test(integration): opt the config pass-through spend-log case into auth (rc/1.104.0 backport of #44265)

[yuneng-berri]

TLDR

Problem this solves:

• Scheduled CircleCI integration-extensions has been red since 2026-10-01 06:09 UTC
• The config pass-through spend-log test relied on a config wins side effect

How it solves it:

• The test's config route now sets auth: true explicitly
• No product code changes, so pass-through auth stays as before config wins

User Flow

Before: a config pass-through route without auth is treated the way it was before config wins, and the test expects otherwise

1. The admin declares /audit-pt under general_settings.pass_through_endpoints with no auth key
2. A client sends POST http://localhost:4000/audit-pt with a valid key and the upstream answers 403
3. The client gets the upstream 403 body back, and no failure row shows up in http://localhost:4000/ui/?page=logs

After: the same route with auth: true gets the failure row the test was written for

1. The admin declares /audit-pt with auth: true
2. The client sends the same POST http://localhost:4000/audit-pt and gets the upstream 403 body back
3. http://localhost:4000/ui/?page=logs shows the failed request with the upstream error message

Relevant issues

Follow-up to #43962 and #42695. Backport of #44265 to rc/1.104.0

Pre-Submission checklist

• I have added meaningful tests
• The handful of test files covering my change pass locally
• My PR passes all required CI/CD checks
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review

Screenshots / Proof of Fix

The one test, run twice on each side against a local integration stack (Postgres 16, Redis 7, the test's own 2-worker proxy): pytest --timeout=90 -p no:randomly tests/integration/observability/test_passthrough_upstream_error_visibility.py::test_config_pass_through_route_logs_body_and_strips_query

Before (485ad76)

1. Run 1: Failed: Timeout (>90.0s) from pytest-timeout while polling for the failure row in _spend_error_information
2. Run 2: same timeout at the same frame (92.4s)

After (1d9cd9b)

1. Run 1: passed in 40.2s
2. Run 2: passed in 44.2s
3. LiteLLM_SpendLogs held exactly 2 failure rows afterwards, one per After run and none from the Before runs

Type

✅ Test

Caveats (if any)

Low

• Failures on config pass-through routes without auth still write no spend-log row, which matches behavior before config wins

[greptile-apps]

Confidence Score: 5/5

[Low risk] Test file adds a configuration option to an integration test.

This test-only change appears safe to merge

What we checked:

• Auth could block the test: The test sends the integration master key. The auth code gives that caller the proxy administrator role, which bypasses the pass-through route grant check

Summary

Adds auth: True to the config pass-through route in one integration test. No product code changes

• Keeps the upstream 403 response, error-body logging, query-stripping, and failed spend-row assertions unchanged
• No actionable issues found
• yuneng-berri explicitly acknowledged that config routes without auth still omit failed spend rows. The PR intentionally preserves that behavior

_{Reviews (1) · Last reviewed commit: "test(integration): opt the config pass-t..."}