Skip to content

feat: dynamically discover external rollout engines - #3

Open
Aphoh wants to merge 9 commits into
mainfrom
warnold/dynamic-external-engine-discovery
Open

feat: dynamically discover external rollout engines#3
Aphoh wants to merge 9 commits into
mainfrom
warnold/dynamic-external-engine-discovery

Conversation

@Aphoh

@Aphoh Aphoh commented Aug 12, 2026

Copy link
Copy Markdown
Owner

Summary

  • add --rollout-external-dynamic-discovery-path for synchronous worker control-URL discovery
  • refresh membership immediately before weight updates and recreate control actors only when topology changes
  • reconnect NCCL weight-update groups using the new membership while disconnecting the prior actors
  • retain replaced control actors until the update finishes, then retire them
  • require a shared rollout frontend so request routing remains owned by the external system

Dependency

Depends on #2 and has been rebased onto its latest head. Both PRs target main as requested, so this PR temporarily includes #2 in its GitHub diff. Once #2 merges, this PR reduces to its single dynamic-discovery commit.

Validation

  • tests/test_external_sglang_engines.py: 11 passed
  • all 56 affected discovery, streaming, argument-validation, and custom-generate contract tests passed together after the rebase
  • uvx prek run --all-files: passed
  • the combined implementation was previously exercised end-to-end with Dynamo-managed workers and NCCL weight updates

@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from 582952f to b662494 Compare August 12, 2026 17:59
@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from b662494 to 0e49024 Compare August 12, 2026 18:44
@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from 0e49024 to 7cd0763 Compare August 12, 2026 18:54
@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from 7cd0763 to 5e51100 Compare August 12, 2026 19:00
@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from 5e51100 to 7118c84 Compare August 12, 2026 19:45
@Aphoh
Aphoh force-pushed the warnold/dynamic-external-engine-discovery branch from 7118c84 to 06d397f Compare August 12, 2026 21:33
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant