fix(ingestion): do not retry score batches on 207 partial failure

[moneebullah25]

What does this PR do?

A 207 ingestion response with per-item errors raises APIErrors, which is not an APIError. The score consumer therefore treated it as a generic failure and re-POSTed the whole batch up to 3 times, and the per-item details only reached DEBUG.

This catches APIErrors in ScoreIngestionConsumer._upload_batch, logs one WARNING with the rejected count and up to 5 per-item messages (escaped, truncated to 200 chars), and does not retry. The "Successfully sent" debug line now only fires on real success.

Known limitation (noted in a code comment): per-item 429/5xx entries inside a 207 are not retried individually, since APIErrors does not carry the item id yet. The separate 4xx batch drop is intentionally not touched.

Fixes #1965

Type of change

• Bug fix
• New feature
• Breaking change
• Refactor
• Documentation update
• Tooling, CI, or repo maintenance

Verification

List the main commands you ran:

.venv/bin/ruff check .
.venv/bin/ruff format --check langfuse tests/unit/test_resource_manager.py
.venv/bin/mypy langfuse --no-error-summary
.venv/bin/pytest tests/unit/test_resource_manager.py -q
.venv/bin/pytest tests/unit/test_otel.py -q
.venv/bin/pytest -n auto --dist worksteal tests/unit

Notes: uv run could not be used locally (installed uv 0.9.8 cannot parse this repo's uv.lock/exclude-newer), so tools were run from a venv with the project installed. ruff check passes; mypy reports no errors in the changed file (existing errors elsewhere come from the missing langchain/openai dev extras locally); test_resource_manager.py 13 passed, test_otel.py 65 passed. The full unit run had failures/errors only in modules needing the uninstalled langchain/openai extras (test_prompt, test_langchain, test_openai, etc.) and test_media.py has a pre-existing format diff; CI should cover those.

Checklist

• I self-reviewed the diff using code_review.md.
• I added or updated tests for behavior changes.
• I updated docs, examples, or .env.template if needed.
• I did not hand-edit generated files; if generated files changed, I used the upstream regeneration path.
• I did not commit secrets or credentials.

Confidence Score: 4/5

The PR appears safe to merge, with a non-blocking gap in tests for the changed logs.

Summary

Stops retrying whole score batches after a 207 response with item errors.

• Logs one warning with the rejected count and up to five escaped, shortened error messages.
• Logs “Successfully sent” only when the request succeeds.
• Adds a regression test for no retry and warning output. The test should also check the changed logging safeguards.

Acknowledged limitations:

• moneebullah25 explicitly acknowledged and deferred retries for per-item 429/5xx failures because APIErrors does not carry item IDs.
• moneebullah25 explicitly stated that dropping non-429 batch-level 4xx failures is intentionally unchanged.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Send batch] --> B{Result}
    B -->|Success| C[Log success]
    B -->|207 with item errors| D[Log bounded warning; do not retry]
    B -->|Non-rate-limit 4xx| E[Do not retry]
    B -->|Other failure| F[Retry within existing limit]
    C --> G[Mark queue items done]
    D --> G
    E --> G
    F --> G

Loading

_{Reviews (1) · Last reviewed commit: bf505ce · Reviewed by Greptile}

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

All committers have signed the CLA.