Fix: surface swallowed hash-computation errors in calculate_pkg_hashes

[kevinemoore]

Problem

calculate_pkg_hashes submits per-entry hash computations to a thread pool, then calls concurrent.futures.wait(comp_futures). wait() blocks until every future is done but never reads their results, so any exception raised inside a computation is silently retained on the future and discarded. The observable damage:

• An AccessDenied writing to a cross-region scratch bucket (or any per-entry failure) is swallowed.
• entry.hash is left unset, and package creation falls back to an expensive download-and-hash — with no signal that anything went wrong.

This is item 4.5 of the alias-roles 26.8.0 remediation (the swallowed-error half of the cross-region scratch finding). It is filed and reviewed as a standalone PR, separate from the cross-region scratch grant fix (item 4), which is an IAM/template change in the deployment repo.

Fix

Replace wait() with an as_completed() + f.result() loop, re-raising the first exception so the failure is loud. This mirrors the pattern already used for precomp_futures a few lines above.

Test

Added test_calculate_pkg_hashes_propagates_compute_error: mocks compute_checksum_via_copy to raise a dedicated _ScratchAccessDenied, asserts the exception propagates, and asserts the failed entry's hash stays None (the actual regression symptom). Verified it fails (DID NOT RAISE) against the pre-fix wait() code before the fix was applied.

A /code-review pass ran on the branch's final head; the two test-quality findings it raised (bare-Exception assertion; missing symptom assertion) are folded into the test above.

Test environment note

The pkgpush lambda suite runs standalone (no Postgres/Docker needed). Run locally with uv run pytest under lambdas/pkgpush/ — 58 passed including the new regression test. This differs from the registry suite, which needs Postgres + Docker.

Follow-up

• The CHANGELOG entry currently carries a #PR placeholder link; it will be backfilled with this PR's number.

Greptile Summary

This PR makes package hash computation fail loudly when a worker future raises, rather than silently proceeding with an unset entry hash.

• Replaces passive waiting with as_completed() and result() so worker exceptions propagate.
• Adds a regression test covering the scratch-copy failure path and its unset-hash symptom.
• Documents the behavior change in the pkgpush changelog.

Confidence Score: 5/5

The PR appears safe to merge because it surfaces worker failures as intended without introducing a concrete correctness, security, or lifecycle defect.

The executor context still waits for all submitted work during unwinding, while calling each completed future's result closes the existing exception-swallowing path and the regression test covers the intended failure behavior.

Important Files Changed

Filename	Overview
lambdas/pkgpush/src/t4_lambda_pkgpush/init.py	Replaces exception-swallowing future waiting with explicit result retrieval while preserving executor shutdown behavior.
lambdas/pkgpush/tests/test_hash_calc.py	Adds focused regression coverage proving copy-path exceptions propagate and do not set the failed entry's hash.
lambdas/pkgpush/CHANGELOG.md	Records the corrected package-hashing error behavior, with the acknowledged PR-link placeholder pending replacement.

Sequence Diagram

sequenceDiagram
    participant Push as Package push
    participant Pool as Hash worker pool
    participant Worker as Entry hash computation
    Push->>Pool: Submit entry computations
    Pool->>Worker: Compute checksum
    alt computation succeeds
        Worker-->>Pool: Checksum
        Pool-->>Push: Future completes
        Push->>Push: result() returns
    else computation fails
        Worker-->>Pool: Exception
        Pool-->>Push: Future completes
        Push->>Push: result() re-raises exception
    end

Loading

_{Reviews (1): Last reviewed commit: "Tighten swallowed-hash-error regression ..." | Re-trigger Greptile}

Context used:

• Knowledge Base — Operational event services

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 39.67%. Comparing base (1609144) to head (161e7bb).

❗ There is a different number of reports uploaded between BASE (1609144) and HEAD (161e7bb). Click for more details.

HEAD has 11 uploads less than BASE

Flag	BASE (1609144)	HEAD (161e7bb)
py-shared	1	0
api-python	10	0

Additional details and impacted files

@@             Coverage Diff             @@
##           master    #5249       +/-   ##
===========================================
- Coverage   54.70%   39.67%   -15.03%     
===========================================
  Files         869      757      -112     
  Lines       36337    26283    -10054     
  Branches     6424     6424               
===========================================
- Hits        19877    10428     -9449     
+ Misses      14716    14111      -605     
  Partials     1744     1744               

Flag	Coverage Δ
api-python	?
catalog	34.87% <ø> (ø)
lambda	85.38% <100.00%> (+<0.01%)	⬆️
py-shared	?

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.