By name: effect-ts-concurrency description: Concurrent operations in Effect-TS v4 — fiber management, interruption handling, bounded parallelism using Effect primitives.
v4 Renames
| v3 | v4 |
|---|---|
fork |
forkChild |
forkDaemon |
forkDetach |
forkAll |
DELETED → use Effect.forEach with forkChild per element |
forkWithErrorHandler |
DELETED |
Core Principles
- Concurrency must be intentional and bounded to prevent resource exhaustion
- Fibers are lightweight but not free — avoid unbounded creation
- Interruption must be handled properly for resource safety (finalizers must run)
- Choose coordination primitives based on actual coordination needs — don't over-coordinate
- Prefer collection concurrency (
Effect.forEach) with bounds over manual fork management Semaphorefor resource limiting,Queuefor producer/consumer with backpressureDeferredfor one-time synchronization between fibers,Reffor coordinated state updates- Proper error propagation in concurrent contexts — don't swallow errors from fibers
Preferred Primitives (v4)
| Primitive | Use for |
|---|---|
Effect.forEach + concurrency |
Bounded parallelism over collections (replaces deleted forkAll) |
Effect.forkChild |
Supervised child fiber — tied to parent Scope |
Effect.forkDetach |
Detached fiber with explicit supervision strategy |
Semaphore |
Resource limiting (DB connections, file handles) |
Queue (bounded/sliding) |
Producer/consumer with backpressure |
Deferred |
One-time synchronization between fibers |
Ref + atomic ops |
Shared state with Effect.update / Effect.modify |
Effect.race + cancellation |
Competitive execution with loser cleanup |
Effect.addFinalizer / ensure |
Resource cleanup on fiber interruption |
Effect.scoped |
Deterministic cleanup for concurrent resources |
Anti-patterns
| Pattern | Detect | Severity |
|---|---|---|
| v3 API usage | fork, forkDaemon, forkAll, forkWithErrorHandler |
HIGH |
| Unbounded concurrency | Many fibers without Semaphore or concurrency limits | HIGH |
| Ignoring interruption | No finalizers/cleanup on fiber interruption | HIGH |
| Lost updates | Concurrent Ref modification without atomic ops (Effect.update/modify) |
HIGH |
| Fire-and-forget | Effect.forkDetach without supervision or error handling |
HIGH |
| Missing error propagation | Concurrent ops swallowing errors | HIGH |
| Blocking in fibers | Sync blocking calls preventing fiber yielding | HIGH |
| Unbounded Queues | Queue.unbounded without backpressure consideration |
MEDIUM |
| Race conditions | Check-then-act on shared state without atomic ops | MEDIUM |
| Misusing primitives | Queue for simple state sharing (use Ref), Deferred for repeated signaling (use Queue) | LOW |
| Over-coordination | Semaphore/Queue when simple parallelism with bounds suffices | LOW |
Severity
| Level | Criteria |
|---|---|
| HIGH | Resource exhaustion, memory leak, deadlock, data corruption, v3 API won't compile |
| MEDIUM | Degradation under load, stuck fibers, potential race conditions |
| LOW | Over-coordination, suboptimal primitive choice — correct but not idiomatic |
Output per finding
- File:line location
- Concurrency issue (from table)
- Recommended bounded/safe alternative with v4 API
- Risk level
Guardrails
- Never remove interruption handling or finalizers.
- Preserve original concurrency semantics when bounding.
- Avoid coordination primitives where simple bounded parallelism suffices.
- Do not eliminate necessary concurrency — bound and supervise instead.