Add I/O Priority Configuration for process group in Linux Containers

[utam0k]

Description

This PR introduces a new configuration option to set the I/O priority for the entire process group within a Linux container. By allowing users to configure I/O priority, we enable better resource management and performance tuning for various workloads, particularly in Kubernetes environments where multiple containers may be running concurrently.

One specific use case is when running batch processing jobs alongside application Pods in a Kubernetes cluster. Batch jobs are often I/O-intensive and can cause resource contention, negatively affecting the performance of other containers running on the same node. By allowing I/O priority configuration for each container, cluster administrators can minimize the impact of I/O-intensive batch jobs on application Pods.

By adding I/O priority configuration to the OCI Runtime spec, we enable Kubernetes (and other container orchestration platforms) to take advantage of this feature and better manage container I/O resources. This can lead to improved overall performance and resource utilization on nodes running multiple containers with different I/O requirements.

Implement

This feature can be implemented using the ioprio_set system call on Linux, which allows setting the I/O priority for a given process or process group. In Go, the syscall package can be used to invoke the ioprio_set system call. A sample implementation in Go could look like this: https://go.dev/play/p/KsscUN8RITD

[AkihiroSuda]

• https://github.com/opencontainers/runtime-spec/blob/main/schema/config-linux.json has to be updated too.
• I'd like to see a POC of runc, crun, or youki before merging this

[utam0k]

@AkihiroSuda Thanks for your quick review 😻

https://github.com/opencontainers/runtime-spec/blob/main/schema/config-linux.json has to be updated too.

Thanks, I'll update it later.

I'd like to see a POC of runc, crun, or youki before merging this

I might try this against youki in my personal repository. This will take some time, so you'll have to wait.
I'd like to hear the maintainer's first impressions of the proposal. If it's not worth it from the start, there's no point in having a prototype implementation.

[utam0k]

@AkihiroSuda I have implemented it to runc.
opencontainers/runc#3783

[AkihiroSuda]

Thanks

[AkihiroSuda]

Please update https://github.com/opencontainers/runtime-spec/blob/main/schema/config-linux.json

[utam0k]

Please update https://github.com/opencontainers/runtime-spec/blob/main/schema/config-linux.json

@AkihiroSuda
I completely forgot about it 🤦I have updated it.

[AkihiroSuda]

@opencontainers/runtime-spec-maintainers PTAL

[giuseppe]

should it be an enumeration of the supported values?

[utam0k]

@giuseppe Thanks for your review. It would be better! I have fixed it.

[giuseppe]

LGTM

[thaJeztah]

I think the enum is defined in the wrong location, because this enum would now apply to ioPriority as a whole, but the enum defines values for class only.

So class should probably be defined as "#/definitions/IOPriorityClass" (e.g.) instead of a "string". Similar to

runtime-spec/schema/defs-linux.json

Line 102 in 720792f

"$ref": "#/definitions/SeccompOperators"

runtime-spec/schema/defs-linux.json

Lines 77 to 88 in 504f70e

	"SeccompOperators": {
	"type": "string",
	"enum": [
	"SCMP_CMP_NE",
	"SCMP_CMP_LT",
	"SCMP_CMP_LE",
	"SCMP_CMP_EQ",
	"SCMP_CMP_GE",
	"SCMP_CMP_GT",
	"SCMP_CMP_MASKED_EQ"
	]
	},

[thaJeztah]

Oh! I think #1206 was meant to address that, correct @giuseppe ?