Home › npm & Node.js

Error: socket hang up (ECONNRESET) (Node.js) — how to fix

Error: socket hang up
    at connResetException (node:internal/errors:787:14)
    at TLSSocket.socketOnEnd (node:_http_client:519:23)
    at TLSSocket.emit (node:events:530:35) {
  code: 'ECONNRESET'
}

Error: read ECONNRESET
    at TLSWrap.onStreamRead (node:internal/stream_base_commons:218:20) {
  errno: -104,
  code: 'ECONNRESET',
  syscall: 'read'
}

Quick fix: Retry idempotent requests once on ECONNRESET. On your Node server set server.keepAliveTimeout above the load balancer's idle timeout (e.g. 65000 with a 60s ALB) and server.headersTimeout higher still. Check you use https:// for TLS ports, and look at the server/proxy logs for crashes or timeouts at the same moment.

What it means

ECONNRESET means the peer reset the connection with a TCP RST. socket hang up is Node's message when the socket ended before a response arrived. To your code both mean the same: the request may or may not have reached the server, and there's no response.

Find out which side you are on:

Cause 1: keep-alive race (intermittent, after idle periods)

Clients reuse idle keep-alive sockets. If the server closes an idle socket at the same moment the client sends a new request on it, you get ECONNRESET. Node 19+ enables keep-alive by default in the global agent, and fetch (undici) always pools connections, so this got more common.

Client-side fixes:

// retry idempotent requests once
async function getWithRetry(url, tries = 2) {
  try { return await fetch(url); }
  catch (e) {
    if (tries > 1 && e.cause?.code === 'ECONNRESET') return getWithRetry(url, tries - 1);
    throw e;
  }
}

With http.Agent you can make the client drop idle sockets before the server does: new http.Agent({ keepAlive: true, timeout: 4000 }). Keep it shorter than the server's keep-alive timeout. axios users can add axios-retry.

Cause 2: Node server behind a load balancer

Node's server.keepAliveTimeout defaults to 5 seconds. AWS ALB, GCP and nginx keep upstream connections longer (ALB: 60s). The proxy reuses a connection that Node already closed, so clients see 502 and the proxy logs a reset. Make Node wait longer than the proxy:

const server = app.listen(3000);
server.keepAliveTimeout = 65_000;   // > LB idle timeout (60s)
server.headersTimeout   = 66_000;   // must be > keepAliveTimeout

Cause 3: the server timed out or crashed

Long requests can be killed by a timeout anywhere along the path: Node's server.requestTimeout (300s default since Node 18), an nginx proxy_read_timeout (60s), a cloud LB, or the upstream app. A process crash or OOM kill mid-request also resets the socket. Check the server logs and dmesg at the exact time. For long jobs, return 202 Accepted and poll, or stream progress. See 504 Gateway Timeout.

Cause 4: protocol mismatch

Cause 5: the server rejects the request

Some servers and WAFs reset instead of returning 4xx: request bodies over the size limit, missing Content-Length, a blocked User-Agent, rate limits, or too many concurrent connections from one IP. Throttle concurrency (e.g. p-limit) and compare with a curl request.

Server side: don't crash on client resets

An unhandled 'error' on a socket or stream crashes the process with Error: read ECONNRESET. Handle it:

req.on('error', (err) => { if (err.code !== 'ECONNRESET') console.error(err); });
server.on('clientError', (err, socket) => socket.destroy());

With stream.pipeline() errors are forwarded to the callback instead of being thrown.

Last updated 2026-11-15.