Repair Keep-Alive Socket Leaks

The problem

Your application suffers from intermittent ECONNRESET errors, socket hang-ups, or rising file descriptor counts that eventually exhaust your connection pool. These issues often appear after idle periods or upstream deployments because the HTTP client lifecycle doesn't align with the infrastructure's timeout boundaries.

What it does

• Identifies leaked HTTP clients and unconsumed response bodies that cause pool exhaustion.
• Synchronizes client-side idle timeouts with upstream load balancer and proxy boundaries.
• Implements cancellation-safe cleanup to prevent stranded sockets after aborted requests.
• Establishes deterministic regression tests for connection reuse and recovery.
• Standardizes client ownership to prevent per-request agent construction.

Frameworks & tools

Works with any repository using standard HTTP clients like Axios, node-fetch, httpx, requests, or aiohttp. Utilizes Bash for local verification and repository-specific testing frameworks.

Why this beats prompting it yourself

Generic AI advice often suggests simply increasing pool sizes, which only masks leaks. This skill enforces a strict connection contract, tracing ownership from construction to shutdown and proving the fix with local synthetic fixtures that simulate upstream failures.

Use cases

• Fixing first-request failures that occur after service idle periods.
• Stopping memory and file descriptor growth in long-running worker processes.
• Ensuring streaming responses release sockets even when a consumer aborts early.
• Hardening clients against ECONNRESET during upstream rolling deployments.

Known limitations

Cannot modify production infrastructure like load balancers or DNS settings. Does not support automatic retries for non-idempotent requests unless a safety contract is already present in the code.