Python: fix: unique temporary filenames for concurrent FileCheckpointStorage saves (#8182)

[Atik (ptimizeroracle)]

description

fixes #8182

every writer used the same temp name, {id}.json.tmp. so when two saves of one checkpoint id run at the same time, one os.replace eats the temp file the other writer is about to replace, and that save dies with a raw FileNotFoundError

now each writer gets its own temp name, {id}.json.{pid}.{uuid}.tmp. still ends in .tmp so the *.json globs in list_checkpoints and list_checkpoint_ids never pick these up

validation

before: 20 concurrent saves of one id, several of them crash in os.replace
after: same probe, 0 errors

added test_file_checkpoint_storage_concurrent_saves_same_id (10 concurrent saves, no exceptions, final state loads)

test_checkpoint.py 47/47 local, ruff clean

i only changed the temp name. no exception type changes, that question stays in the issue thread

ai assistance: i led this contribution and used a coding agent to help write the change and its test. validation was run locally. happy to use a different suffix format if you have one

[Copilot]

🟡 Changes recommended

Failed writes can leave an unbounded number of uniquely named temporary files.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Prevents concurrent checkpoint saves from racing on a shared temporary file.

Changes:

• Generates a unique temporary filename per save.
• Adds a concurrent same-ID save regression test.

File summaries

File	Description
_checkpoint.py	Uses PID/UUID-based temporary filenames.
test_checkpoint.py	Tests concurrent saves and final-state loading.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 1
• Review effort level: Balanced

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[Atik (ptimizeroracle)]

@microsoft-github-policy-service agree

[Atik (ptimizeroracle)]

addressed in 6b7311d: failed writes now unlink their own temp file before re-raising, so failures don't accumulate. added a test that simulates a replace failure and asserts no .tmp leftovers. 48/48 local.

[Atik (ptimizeroracle)]

closing this duplicate in favor of #7757, which already covers the same concurrent-save temp-file race and is under active review

the reproduction in #8182 and the regression tests here remain available as additional evidence

thanks for the ongoing work on the fix