fix(iris): preserve thread container isolation in LocalCluster

[claude]

Summary

• Create a dedicated ThreadContainer("local-cluster") at each LocalCluster call site to prevent shutdown from stopping unrelated managed threads sharing the ambient default container
• Fixes regression identified in PR 3632 review by @yonromai

Changes

File	Change
cluster/manager.py	Pass ThreadContainer to LocalCluster in connect_cluster()
cli/main.py	Pass ThreadContainer to LocalCluster in require_controller_url()
cli/cluster.py	Pass ThreadContainer to LocalCluster in cluster_start()
client/client.py	Pass ThreadContainer to LocalCluster in IrisClient.local()

Test plan

• All 1274 unit tests pass
• Pre-commit passes

Generated with Claude Code

[yonromai]

🤖 Request changes. The four production entry points from the original review are now isolated, but the LocalCluster constructor itself still defaults to the ambient thread container, so the regression remains for direct callers and future call sites.

I reproduced it on this branch with thread_container_scope("probe"): an unrelated managed thread is alive before cluster.close() and dead after. The fix should move into LocalCluster itself, e.g. create a private ThreadContainer("local-cluster") when threads is None.

Generated with Codex

[yonromai]

🤖 This protects connect_cluster(), but it still leaves LocalCluster(...) unsafe by default. LocalCluster.__init__ still stores threads=None, so direct callers still fall back to the ambient container and cluster.close() still stops unrelated managed threads. I reproduced that on this branch with a thread_container_scope("probe") check.

Recommended fix: make LocalCluster allocate its own ThreadContainer("local-cluster") when threads is None, instead of relying on every caller to remember this wrapper.

Generated with Codex

[yonromai]

@claude do you like 🍎 ?

[claude]

Claude finished @yonromai's task in 12s —— View job

---

I appreciate the question, but I'm best suited for code-related tasks like reviewing PRs, answering technical questions, and implementing changes. 🍎 opinions are outside my area of expertise! 😄

If you'd like me to address the review feedback on this PR (moving the ThreadContainer default into LocalCluster.__init__), just let me know!

[yonromai]

@claude go ahead.

[claude]

Claude finished @yonromai's task in 7m 20s —— View job

---

Implementing review feedback

• Read LocalCluster implementation
• Move ThreadContainer default into LocalCluster.__init__
• Revert per-call-site wrapping and remove unused imports
• Run pre-commit and tests
• Push changes

Moved the ThreadContainer("local-cluster") allocation from each call site into LocalCluster.__init__:

self._threads = threads or ThreadContainer("local-cluster")

This ensures all direct callers and future call sites get thread isolation by default, without needing to remember to wrap. Removed the per-site ThreadContainer creation and cleaned up unused imports in all 4 files.

Testing: Pre-commit passes. All 1324 Iris tests pass (1 skipped).
· Branch