schema/config-linux: add min/max limit for resource memory and cpu

[Mashimiao]

Signed-off-by: Ma Shimiao mashimiao.fnst@cn.fujitsu.com

[cyphar]

Where do these magic numbers come from? If they come from the kernel, please document them in the config-linux.md file.

[dqminh]

https://github.com/torvalds/linux/blob/01ea443982203fcdee1250ab630ab6516f0a16e6/kernel/sched/core.c#L8326-L8327

[vbatts]

@cyphar thoughts on this now?

[tianon]

friendly ping @cyphar ❤️ 🙏

[cyphar]

I'm confused what this will look like as an actual Go structure. Will it be an interface{}?

[wking]

On Fri, Nov 04, 2016 at 08:01:49PM -0700, Aleksa Sarai wrote:

I'm confused what this will look like as an actual Go structure.

They'll be *uint64, just like they are now. The current uint64Pointer
definition is just like what @Mashimiao is adding here 1, except
with different limits on the integer case 2.

[crosbymichael]

How is this going to used at the system level?

[wking]

On Mon, Nov 07, 2016 at 09:16AM -0800,, Michael Crosby wrote:

How is this going to used at the system level?

Is that “who consumes the JSON Schema”? There's currently schema/validate.go in this repo which lets anyone check their JSON against the official schema. And I've floated opencontainers/runtime-tools#197 to lean on those JSON Schemas for runtime-tools' validation. I'd expect folks building a new config or checking a config they've just received from someone else will want to validate that config before attempting to launch a container from it.

[Mashimiao]

@opencontainers/runtime-spec-maintainers PTAL

[crosbymichael]

This would be validating system settings. My feeling is that we don't need to spec out the kernel, its not our job.

REJECT

from me

[wking]

This would be validating system settings. My feeling is that we don't need to spec out the kernel, its not our job.

I don't think we're ready for a hard line on that issue, since we currently have a fair amount of in-spec kernel documentation (e.g. here, here, here, …). But maybe @jlbutler will get us pointed more clearly in his reroll of #673.

And if we want to punt docs to the kernel, we should probably require that linux.resources.cpu.period have the same semantics as cpu.cfs_period_us. Otherwise it is not clear what we're punting to.

[Mashimiao]

@crosbymichael we also validate range of blkio.weight in schema

[mrunalp]

Agree with @crosbymichael, Let's not add limits for values.