[AdarshK15]

Summary

Blueprint & Daily Tests

• Unified TPU Slurm Blueprint (examples/machine-learning/tpu-v5e-v5p-v6e/):
  • Added tpu-slurm-blueprint.yaml, tpu-slurm-deployment.yaml, and README.md supporting:
    TPU v6e (ct6e-standard-4t),
    TPU v5p (ct5p-hightpu-4t), and
    TPU v5e (ct5lp-hightpu-4t) with Slurm 26.05 (aci-tpu-v5e-v5p-v6e-u2204-slurm-2605-amd64) and Managed Lustre.
  • Changing tpu_machine_type automatically selects the required boot disk type (hyperdisk-balanced for v6e, pd-balanced for v5p/v5e) and default static accelerator_topology (2x2x1 for v5p, 2x2 for v6e/v5e) when tpu_accelerator_topology is left null.
• Integration Tests (tools/cloud-build/daily-tests/):
  • Added 3 daily Spot Slurm tests (ml-tpu-v6e-onspot-slurm, ml-tpu-v5p-onspot-slurm, ml-tpu-v5e-onspot-slurm).
  • Updated test-tpu-slurm.yml to check ConstrainDevices=no for dual-core-per-chip TPUs (tpu7x, v5p) and ConstrainDevices=yes for single-core-per-chip TPUs (v6e, v5e).

---

Key Changes & Bug Fixes

• conf_v2605.py:
  • Added enable_async_reply to SlurmctldParameters when TPU nodesets are present.
  • Removed Type=<tpu_type> from cloud_gres.conf entries (Name=tpu File=/dev/vfio/devices/vfio[0-3]) to match Slurm 26.05's gres/tpu plugin expectations.
• mig_flex.py & resume.py:
  • Added resume_tpu_chunks() to provision multiple TPU slices concurrently via util.map_with_futures instead of sequentially, isolating per-slice failures.
  • Added thread synchronization (_tpu_shared_resource_lock) and a short wait on HTTP 409 ("already exists") in get_instance_template_copy() and _ensure_tpu_workload_policy() to prevent race conditions during concurrent TPU slice resume.
  • Fixed zonal mig_self_link construction during createInstances rollback so _delete_tpu_mig() uses the zonal API when cleaning up failed zonal TPU MIGs.
• slurmsync.py:
  • Updated get_node_action() for terminated TPU instances to transition directly to NodeActionPowerDownForce with a clear reason ("Spot instance preempted" or "Instance terminated") so allocated/mixed nodes power down and clean up their MIGs immediately instead of taking multiple sync cycles.
• util.py:
  • Fixed Lookup.get_jobs() to split scontrol show jobs output on \n\n+(?=JobId=\d+) instead of \n\n (and ReqNodeList=(\S+)), preventing crashes when job Command= fields contain blank lines.
• find_available_zone.sh:
  • Fixed jq quota parsing in check_filestore_quota() and check_lustre_quota() (| vs or operator precedence in the .dimensions fallback, and defaulting missing .details.value fields to "0" before converting to number).

[gemini-code-assist]

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces support for TPU v5e, v5p, and v6e within the Slurm on GCP framework. It provides new blueprints for these TPU generations, enhances the reliability of concurrent TPU slice management, and includes critical bug fixes for job parsing and quota management. These changes ensure robust deployment and operational stability for high-performance TPU workloads on Slurm.

Highlights

• New TPU Support: Added comprehensive Slurm support for TPU v5e, v5p, and v6e, including a new blueprint and deployment configuration.
• Concurrency Improvements: Updated TPU slice resumption to use concurrent processing with thread synchronization to prevent race conditions.
• Slurm Configuration Updates: Updated Slurm 26.05 configuration generation to include 'enable_async_reply' and adjusted GRES configuration to align with the new plugin expectations.
• Reliability and Bug Fixes: Improved error handling for TPU instance termination, fixed job lookup parsing issues, and corrected quota parsing logic in deployment scripts.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize the Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counterproductive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[AdarshK15]

/gemini review

[AdarshK15]

/gemini review

[arpit974]

@AdarshK15,
pre commits are failing,
please check