[anth-volk]

Fixes #4085
Fixes #4071

#4089 is stacked on this PR. It changes which record represents a failed simulation worker, which moves some failures between groups, and relies on this PR's algorithm 2 instead of needing algorithm 3, so it should merge right after this one.

What was wrong

The Public API's daily telemetry error report fingerprints an exception by its service family, its exception class and its three innermost stack frames in application packages. Each frame was keyed as module path:function name, so moving a function to another module gave every exception raised through it a new fingerprint: the group's issue stopped being updated and a duplicate opened. The recent file-length splits moved many functions into new modules, and made some private helpers public.

What changed

• normalization.fingerprint_frames still chooses the three innermost frames whose paths are in application packages (or, without any, the three innermost of any kind), but now keys each one by its function name alone, with leading underscores removed. _resolve_region and resolve_region give the same key. Any other rename still changes the fingerprint, which is accepted. Two functions with the same name in different modules now give the same key; with three frames, the exception class and the service family also in the fingerprint, this should rarely merge different failures.
• frame_key, which returned path:function, is replaced by frame_location, which returns the normalized path and the function name separately, and function_key, which removes the leading underscores.
• FINGERPRINT_ALGORITHM is now 2. The issue state comment and the issue details record it.
• gcp/observability/README.md describes the new frame keys and what happens to issues from algorithm 1. changelog.d/telemetry-fingerprint-function-names.changed.md describes the change.

Message and failed-request groups

Their fingerprint inputs are unchanged. Their fingerprints still change, because the HMAC input starts with FINGERPRINT_ALGORITHM, the issue state records one algorithm number for every kind, and the sync matches an issue's state only when it names the current algorithm. Keeping them would need a separate algorithm number per kind, which isn't worth it for the two such issues that exist (#4067 and #4070).

Issues opened under algorithm 1

Five exist: #4067 (message group), #4068 (closed as completed), #4069, #4070 (failed-request group) and #4071. sync._issue_fingerprint reads an issue's fingerprint from its state comment only when the state names the current algorithm; otherwise it reads the fingerprint at the end of the title. The job never changes titles, so these issues are indexed under their algorithm 1 fingerprints, which no algorithm 2 group produces. As a result:

• Matching: none of them matches a group. Nothing else iterates over unmatched issues, so they are left as they are.
• Updating: their marked sections keep the last algorithm 1 report and are no longer refreshed.
• Reopening: [telemetry] policyengine-api: StaleDataError on /<country_id>/report (20e14b2f6c33dcfa) #4068 won't reopen if its StaleDataError returns. A new issue opens instead, without the regression label.
• New issues: a group that still fails opens a new issue once it meets the thresholds, within the limit of five new issues per run.
• Nothing fails: the old state comments still parse, and an issue whose fingerprint matches nothing is skipped.

Someone can close the five by hand. Alternatively, an issue can be carried over by replacing the fingerprint at the end of its title with its group's algorithm 2 fingerprint: the job then matches it by its title and rewrites its marked section with an algorithm 2 state. A test covers this.

Testing

The first commit adds the tests and they fail; the second commit is the fix.

• test_records.py: an exception raised through functions that moved to other modules gives the same fingerprint, as does a helper renamed from _resolve_region to resolve_region, in its old or new module. A renamed function (_find_region) still gives a different fingerprint; this passes before and after. The exception's fingerprint inputs are now the function names.
• test_normalization.py: the three existing frame-selection tests expect function names, and a new test checks that leading underscores are removed (__getitem__ becomes getitem__). Frames are still chosen by path: library frames are skipped when there are application frames.
• test_rendering.py: the state comment and the issue details name algorithm 2.
• test_sync.py: an open, or completed and closed, issue from algorithm 1 for the same failures is neither updated, reopened nor commented on, and a new issue opens. An algorithm 1 issue whose title was changed to the new fingerprint is updated and gets an algorithm 2 state.
• The fix commit also renames frame_key to frame_location in the six frame-extraction tests, which now compare (path, function) pairs. Their expected paths and functions are unchanged.
• pytest tests/unit/telemetry_error_issues: 254 passed on master; 12 failed and 250 passed after the first commit; 262 passed after the second.
• ruff format and lint, the quality guards, the rules guard against master, the rules tool tests, the generated-files check and uv lock --check pass. No inline suppressions were added (94 noqa, 9 type or pyright ignores, as on master).

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.83%. Comparing base (ad26e90) to head (fe8ad13).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #4086       +/-   ##
===========================================
+ Coverage   50.82%   87.83%   +37.01%     
===========================================
  Files         254      270       +16     
  Lines       14635    16058     +1423     
  Branches     2339     2626      +287     
===========================================
+ Hits         7438    14105     +6667     
+ Misses       6570     1284     -5286     
- Partials      627      669       +42     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.