tech. A lot of choices are already baked in 22:27:37 It's hard to figure out which changes are too big for 1.0 and which can fit in underneath 22:28:01 stevvooe: It's difficult to make a hard rule for that. It takes maintainer wisdom 22:28:43 stevvooe: sometime we need to disagree and commit 22:28:49 vbatts: 1.0000000000 22:29:02 * vbatts waves hands wildly 22:29:32 stevvooe: sorry to put you on the spot. Don't stop the good stuff. But let the conversations breathe 22:30:05 sarah-novotny: thanks for having this conversation in the open 22:30:22 #topic chainID PR 22:30:37 smoke bomb? 22:30:42 #link https://github.com/opencontainers/image-spec/pull/486 22:30:49 vbatts: I'm hoping to get there today 22:31:24 #topic annotations doc 22:31:36 vbatts: I have an annotations doc in the works to try and DRY up the doc 22:31:52 #link https://github.com/opencontainers/image-spec/pull/340 22:31:58 ^ related to DRY annotations 22:32:23 vbatts: there's also not much clarity on the meaning of some of the arch/platform fields 22:32:42 vbatts: if we agree that manifest-lists are just descriptor lists, that might be a hinge point 22:32:54 vbatts: we should genericize the "manifests" property name 22:33:11 vbatts: maybe someone points one at an SQLite database 22:33:27 vbatts: anything that references manifests would be further metadata 22:33:40 stevvooe: I'm not quite ready to deprecate platform 22:33:53 stevvooe: we might be able to map platfom 1:1 to annotations 22:34:16 stevvooe: but for 1.0 I think adding annotations to descriptors would be just enough (without breaking existing consumers) 22:34:22 stevvooe: did you need anything else? 22:34:29 vbatts: maybe some spec wording updates? 22:34:46 vbatts: if we stay compatible with the Docker format and just change the spec wording, I understand that 22:35:13 vbatts: so keep "manifests" but weaken the manifest assumption in the surrounding spec language 22:35:19 philips: I think that plan seems fine to me 22:35:49 stevvooe: do you want to resumbit the PR for this discussion? 22:35:58 stevvooe: or do we need to make other decisions first? 22:36:05 vbatts: I have another comparison PR 22:36:26 vbatts: the only other piece would be the index.json file or embedding in oci-layout 22:36:54 stevvooe: I'm fine with index.json 22:37:15 stevvooe: we declare the media type for index.json and say "it might have these annotations" as a secondary step 22:37:21 stevvooe: I don't see a need to make it more complex 22:37:47 vbatts: I'll work that up and push it up later 22:38:15 vbatts: no other topics outside of some outstanding review 22:38:18 philips: me neither 22:38:28 #topic next meeting 22:38:32 mrunalp: call next week? 22:38:35 vbatts: not me 22:38:38 philips: me neither 22:38:44 mrunalp: Jan 4th for the next call? 22:38:46 philips: sure 22:38:50 stevvooe: fine 22:38:57 vbatts: not me, but it's fine if you have one then 22:39:03 mrunalp: See you next year 22:39:06 #endmeeting