Processes Best Practices

Process and worker boundaries protect security and stability - spawn safely, size pools honestly, and shut down gracefully under orchestrator signals.

How to Use This List

• Mandatory for any code using child_process or shell: true.
• Revisit when enabling cluster module or PM2 cluster mode.
• Count total DB connections as pool size × workers × pods.
• Pair with Graceful Shutdown runbooks.

A - child_process Security

• Prefer spawn/execFile with argument arrays over exec with shell. Eliminates injection on user-influenced input.
• Never use shell: true with interpolated user strings. Code review blocker.
• Set maxBuffer consciously on execFile or avoid buffered APIs for large output. Stream with spawn instead.
• Reap child processes - always await exit or handle error. Avoid zombies and FD leaks.
• Run children as least-privilege OS user in production. Not root unless absolutely required.

B - worker_threads and CPU

• Use fixed worker pools, not per-request Worker creation. Cap at CPU core count unless profiled otherwise.
• Keep I/O on main thread; CPU transforms in workers. DB and HTTP stay in primary isolate.
• Transfer ArrayBuffers when posting large binary to avoid clone cost. Use transfer list in postMessage.
• Handle worker.on('error') and restart pool workers on failure. Prevent pool shrink to zero silently.
• Do not assume native addons are thread-safe. Consult addon docs before worker use.

C - cluster and Scaling

• Avoid double-scaling cluster workers inside many K8s replicas. Pick orchestrator or cluster, rarely both at full scale.
• Use external store (Redis) for sessions and rate limits with multiple workers. No in-memory assumptions.
• Log process.pid in structured logs. Debug which worker handled a request.
• Refork workers with backoff on crash loops. Prevent CPU thrash on misconfiguration.
• Size connection pools: poolMax × workers × replicas. Stay under database max connections.

D - Signals and Lifecycle

• Handle SIGTERM and SIGINT with idempotent shutdown function. Kubernetes uses SIGTERM.
• Call server.close() and await DB pool.end() before exit. Close BullMQ workers too.
• Set force-exit timer below platform grace period. Typically 25s on 30s K8s grace.
• Fail readiness probe when draining starts. Stop new traffic before shutdown work.
• Treat unhandledRejection as fatal in production APIs. Separate policy from graceful SIGTERM.

E - Architecture Choices

• Default to in-process async I/O for CRUD APIs. Processes/workers are escalation, not default.
• Use queues for durable background work, not fork per email. BullMQ over blind child_process.
• Document ADR when choosing cluster on bare metal vs pod-only scaling. Team alignment on ops model.
• Isolate untrusted code in OS sandbox/container, not just worker_thread. Workers share process privileges.
• Test shutdown in CI integration tests at least once per release train. Catch hang on keep-alive.

FAQs

Related

• child_process - spawn vs exec
• Process Signals & Shutdown - SIGTERM patterns
• worker_threads - CPU pools
• Graceful Shutdown - operations

External Related Links

• Node.js API Documentation - Complete reference for core modules and APIs.
• Node.js Best Practices (GitHub) - Curated production best practices.
• NodeSource Blog & Resources - Enterprise Node.js performance and security content.
• Node.js Design Patterns, 4th Edition - Production-grade application patterns.
• Node.js Cookbook, 5th Edition - Practical server-side recipes.

Stack versions: This page was written for Node.js 24.18.0 (Active LTS), npm 10+, TypeScript 5.6+, Express 5, Fastify 5, and NestJS 11.