enable istio as a provider + configuring destinationRule

[Gregory-Pereira]

cc @liu-cong

[linux-foundation-easycla]

The committers listed above are authorized under a signed CLA.

• ✅ login: Gregory-Pereira / name: Gregory Pereira (f9eef08, 4e84c84, 050fd49)

[k8s-ci-robot]

Welcome @Gregory-Pereira!

It looks like this is your first PR to kubernetes-sigs/gateway-api-inference-extension 🎉. Please refer to our pull request process documentation to help your PR have a smooth ride to approval.

You will be prompted by a bot to use commands during the review process. Do not be afraid to follow the prompts! It is okay to experiment. Here is the bot commands documentation.

You can also check if kubernetes-sigs/gateway-api-inference-extension has its own contribution guidelines.

You may want to refer to our testing guide if you run into trouble with your tests not passing.

If you are having difficulty getting your pull request seen, please follow the recommended escalation practices. Also, for tips and tricks in the contribution process you may want to read the Kubernetes contributor cheat sheet. We want to make sure your contribution gets all the attention it needs!

Thank you, and welcome to Kubernetes. 😃

[k8s-ci-robot]

Hi @Gregory-Pereira. Thanks for your PR.

I'm waiting for a kubernetes-sigs member to verify that this patch is reasonable to test. If it is, they should reply with /ok-to-test on its own line. Until that is done, I will not automatically test new commits in this PR, but the usual testing commands by org members will still work. Regular contributors should join the org to skip this step.

Once the patch is verified, the new status will be reflected by the ok-to-test label.

I understand the commands that are listed here.

Details

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-sigs/prow repository.

[netlify]

✅ Deploy Preview for gateway-api-inference-extension ready!

Name	Link
🔨 Latest commit	050fd49
🔍 Latest deploy log	https://app.netlify.com/projects/gateway-api-inference-extension/deploys/68d042d24286f5000897277a
😎 Deploy Preview	https://deploy-preview-1381--gateway-api-inference-extension.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify project configuration.

[Gregory-Pereira]

cc @SergeyKanzhelev @ahg-g

[kfswain]

Meta-question here; does a destination rule have to be provided at the time of inferencePool creation? I don't fully understand the usecase here, and so, to me, it looks like we are just extracting all the DestinationRule fields into the helm chart. That may set a pattern we don't want, where the helm chart is essentially just a wrapper over the direct K8s manifest.

Said another way; what are we simplifying by exposing the entire CRD in the helm chart?

[Gregory-Pereira]

Meta-question here; does a destination rule have to be provided at the time of inferencePool creation?

The key reason why I wanted this change, is because to properly route traffic with Istio we need a DestinationRule. The .host value for that DestinationRule needs to be the EPP service address, and so the DestinationRule is dependent on the name of the EPP service and thus the release name used when installing the GIE charts.

What we do currently (which is definitely not ideal) is create the destination rule in the infra charts, (mapping gateway traffic to a service address that does not exist yet). Then we create the release of the model-service the GIE charts which will create the other pieces of the stack, and we apply a bogus label to the HTTPRoute and or Gateway to force reconciliation.

Said another way; what are we simplifying by exposing the entire CRD in the helm chart?

In my eyes we are simplifying the user experience. AFAICT, this chart does not support extraDeploy or any method to apply or track miscellaneous but related manifests. The alternatives I see here are creating an individual DestinationRule out of band after chart application, which relies on scripting rather than coordinating helmfile releases, or guessing the service address ahead of time in infra and applying it there.

Definitely open to alternatives if you had anything better in mind.

[pierDipi]

Do we have to use Istio specific APIs (DestinationRule) or standard GW APIs can be used instead?
(for example: BackendTrafficPolicy and BackendTLSPolicy
https://gateway-api.sigs.k8s.io/api-types/backendtrafficpolicy/
https://gateway-api.sigs.k8s.io/api-types/backendtlspolicy/?h=backendtlspolicy)

[Gregory-Pereira]

Definitely open to it if you want to contribute an alternative, this is just beyond the scope of this initial implementation for integration with llm-d

[liu-cong]

Great discussions between @Gregory-Pereira and @kfswain .

The part that it creates a DestinationRule pointing to the EPP service makes sense to me (unless there are better ways to do it), similar to how we create the GCPBackendPolicy and HealthCheckPolicy for gke. To me these are necessary resources to make things work, while they are not "mandatory" info that users need to know unless they want to customize.

But if that's our goal (hide unnecessary complexity for most initial use cases), then we should probably provide a [optional] default config that works out of the box, instead of asking the user to provide the entire DestinationRule config. The current gke example provides default config that works out of the box. Then the question becomes what's the minimal DestinationRule config required to get things to work? Is it just the host part?

[danehans]

This PR needs a link to an issue or a more detailed description of what the problem is and how this PR fixes it.

/hold

[danehans]

xref: #1475 that refactors the setup guide and may affect this PR.

[Gregory-Pereira]

Any update on this? I would like to know if the community is open to this feature, to prevent me from continually having to rebase if there is no interest here.

[nirrozenbaum]

Any update on this? I would like to know if the community is open to this feature, to prevent me from continually having to rebase if there is no interest here.

I’d be happy to see this PR gets in. Istio is one of our main partners and ideally installing IGW through helm with istio as gateway should work out of the box. without this PR it’s required to create the DestinationRule manually.

[nirrozenbaum]

/ok-to-test

[kfswain]

/approve

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: Gregory-Pereira, kfswain

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Details

Needs approval from an approver in each of these files:

• OWNERS [kfswain]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[liu-cong]

/unhold

/lgtm