bug: flaky tests due to global.fetch mock pollution in parallel test execution

[la14-1]

Summary

Two tests consistently fail when run as part of the full suite (bun test) but pass individually:

1. hetzner-cov.test.ts: "cleans up orphaned primary IPs on resource_limit_exceeded and retries"
2. digitalocean-token.test.ts: "attempts OAuth recovery on 401 before throwing" (Expected callCount=2, Received=4)

Root Cause

Bun runs test files in the same process. global.fetch mock assignments in one test file can race with or pollute another file's fetch mock when tests run concurrently. The hetzner test throws the raw API error instead of entering the retry path, and the DO test sees double the expected fetch calls.

Both tests pass in isolation and in small groups. CI may not reproduce this if the test runner scheduling differs.

Suggested Fix

• Use mock.module() to isolate fetch per-test-file, OR
• Add a preload that saves/restores global.fetch around each test file, OR
• Run these specific test files serially (bun test serialization)

[la14-1]

Acknowledged — this is a valid test isolation bug. Parallel test files sharing global.fetch in the same Bun process can cause mock pollution and non-deterministic failures.

This likely needs per-file fetch scoping (e.g., mock.module or test-level fetch wrappers) rather than raw global.fetch = assignments.

Triaging for the next refactor cycle.

-- refactor/community-coordinator

[la14-1]

Multiple PRs are addressing this:

• fix(test): resolve flaky fetch mock tests caused by concurrency #3394 (resolve flaky fetch mock tests caused by concurrency)
• fix(tests): use URL-based mock routing to fix flaky tests #3379 (URL-based mock routing)
• fix(tests): filter telemetry fetch calls in DO/Hetzner mocks #3391 (filter telemetry fetch calls in DO/Hetzner mocks)
• fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376 (prevent telemetry singleton from polluting parallel test fetch mocks)

-- refactor/community-coordinator