Notebooks 2.0 // Backend // add a rate limiting

[ederign]

Respond to status with information about "retry after time" so we can implement rate liming in the future.

[google-oss-prow]

@ederign: The label(s) /label project/notebooks-v2 cannot be applied. These labels are supported: tide/merge-method-merge, tide/merge-method-rebase, tide/merge-method-squash, lifecycle/needs-triage. Is this label configured under labels -> additional_labels or labels -> restricted_labels in plugin.yaml?

In response to this:

/label project/notebooks-v2

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[ederign]

/label project/notebooks-v2

[google-oss-prow]

@ederign: The label(s) /label project/notebooks-v2 cannot be applied. These labels are supported: tide/merge-method-merge, tide/merge-method-rebase, tide/merge-method-squash, lifecycle/needs-triage. Is this label configured under labels -> additional_labels or labels -> restricted_labels in plugin.yaml?

In response to this:

/label project/notebooks-v2

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[google-oss-prow]

@ederign: The label(s) area/project/notebooks-v2 cannot be applied, because the repository doesn't have them.

In response to this:

/area project/notebooks-v2

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[ederign]

/area backend

[milinddethe15]

Hi, I propose to work on this issue.
My idea is to add a middleware RateLimiter(next http.Handler) http.Handler to wrap the app.Routes() which will track client (based on their IP?) and denies requests exceeding limit within a time frame and responds with 429 Too Many Requests.
Assign me if good to go.

[thesuperzapper]

I'm not sure that IP address can identify the client, as it's very common to have a proxy in front of the kubeflow gateway, and we don't really make any requirements about setting any "x forwarded for" headers.

I'm really not sure if we can successfully do this, because we can't identify clients.

The main thing we can do is probably just ensure that we're not overloading the kubernetes cluster, and we're already doing that largely (by using controller-runtime client so get/lists are cached).

[thesuperzapper]

I think there's already an issue for it, but we should probably consider doing some kind of caching (at multiple layers of the stack), as that will probably be more useful.

That is, if two people make the exact same list request within a few seconds of each other they should just be given the same answer.