fix(vllm): declare entry-stage engine_input_source in GLM-Image NIXL config (cherry-pick #12709) - #12719
Merged
pvijayakrish merged 1 commit intoAug 5, 2026
Conversation
…config (#12709) Signed-off-by: Guan Luo <41310872+GuanLuo@users.noreply.github.com> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Signed-off-by: Guan Luo <41310872+GuanLuo@users.noreply.github.com>
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| # Entry stage: input comes from the request, not from another stage. | ||
| # Declared explicitly because the stage_args format applies no schema | ||
| # defaults, and vLLM-Omni reads this field unconditionally. | ||
| engine_input_source: [] |
Contributor
There was a problem hiding this comment.
🔍 Fix may only clear the first startup failure
The PR description itself notes that DYN-3746 argues this config needs a broader migration (stage_args → stages, nested engine_args:/runtime: forms, and a different connector declaration form) and that end-to-end validation on two H100s has not been performed. If that analysis still holds, the AR stage may fail at a later init step for a different missing key; a follow-up run of disagg_omni_glm_image_nixl.sh on the 1.4.0 image is warranted before assuming the pipeline reaches ready.
Was this helpful? React with 👍 or 👎 to provide feedback.
![dynamo-review-agent[bot]](https://avatars.githubusercontent.com/u/201626793?s=60&v=4){kind=small}
|
🎯 Code Coverage (details) 🔗 Commit SHA: eca489e | Docs | Datadog PR Page | Give us feedback! |
Merged
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
release/1.4.0Missing key engine_input_source, sodisagg_omni_glm_image_nixl.shcould not start and/v1/images/generationsnever became ready.Fixes DYN-3746
Original PR
0c4fabc26b1baea86846eb2e3bd4dc89e4e9dc0cWhat the fix does
Declares
engine_input_source: []on the entry stage. Thestage_argsformat applies no schema defaults and vLLM-Omni reads the field unconditionally, so omegaconf raisedConfigAttributeErroratstage_args[0].engine_input_source, killing the orchestrator thread.Applied to
release/1.4.0with no conflicts — the config file is identical tomainon this branch.Test plan
Verified after the pick: the file parses, and the entry stage now declares the field while stage 1 keeps its existing source.
Byte-identical to the merged state on
main.pre-commit run --files <changed>passes clean.Not verified end-to-end. This needs a two-GPU H100 run of
disagg_omni_glm_image_nixl.shon the 1.4.0 image to confirm the AR stage now reaches ready — the reproduction in DYN-3746.Note
DYN-3746's write-up argues the config needs a fuller migration than this one field (
stage_args→stages, plus theengine_args:/runtime:nesting and the connector declaration form). The ticket was closed against this PR, so that analysis may have been superseded — but if it still holds,release/1.4.0will need the follow-up too, and this pick only clears the first failure.🤖 Generated with Claude Code