Barrier Concurrency Testing
Deterministic race condition testing -- no flaky timing-based tests.
Instead of setTimeout (flaky) or sleep (slow), use barriers to pause execution at exact interleave points. The test controls when each concurrent operation proceeds, making race condition tests deterministic and reproducible on every run.
When to use: Testing concurrent operations, flush conflicts, parallel mutations, race windows between read and write, lock contention scenarios, any code where timing affects correctness.
When not to use: Sequential-only code, simple unit tests, UI components, read-only operations, code with no concurrency concerns.
Rationalizations (Do Not Skip)
| Rationalization | Why It's Wrong | Required Action |
|---|---|---|
| "setTimeout is good enough" | Timing-based tests are inherently flaky -- they pass 99 times, fail on CI | Use barriers for deterministic control |
| "It passed 10 times, it's fine" | Heisenbugs hide in timing windows that haven't been hit yet | Barrier at every interleave point |
| "We don't have race conditions" | Any concurrent code has race windows | Write a barrier test to prove it |
| "Barriers are too complex" | 15 lines of setup prevents hours of debugging flaky failures | Copy the pattern from references |
What To Protect (Start Here)
Before generating barrier tests, identify which concurrency decisions apply to your code:
| Decision | Question to Answer | If Yes → Use |
|---|---|---|
| Concurrent writes must not lose data | Can two writes to the same resource happen simultaneously? | assertPreservesConcurrentItems |
| Failed operations must not corrupt state | Can a failure leave behind partial or inconsistent state? | assertPreservesOnFailure |
| Event/message ordering must be preserved | Does processing order affect correctness? | assertSequenceContinuity |
| Retry state must reset on new input | Can stale retry counts or error flags affect new operations? | assertRetryCountReset |
Do not generate tests for decisions the human hasn't confirmed. A barrier test without a named invariant is coverage theater — it proves the barrier works, not that the system is correct.
Included Utilities
// Barrier infrastructure + invariant assertions + test data generators
import {
createBarrier,
createTrackedBarrier,
releaseAllBarriers,
assertPreservesConcurrentItems,
assertPreservesOnFailure,
assertSequenceContinuity,
assertLastSequenceCorrect,
assertRetryCountReset,
createTestItems,
} from './test-fixtures.ts';Violation Rules
inadequate_barrier_coverage
Race conditions MUST have barrier tests at each interleave point. If there are N interleave points in the code, there must be N barrier test cases. Severity: must-fail
flaky_timing_test
NEVER use setTimeout, sleep, or arbitrary delays for concurrency testing. Use barriers for deterministic control. Severity: must-fail
Companion Skills
This skill provides deterministic timing control for concurrency tests, not concurrency architecture guidance. For broader methodology:
- Search
concurrencyon skills.sh for lock-free patterns, actor models, and thread safety analysis - Concurrent state machines need transition coverage — use model-based-testing for N*N transition matrices and guard truth tables
- Concurrent failure recovery needs resilience testing — use fault-injection-testing for circuit breaker, retry policy, and queue preservation under concurrent load
Quick Reference
| Scenario | Pattern | Example |
|---|---|---|
| Block before transaction | barrier.wait() in transaction start | Verify concurrent writes preserved |
| Block during I/O | deferred.promise as mock return | Verify queue depth during processing |
| Multiple interleave points | Multiple barriers, release in sequence | Test all orderings of concurrent ops |
| Cleanup | afterEach(() => releaseAllBarriers()) | Prevent test hangs |
See patterns.md for full code examples, deferred promise alternative, and framework adaptation guide.