remove Model field from LLMRequest #782

nirrozenbaum · 2025-05-05T06:06:40Z

This PR removes the Model field from Scheduling.LLMRequest struct.
from the scheduler point of view, it doesn't care about the original requested model name, only about the resolved model after traffic splitting that was done in a higher level.
we can see that Model field was used in unit-tests only and that it's always set to be identical to ResolvedTargetModel. Scheduler doesn't use this field.

In addition to removing Model field from LLMRequest struct, this PR renames ResolvedTargetModel to TargetModel in LLMRequest from the same reasons. scheduler plugins don't care about traffic splitting and "resolved" model, only about what is target model.

unit-tests and other usage places were updated accordingly.

nirrozenbaum · 2025-05-05T06:07:00Z

cc @kfswain

netlify · 2025-05-05T06:07:05Z

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

Name	Link
🔨 Latest commit	`0d00997`
🔍 Latest deploy log	https://app.netlify.com/sites/gateway-api-inference-extension/deploys/6820d800237f460008e56177
😎 Deploy Preview	https://deploy-preview-782--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 site configuration.

LukeAVanDrie · 2025-05-05T17:38:53Z

I am working on the Flow Controller POC. We currently enforce the dispatch policy (for fairness) by model, not resolved target model. I will either need to use a different type when enqueuing a request than LLMRequest or will need model to remain as a field on this struct.

The Flow Controller is not tightly coupled to the Scheduler, so either approach is fine.

nirrozenbaum · 2025-05-05T17:50:19Z

I am working on the Flow Controller POC. We currently enforce the dispatch policy (for fairness) by model, not resolved target model. I will either need to use a different type when enqueuing a request than LLMRequest or will need model to remain as a field on this struct.

The Flow Controller is not tightly coupled to the Scheduler, so either approach is fine.

@LukeAVanDrie thanks for bringing this up.
can you elaborate on what other fields you have in Flow Controller? (do you have a link to your git branch?)
I mean, do you have a subset of the fields that are used in both? do you have additional fields that are not used in Scheduler and you need in Flow Control?

Ideally, each layer should get the data it needs for its mission, and only that.
if there is a complete match (with the exception of Model field) then sure, let's leave it as is.
but I suspect you might need only parts of the fields or additional fields

kfswain · 2025-05-05T22:14:27Z

but I suspect you might need only parts of the fields or additional fields

To my understanding, @LukeAVanDrie needs the Model name, as that is the unique identifier we currently use to separate one 'use case' from another.

Ideally, each layer should get the data it needs for its mission, and only that.
Completely agreed here. That is the endstate

pkg/epp/handlers/request.go

LukeAVanDrie · 2025-05-05T23:55:30Z

@LukeAVanDrie thanks for bringing this up. can you elaborate on what other fields you have in Flow Controller? (do you have a link to your git branch?) I mean, do you have a subset of the fields that are used in both? do you have additional fields that are not used in Scheduler and you need in Flow Control?

A subset (model, criticality) and some new fields: a reference to the request prompt size in bytes and a reference to the request context.

Ideally, each layer should get the data it needs for its mission, and only that. if there is a complete match (with the exception of Model field) then sure, let's leave it as is. but I suspect you might need only parts of the fields or additional fields

This is a good point. I have removed my dependency on scheduling.LLMRequest and am defining my own type for the FlowController input. My POC's fate is no longer tied to this PR.

nirrozenbaum · 2025-05-08T07:47:07Z

ready for review after #781 was merged
/unhold

nirrozenbaum · 2025-05-08T07:47:27Z

pkg/epp/requestcontrol/director.go

 		if reqCtx.ResolvedTargetModel == "" {
 			return reqCtx, errutil.Error{Code: errutil.BadConfiguration, Msg: fmt.Sprintf("error getting target model name for model %v", modelObj.Name)}
 		}
+		reqCtx.Request.Body["model"] = reqCtx.ResolvedTargetModel // Update target model in the body.


this was moved from PostDispatch to here.
in this point the resolution of the target model is done, it seems to belong here more naturally.

danehans · 2025-05-08T23:49:43Z

@nirrozenbaum needs a rebase.

danehans

/lgtm

ahg-g · 2025-05-11T16:02:23Z

/approve
/lgtm

k8s-ci-robot · 2025-05-11T16:02:35Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: ahg-g, nirrozenbaum

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

The pull request process is described here

Needs approval from an approver in each of these files:

~~OWNERS~~ [ahg-g]

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

Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

ahg-g · 2025-05-11T17:29:47Z

/lgtm

* Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to NewSchedulerContext API change Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Populate the RequestId field of LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates to tests Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added PostResponse plugins to scheduler config Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added scheduler.OnResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added dispatcher.HandleResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Refactored server response header handling to invoke PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added simple test for PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Setup the logger in the SchedulerContext appropriately for reponses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to rebase issues * merge functions in env utils (#819) Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * generalize scheduling cycle state concept (#818) * generalize scheduling cycle state concept Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * typo Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make linter happy Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make prefix state struct internal to package instead of public Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * remove Model field from LLMRequest (#782) * remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Insure that wanted response header messages have all of the response headers in them Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> --------- Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> Co-authored-by: Nir Rozenbaum <nirro@il.ibm.com>

* remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

* Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to NewSchedulerContext API change Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Populate the RequestId field of LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates to tests Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added PostResponse plugins to scheduler config Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added scheduler.OnResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added dispatcher.HandleResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Refactored server response header handling to invoke PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added simple test for PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Setup the logger in the SchedulerContext appropriately for reponses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to rebase issues * merge functions in env utils (kubernetes-sigs#819) Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * generalize scheduling cycle state concept (kubernetes-sigs#818) * generalize scheduling cycle state concept Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * typo Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make linter happy Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make prefix state struct internal to package instead of public Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * remove Model field from LLMRequest (kubernetes-sigs#782) * remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Insure that wanted response header messages have all of the response headers in them Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> --------- Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> Co-authored-by: Nir Rozenbaum <nirro@il.ibm.com>

* remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

* Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to NewSchedulerContext API change Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Populate the RequestId field of LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates to tests Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added PostResponse plugins to scheduler config Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added scheduler.OnResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added dispatcher.HandleResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Refactored server response header handling to invoke PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added simple test for PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Setup the logger in the SchedulerContext appropriately for reponses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to rebase issues * merge functions in env utils (kubernetes-sigs#819) Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * generalize scheduling cycle state concept (kubernetes-sigs#818) * generalize scheduling cycle state concept Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * typo Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make linter happy Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make prefix state struct internal to package instead of public Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * remove Model field from LLMRequest (kubernetes-sigs#782) * remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Insure that wanted response header messages have all of the response headers in them Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> --------- Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> Co-authored-by: Nir Rozenbaum <nirro@il.ibm.com>

* remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

* Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to NewSchedulerContext API change Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Populate the RequestId field of LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates to tests Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added PostResponse plugins to scheduler config Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added scheduler.OnResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added dispatcher.HandleResponse to handle responses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Refactored server response header handling to invoke PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Added simple test for PostResponse plugins Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Setup the logger in the SchedulerContext appropriately for reponses Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Updates due to rebase issues * merge functions in env utils (kubernetes-sigs#819) Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * generalize scheduling cycle state concept (kubernetes-sigs#818) * generalize scheduling cycle state concept Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * typo Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make linter happy Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * make prefix state struct internal to package instead of public Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * remove Model field from LLMRequest (kubernetes-sigs#782) * remove Model field from LLMRequest Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * rebase handling Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> --------- Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> * Added the LLMResponse struct and RequestId to LLMRequest Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> * Insure that wanted response header messages have all of the response headers in them Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> --------- Signed-off-by: Shmuel Kallner <kallner@il.ibm.com> Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com> Co-authored-by: Nir Rozenbaum <nirro@il.ibm.com>

k8s-ci-robot added the cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. label May 5, 2025

k8s-ci-robot requested review from ahg-g and robscott May 5, 2025 06:06

k8s-ci-robot added the size/M Denotes a PR that changes 30-99 lines, ignoring generated files. label May 5, 2025

kfswain reviewed May 5, 2025

View reviewed changes

pkg/epp/handlers/request.go Outdated Show resolved Hide resolved

k8s-ci-robot added do-not-merge/hold Indicates that a PR should not merge because someone has issued a /hold command. needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. labels May 6, 2025

nirrozenbaum force-pushed the rm-model branch from 4313476 to 00111d9 Compare May 8, 2025 07:44

k8s-ci-robot removed the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label May 8, 2025

k8s-ci-robot removed the do-not-merge/hold Indicates that a PR should not merge because someone has issued a /hold command. label May 8, 2025

nirrozenbaum commented May 8, 2025

View reviewed changes

nirrozenbaum requested a review from kfswain May 8, 2025 07:47

k8s-ci-robot added the needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. label May 8, 2025

danehans reviewed May 8, 2025

View reviewed changes

k8s-ci-robot assigned danehans May 8, 2025

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 8, 2025

nirrozenbaum force-pushed the rm-model branch from 00111d9 to a2c19cf Compare May 9, 2025 07:20

k8s-ci-robot removed lgtm "Looks good to me", indicates that a PR is ready to be merged. needs-rebase Indicates a PR cannot be merged because it has merge conflicts with HEAD. labels May 9, 2025

nirrozenbaum requested a review from danehans May 9, 2025 07:21

k8s-ci-robot assigned ahg-g May 11, 2025

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 11, 2025

k8s-ci-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label May 11, 2025

nirrozenbaum added 2 commits May 11, 2025 19:59

remove Model field from LLMRequest

a42cb79

Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

rebase handling

0d00997

Signed-off-by: Nir Rozenbaum <nirro@il.ibm.com>

nirrozenbaum force-pushed the rm-model branch from a2c19cf to 0d00997 Compare May 11, 2025 17:01

k8s-ci-robot removed the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 11, 2025

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label May 11, 2025

k8s-ci-robot merged commit 8df511a into kubernetes-sigs:main May 11, 2025
8 checks passed

nirrozenbaum deleted the rm-model branch May 11, 2025 17:34

remove Model field from LLMRequest #782

remove Model field from LLMRequest #782

Uh oh!

Conversation

nirrozenbaum commented May 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nirrozenbaum commented May 5, 2025

Uh oh!

netlify bot commented May 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

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

Uh oh!

LukeAVanDrie commented May 5, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

nirrozenbaum commented May 5, 2025

Uh oh!

kfswain commented May 5, 2025

Uh oh!

Uh oh!

LukeAVanDrie commented May 5, 2025

Uh oh!

nirrozenbaum commented May 8, 2025

Uh oh!

nirrozenbaum May 8, 2025

Choose a reason for hiding this comment

Uh oh!

danehans commented May 8, 2025

Uh oh!

danehans left a comment

Choose a reason for hiding this comment

Uh oh!

ahg-g commented May 11, 2025

Uh oh!

k8s-ci-robot commented May 11, 2025

Uh oh!

ahg-g commented May 11, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

nirrozenbaum commented May 5, 2025 •

edited

Loading

netlify bot commented May 5, 2025 •

edited

Loading

LukeAVanDrie commented May 5, 2025 •

edited

Loading