Skip to content

fix(gms): align failover tests and runtime contracts - #15049

Draft
hutm wants to merge 3 commits into
hutm/gms-hardening-13-sglang-reservationsfrom
hutm/gms-hardening-14-sglang-recovery
Draft

hutm wants to merge 3 commits into
hutm/gms-hardening-13-sglang-reservationsfrom
hutm/gms-hardening-14-sglang-recovery

Conversation

@hutm

@hutm hutm commented Sep 18, 2026

Copy link
Copy Markdown
Contributor

Purpose

  • Align SGLang tests with native pause routes.
  • Resolve the default CUDA ordinal without recursive lookup.
  • Match the current vLLM persistent-KV context contract.

Why this is separate

This is hardening step 14 of 15. It keeps one reviewable failure or performance boundary together and builds on #15048. The next step is #15050.

Migration note

This existing draft PR was updated in place on 2026-09-22 so its discussion and review history remain available. The stack was rebased onto current Dynamo main and its content was remapped from the simplified local implementation. Previous public heads are preserved in local archive refs.

Validation

The complete remapped stack preserves the reviewed local simplified implementation and adds only repository-required isort/Black formatting; every PR remains non-empty. Per-PR CI and focused correctness checks are being rerun after the restack. Fresh latency and throughput benchmarking is intentionally deferred.

Tracks #15035.

@copy-pr-bot

copy-pr-bot Bot commented Sep 18, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@github-actions github-actions Bot added the fix label Sep 18, 2026
@hutm
hutm force-pushed the hutm/gms-hardening-13-sglang-reservations branch from 0d47c1b to f132644 Compare September 22, 2026 16:37
@hutm
hutm force-pushed the hutm/gms-hardening-14-sglang-recovery branch from 70caabe to a8c5644 Compare September 22, 2026 16:37
@github-actions github-actions Bot added the backend::sglang Relates to the sglang backend label Sep 22, 2026
@hutm hutm changed the title fix(gms/sglang): preserve cache ownership through recovery fix(gms): align failover tests and runtime contracts Sep 22, 2026
@hutm
hutm force-pushed the hutm/gms-hardening-14-sglang-recovery branch from a8c5644 to 99bc03f Compare September 22, 2026 16:49
@hutm
hutm force-pushed the hutm/gms-hardening-13-sglang-reservations branch from bd57d63 to b27f3e5 Compare September 22, 2026 17:09
@hutm
hutm force-pushed the hutm/gms-hardening-14-sglang-recovery branch from 99bc03f to aec6159 Compare September 22, 2026 17:09
Signed-off-by: mkhadkevich <mkhadkevich@nvidia.com>
Signed-off-by: mkhadkevich <mkhadkevich@nvidia.com>
Signed-off-by: mkhadkevich <mkhadkevich@nvidia.com>
@hutm
hutm force-pushed the hutm/gms-hardening-13-sglang-reservations branch from b27f3e5 to f22d537 Compare September 22, 2026 17:19
@hutm
hutm force-pushed the hutm/gms-hardening-14-sglang-recovery branch from aec6159 to e29097c Compare September 22, 2026 17:19

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

backend::sglang Relates to the sglang backend fix size/M

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant