Skip to content
Get packs

TPU Partition Overlap Troubleshooter

Diagnose GKE TPU slice failures caused by partition or node overlap with another slice

Source Download skill
Repository
google/skills
Stars
21,782
Licence
Apache-2.0
Files
2 files

What it does

When a Google Kubernetes Engine (GKE) TPU slice fails to form with the reason SliceCreationFailed because its nodes or partitions are already used by another slice or JobSet, this skill guides the agent through a structured, read-only investigation. The agent confirms the failure, rules out other causes, finds the conflicting slice, and explains which documentation section describes the fix, leaving the changes to the user. It is written for people who already operate GKE clusters with TPUs.

Great for

  • Engineers running TPU workloads on GKE with dynamic slicing who see SliceCreationFailed errors
  • Platform or ML infrastructure teams triaging slice conflicts across shared TPU node pools
  • Operators who want a consistent diagnostic checklist with links to official docs

Skip it if

  • Quota exhaustion, unresponsive TPU nodes, or general slice provisioning issues without node overlap, which the skill itself says to handle elsewhere (reason: out of scope and could lead to wrong fixes)
  • Anyone without GKE, gcloud, or kubectl access, since the skill cannot troubleshoot without live cluster data (reason: it depends on running cluster queries)

Try asking

  • “My Slice custom resource shows SliceCreationFailed. Help me find which other slice is using the same nodes.”
  • “Diagnose why slice my-slice-1 in cluster prod-tpu in us-central1 failed to form.”
  • “Check whether the partitions requested by my slice overlap with a running JobSet and tell me the doc section for the fix.”
Step-by-step guide: how to use TPU Partition Overlap TroubleshooterInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 2 original files, the Apache-2.0 licence and an attribution note.

    gke-ai-troubleshooting-tpu-partition-overlap.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip gke-ai-troubleshooting-tpu-partition-overlap.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip gke-ai-troubleshooting-tpu-partition-overlap.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Provide the cluster name, location, and failed slice name up front so the agent does not need to guess.
  • Use read-only credentials; the skill itself forbids draining, deleting, or re-creating nodes, but scoped permissions add a safety layer.
  • Ask the agent to cite the documentation section for any recommended fix and apply changes yourself after reviewing them.

What's inside

  • references/attribution.md
  • SKILL.md

Redistributed unmodified from google/skills under the Apache-2.0 licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.