Skip to content

[Bug] Route the category-remote-backend E2E through a real gateway route - #3544

Open
theohsiung wants to merge 4 commits into
vllm-project:mainfrom
theohsiung:fix/3543-category-e2e-route
Open

[Bug] Route the category-remote-backend E2E through a real gateway route#3544
theohsiung wants to merge 4 commits into
vllm-project:mainfrom
theohsiung:fix/3543-category-e2e-route

Conversation

@theohsiung

Copy link
Copy Markdown
Collaborator

Summary

category-remote-backend has never passed — its job failed on the PR that
introduced it (#3138) and has kept failing, so it reported coverage of the
remote category backend that it did not provide.

The router was never at fault. It starts cleanly and logs a correctly selected
backend; the request never reaches it. The profile named its model
base-model, while the shared AIGatewayRoute matches x-ai-eg-model exactly
against math-, science-, social-, humanities-, law- and
general-expert. Anything outside that list has no route, so the gateway
answers 404 — and because the router looks healthy, it reads like a router bug.

One line of substance: the model becomes general-expert, which
e2e/profiles/ai-gateway/values.yaml already uses for the same reason. The
rest is a comment so the next profile on this stack does not repeat it.

Validation

Run on a kind cluster, before and after:

before:  ❌ FAILED - category-backend-routing (2.264s)  Unexpected status code: 404
after:   ✅ PASSED - category-backend-routing (2.245s)

Closes #3543

The profile has never passed. Its job failed on the PR that introduced it
and has kept failing, so it reported coverage of the remote category
backend that it did not provide.

The router was never at fault: it starts cleanly and logs a correctly
selected backend, and the request simply never reaches it. The profile
named its model base-model, while the shared AIGatewayRoute matches
x-ai-eg-model exactly against math-, science-, social-, humanities-, law-
and general-expert. A model outside that list has no route, so the
gateway answers 404 - and because the router looks healthy, the failure
reads like a router bug.

Verified on a kind cluster: the profile now passes, where the same
profile 404s without this change.

Signed-off-by: theohsiung <theobear870924@gmail.com>
@netlify

netlify Bot commented Sep 7, 2026

Copy link
Copy Markdown

Deploy Preview for vllm-semantic-router ready!

Name Link
🔨 Latest commit cb1593e
🔍 Latest deploy log https://app.netlify.com/projects/vllm-semantic-router/deploys/6a9ed27b9b9c6f0008c7ac21
😎 Deploy Preview https://deploy-preview-3544--vllm-semantic-router.netlify.app
📱 Preview on mobile
Toggle QR Code...

QR Code

Use your smartphone camera to open QR code link.
🤖 Make changes Run an agent on this branch

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

@github-actions github-actions Bot added the pr/blocked Blocked on a named decision, dependency, or required check. label Sep 7, 2026
@theohsiung theohsiung added the wg/router-models-inference-runtime Owned by the Router Models and Inference Runtime Workgroup. label Sep 7, 2026
The profile had two independent faults, and a roomier machine only shows
the second.

At 2Gi the router is OOMKilled on a CI runner, restarts, never becomes
ready, and the helm wait burns its full 60m timeout - which is what the
1h23m job actually was, not a slow test. It still loads the local mmbert
embedding model the response cache is configured with, and taking the
classifier off-box does not avoid that: the embedding models are gated
separately from the classifier backend.

A local run survives 2Gi and fails later at the gateway with a 404
instead, which is how one broken profile looked like two different bugs.

Signed-off-by: theohsiung <theobear870924@gmail.com>
…-route

Signed-off-by: theohsiung <theobear870924@gmail.com>

# Conflicts:
#	e2e/profiles/category-remote-backend/values.yaml
@github-actions github-actions Bot added pr/needs-rebase Needs rebase or conflict resolution. wg/router-models-inference-runtime Owned by the Router Models and Inference Runtime Workgroup. and removed wg/router-models-inference-runtime Owned by the Router Models and Inference Runtime Workgroup. pr/blocked Blocked on a named decision, dependency, or required check. labels Sep 7, 2026

@WUKUNTAI-0211 WUKUNTAI-0211 left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM.

@mergify

mergify Bot commented Sep 8, 2026

Copy link
Copy Markdown
Contributor

Queued — the merge queue status continues in this comment ↓.

@wilsonwu wilsonwu left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@mergify

mergify Bot commented Sep 8, 2026

Copy link
Copy Markdown
Contributor

Merge Queue Status

  • Entered queue2026-09-08 02:24 UTC · Rule: default · triggered by @wilsonwu with the merge queue checkbox
  • 🚫 Left the queue2026-09-08 02:25 UTC · at cb1593e661f39dceffe285c1cffff7b20c29f9a5

This pull request spent 7 seconds in the queue, with no time running CI.

Reason

The pull request can't be updated

For security reasons, Mergify can't update this pull request. Try updating locally.
GitHub response: refusing to allow a GitHub App to create or update workflow .github/workflows/integration-test-k8s.yml without workflows permission

Hint

You should update or rebase your pull request manually. If you do, this pull request will automatically be requeued once the queue conditions match again.
If you think this was a flaky issue, you can requeue the pull request, without updating it, by posting a @mergifyio queue comment.

Tick the box to put this pull request back in the merge queue (same as @mergifyio queue).

  • Requeue this pull request

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

pr/needs-rebase Needs rebase or conflict resolution. wg/router-models-inference-runtime Owned by the Router Models and Inference Runtime Workgroup.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Bug] The category-remote-backend E2E profile 404s and has never passed

3 participants