chore(tests): finish t.Parallel() adoption sweep across remaining packages

[cristim]

Summary

t.Parallel() adoption is partial — only pkg/exchange/{auto,exchange,reshape}_test.go, providers/aws/services/ec2/client_test.go, and internal/api/validation_test.go have been audited and parallelized. Other packages keep sequential execution because each needs a per-test-file shared-state audit before it's race-safe to parallelize.

Captured in: known-issues.md → "Outstanding follow-ups" → "t.Parallel() adoption for remaining packages".

Current state

Adoption complete:

• pkg/exchange/{auto,exchange,reshape}_test.go
• providers/aws/services/ec2/client_test.go
• internal/api/validation_test.go

Pending audit + parallelization (each carries shared-state risk):

• internal/api/ other test files — handler fixtures + shared mocks; not race-safe without review.
• internal/config/*_test.go integration tests — share a Postgres container; cannot naively parallelize.
• internal/server/app_test.go — uses package-level vars (runMigrations, migrationsTimeout) that are not race-safe.
• Any test file using os.Setenv / t.Setenv for process-wide state — needs verification that the variable scope is per-test (Go ≥1.17 t.Setenv is correct).

Why this matters

go test ./... for the project currently runs each package sequentially. Per-package parallelism would meaningfully reduce CI wall-clock without materially changing test design. Expected speedup is incremental (a few minutes off the suite) but compounds across CI runs.

Steps to verify the gap

grep -rL "t.Parallel" $(find . -name "*_test.go" | xargs grep -l "func Test" | grep -v vendor) | head
# Lists test files that currently never call t.Parallel

Proposed approach

Per-package incremental audit commits, NOT a single sweeping change:

1. Pick one package at a time (start with internal/api/ test files that don't touch the shared Postgres container).
2. Audit each test for: os.Setenv (use t.Setenv instead), package-level fixtures, shared mocks (move to per-test setup), shared file paths.
3. Add t.Parallel() to test functions and (where needed) sub-tests.
4. Run go test -race -count=2 ./<pkg>/... to confirm no flakes.
5. Commit per package with chore(tests): parallelize <pkg>.

For integration tests sharing a Postgres container: use sub-test t.Run with t.Parallel() only after introducing per-test-schema isolation, OR keep them sequential and parallelize only unit-level tests within the same package.

References

• Known-issue doc: known-issues.md → "t.Parallel() adoption for remaining packages".
• Already-done examples: pkg/exchange/auto_test.go, internal/api/validation_test.go.

Severity

Low — quality-of-life CI improvement, no functional impact.

Effort

Variable per package — small for unit-only packages (~30 min audit + parallelize), larger for integration-test packages (require schema-isolation or stay sequential). Best handled as a series of small per-package commits.

[cristim]

Progress update — PR #70 landed (one package)

PR #70 → sha 2e33c88 adopted t.Parallel() across all three test files in pkg/common/ — the safest leaf-package starting point per the issue's "one package per PR" guidance.

Deploy state:

• Azure Container Apps + GCP Cloud Run deployed 2e33c88 at 2026-04-25T13:33:16Z (both green).
• AWS Lambda did not trigger a new deploy for 2e33c88 — likely because the change is Go-tests-only and the deploy workflow's path filters skip it. The previous AWS deploy at sha 599e58f is still live; the parallel-tests change has no runtime impact, so this is fine.

Verification path: test-only change, no runtime behaviour to curl. CI ran the parallelized tests successfully on the merge.

Issue stays open — many packages remain. Suggested next leaves to sweep (mirroring the same "no shared mutable state" criterion):

• pkg/errors/ — already has a draft PR chore(tests): parallelize pkg/errors test files #83 in flight
• internal/credentials/ test files
• internal/commitmentopts/ (mostly clean post-PR chore(commitmentopts): MemoryDB probe target verification + page-cap test coverage #69)
• internal/api/ (largest; needs care because of shared HTTP test fixtures)

Closing this issue is appropriate once the sweep is complete; for now it's tracking the remaining packages.

[cristim]

t.Parallel adoption progress (PR #83 merged at d03b5e3a6fc54529cfcee975cd263cc7fb2affe0)

This PR parallelized pkg/errors/ test files (49 t.Parallel calls — 9 top-level + 40 sub-tests).

Cumulative progress on issue #58:

• ✅ pkg/exchange/{auto,exchange,reshape}_test.go (pre-existing)
• ✅ providers/aws/services/ec2/client_test.go (pre-existing)
• ✅ internal/api/validation_test.go (pre-existing)
• ✅ pkg/common/ — PR chore(tests): parallelize pkg/common test files #70
• ✅ pkg/errors/ — PR chore(tests): parallelize pkg/errors test files #83 (this one)

Remaining packages (per the issue body's per-package list, each safe to ship as a separate small PR):

• pkg/{logging,scorer,retry,provider}/ and other leaf packages
• internal/api/ other test files (NOT validation_test.go) — handler fixtures + shared mocks; needs review
• internal/config/*_test.go — Postgres integration; cannot naively parallelize, needs schema isolation
• internal/server/app_test.go — package-level vars; needs review
• Any test file using os.Setenv — replace with t.Setenv per-test

Issue stays OPEN for the next incremental PR.