Skip to content

fix(ai_dynamo): accept CloudAI bool True for multi-worker flags#970

Merged
podkidyshev merged 1 commit into
NVIDIA:mainfrom
dshirron3:fix/ai-dynamo-bool-case
Jul 16, 2026
Merged

fix(ai_dynamo): accept CloudAI bool True for multi-worker flags#970
podkidyshev merged 1 commit into
NVIDIA:mainfrom
dshirron3:fix/ai-dynamo-bool-case

Conversation

@dshirron3

Copy link
Copy Markdown
Contributor

Summary

  • Fix case-sensitive boolean checks for `multiple-workers-per-node` on prefill and decode workers in `ai_dynamo.sh`.
  • CloudAI passes Python `True`/`False` (capitalized) via Slurm CLI args; the script only accepted lowercase `true`, so multi-worker topology was never applied (`gpus-per-worker` stayed at 4, `workers-per-node` stayed at 1).
  • Use bash `,,` lowercase expansion for case-insensitive comparison; align `dcgm-exporter-enabled` with the same pattern.

Test Plan

  • [V] Deploy ai_dynamo with `multiple-workers-per-node = true` on prefill (disagg tp2 topology).
  • [V] Confirm logs show `PREFILL: workers per node: 2` and `Launching 2 prefill worker(s)` per node (not 1).
  • [V] Confirm `Initialized: prefill 26/26` (13 nodes × 2 workers) for MLPerf disagg 405B config.

Additional Notes

Observed on GB200 CloudAI runs: CLI had `--prefill-multiple-workers-per-node True` but script logged `workers per node: 1`."

Signed-off-by: Dan Shirron <dshirron@nvidia.com>
@coderabbitai

coderabbitai Bot commented Jul 16, 2026

Copy link
Copy Markdown
Contributor

Review Change Stack

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Enterprise

Run ID: 31174fc9-f135-4315-a1eb-225f75c53323

📥 Commits

Reviewing files that changed from the base of the PR and between 0de6ca8 and c6809cd.

📒 Files selected for processing (1)
  • src/cloudai/workloads/ai_dynamo/ai_dynamo.sh

📝 Walkthrough

Walkthrough

Changes

Configuration normalization

Layer / File(s) Summary
Normalize boolean configuration checks
src/cloudai/workloads/ai_dynamo/ai_dynamo.sh
vLLM worker allocation and DCGM exporter metric URL checks now compare normalized, case-insensitive boolean configuration values, with safe handling for missing values in worker allocation.

Estimated code review effort: 1 (Trivial) | ~3 minutes

Suggested reviewers: podkidyshev

🚥 Pre-merge checks | ✅ 4
✅ Passed checks (4 passed)
Check name Status Explanation
Title check ✅ Passed The title clearly summarizes the main change: accepting capitalized CloudAI booleans for multi-worker flags.
Description check ✅ Passed The description matches the code changes and explains the boolean handling fix and test context.
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests

Comment @coderabbitai help to get the list of available commands.

@podkidyshev
podkidyshev merged commit 35a1d1a into NVIDIA:main Jul 16, 2026
5 checks passed
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.

2 participants