Allow support for a HTTPLocalConcurrencyLimitPolicy similar to HTTPLocalRateLimitPolicy

[himanshu-garg-zepto]

What problem are you trying to solve?

We want to support concurrency limit on requests and that too in a dynamic way similar to how the rate limit policy works. Currently the concurrency can be primarily controlled at the application layer using back-pressure but we would like to have this as policy at the mesh layer.

How should the problem be solved?

I was thinking we'll need the following steps:

• add a new HTTPLocalConcurrencyLimitPolicy crd
• add http_local_concurrency_limit parallel to http_local_rate_limit in linkerd2-proxy-api/proto/inbound.proto
• add the flow in control plane to watch for k8s crd and serving it in its api
• add the flow in proxy to add a concurrency limiter based on the limit provided by the control plane

Any alternatives you've considered?

Currently we're relying on the global concurrency limit (default 10_000)

How would users interact with this feature?

No response

Would you like to work on this feature?

yes

[himanshu-garg-zepto]

here are the PRs for early review (I'm a new contributor)

linkerd/linkerd2-proxy-api#527
#14910
linkerd/linkerd2-proxy#4416

Will be testing this in our QA cluster for the feature and cross-compatibility and with some reasonable load

[cratelyn]

thank you for filing an issue and for driving this work, @himanshu-garg-zepto. i will note this proposal to the other linkerd maintainers, and determine who might be a good shepard to review this work.

[himanshu-garg-zepto]

thanks a lot @cratelyn

[himanshu-garg-zepto]

Hey @cratelyn were you able to find someone who could guide me through this..?