20:07:45 #startmeeting 2017-05-09 runtime-spec 1.0 preparation 20:07:45 Meeting started Tue May 9 20:07:45 2017 UTC. The chair is wking. Information about MeetBot at http://wiki.debian.org/MeetBot. 20:07:45 Useful Commands: #action #agreed #help #info #idea #link #topic. 20:07:45 The meeting name has been set to '2017_05_09_runtime_spec_1_0_preparation' 20:07:58 #topic intelRdt wording 20:08:35 #link https://github.com/opencontainers/runtime-spec/pull/787 20:08:51 #topic Markdown links 20:08:54 #link https://github.com/opencontainers/runtime-spec/pull/786 20:10:39 #topic JSON Schema CI testing 20:10:47 #link https://github.com/opencontainers/runtime-spec/pull/785 20:10:56 crosbymichael: maybe this should move to runtime-tools? 20:11:06 I don't care where it lives, but want CI for missing commas and such 20:11:20 mrunalp: if it moves, there will be a lag. Maybe leave it in runtime-spec? 20:11:36 #topic RFC 2119 wording for oomScoreAdj 20:12:02 #link https://github.com/opencontainers/runtime-spec/pull/781 20:13:06 #topic RFC 2119 wording for config-linux 20:13:09 #link https://github.com/opencontainers/runtime-spec/issues/746 20:13:23 crosbymichael: I don't have time to work on this now, since I'm focused on the TTY issue 20:13:26 mrunalp: I can take a few 20:14:27 mrunalp: I think the one we just merged looks fine 20:16:34 crosbymichael: Just don't make it too explicit 20:16:45 #topic bind and rbind 20:17:44 #link https://github.com/opencontainers/runtime-spec/pull/771 20:18:18 So drop the mount(8) link, add a table with everything that gets mapped to to MS_*, and say everything else goes to data with comment separators? 20:18:21 crosbymichael: yeah 20:18:36 #topic 20:18:38 #topic RFC 2119 tightening for namespaces 20:18:44 #link https://github.com/opencontainers/runtime-spec/pull/767 20:22:00 I think we need wording about where the new namespaces are created, but am ok punting 20:27:04 #topic schema/defs-linux: Drop 'Capability' type 20:27:09 #link https://github.com/opencontainers/runtime-spec/pull/766 20:27:56 vbatts: if we're using the Linux-kernel enum directly, we don't have to define it 20:28:16 vbatts: but if we define a mapping from a string to an integer, we need to say something about it 20:28:36 crosbymichael: you'd get an error at runtime 20:29:20 tianon: if you add something that isn't implemented by your runtime because your kernel is too new, then the runtime will throw an error for a valid config 20:29:35 mrunalp: so one option is punting to the kernel and getting rid of the strings 20:29:55 crosbymichael: then you'd look at the spec and have no clue what the caps were 20:30:13 crosbymichael: I think the strings are good for visibility and usability 20:30:51 crosbymichael: I also don't want a list of valid caps in the spec, because a new kernel comes out with a new cap, and then we have to wait for a new spec version? 20:31:13 tianon: what about "reasonable effort should be made to map these strings to the kernel constants"? 20:31:22 crosbymichael: so pass-through for unrecognized strings? 20:31:33 tianon: can the runtime do this lookup dynamically? 20:31:42 ^ last line was mrunalp, not tianon 20:31:47 crosbymichael: the runtime can do that 20:32:09 crosbymichael: the spec should just say you have to report an error for unmapped entries 20:32:52 tianon: so we leave the mapping up to the runtime, but they have to match the names, and they MUST error when they can't get a mapping for whatever reason 20:32:59 mrunalp: does anyone want to take that PR up? 20:33:25 #action tianon to update the spec language for capabilities 20:34:00 wking, this may not map to solaris 20:34:13 vbatts, Solaris can add their own platform specific regex 20:34:44 but you'd need to say what you were mapping to 20:34:57 tianon: so maybe say "capability constants in the kernel" and don't get more specific 20:35:18 tianon: we have lots of CAP_ examples, so I don't think we need to be more specific about that 20:35:53 so do we revisit #766? 20:36:09 crosbymichael: we'll come back to that after we've adjusted the spec 20:36:14 vbatts: is there a non-blocker label? 20:36:56 tianon: we currently link to the capabilities man page which has a pretty simple list. 20:37:02 tianon: but I don't see this as a 1.0 blocker 20:38:09 there's a gap between "gets the point across" and "useful for compliance testing" 20:40:26 mrunalp: I don't think we need to worry about folks submitting no-op runtimes for certification 20:40:40 mrunalp: and testers would probably be taking a closer look at those 20:41:07 RobDolinMS: you can assign an issue in the cert repo to me for this 20:41:29 mrunalp: I don't expect this to be a practical issue 20:42:05 vbatts: to crosbymichael's point, even with firmer language for the Linux string, is there an issue with... 20:42:34 crosbymichael: we can hash it out on tianon's PR 20:42:41 ++ 20:42:48 #topic config: Require strictly-positive timeout values 20:42:54 #link https://github.com/opencontainers/runtime-spec/pull/764 20:43:26 crosbymichael: the earlier version of this was depointerizing, which is not what we want 20:43:36 I've spun that part out into https://github.com/opencontainers/runtime-spec/issues/772 20:44:26 #topic specs-go/round_trip_test: Add round-trip testing for the config 20:44:31 #link https://github.com/opencontainers/runtime-spec/pull/759 20:53:01 (new capabilities PR is opened at https://github.com/opencontainers/runtime-spec/pull/790 for hashing out that previous string conversion topic) 20:56:20 #topic config: Drop the redundant config.json filename requirement 20:56:26 #link https://github.com/opencontainers/runtime-spec/pull/756 20:56:47 #topic spec: Define the property-specification syntax 20:56:52 #link https://github.com/opencontainers/runtime-spec/pull/747 21:01:43 #topic config-linux: RFC 2119 wording for disableOOMKiller 21:01:49 #link https://github.com/opencontainers/runtime-spec/pull/745 21:02:06 mrunalp: the not-set lines seem like they could be pulled out (and not be per-property) 21:02:19 that sounds good to me, but I couldn't figure out the wording 21:02:56 #topic config: Lead off with the purpose of the config 21:03:04 #link https://github.com/opencontainers/runtime-spec/pull/737 21:03:10 mrunalp: please rebase 21:03:11 will do 21:03:24 #topic next steps? 21:03:34 crosbymichael: with another hour we can knock out another 10 21:03:55 mrunalp: regular call at 8am, and then another meeting for this at 2pm? 21:04:05 crosbymichael: sounds good. I need to figure out the terminal stuff 21:04:14 tianon: all I did was discover it ;) 21:04:49 vbatts: I'll see, I might be headed to the airport 21:04:58 #endmeeting