---
title: "JMeter Socket Closed Connection Reset"
description: "Fix JMeter connection reset and socket closed errors under load: server limits, idle timeouts, keep-alive, injector ports, and TLS drops. Symptom and fixes."
url: https://docs.jmeter.ai/topics/errors/socket-closed-connection-reset/
lastUpdated: 2026-10-01
source: docs.jmeter.ai
---

# JMeter Socket Closed Connection Reset

## Symptom

Samples fail mid-run or under higher concurrency with messages such as:

- `Connection reset` / `Connection reset by peer`
- `Socket closed` / `SocketException`
- `Broken pipe`
- Often shown as [Non HTTP response code](/topics/errors/non-http-response-code/)

Unlike pure **connection refused**, some samples may succeed first; failures increase with load or longer tests.

## Common causes

| Cause | Detail |
| --- | --- |
| Server or load balancer limits | Max connections, rate limits, or worker exhaustion |
| Idle timeouts | LB/proxy closes keep-alive connections JMeter tries to reuse |
| Injector ephemeral port / file descriptor exhaustion | Many short connections from one host |
| TLS session / middlebox interference | Intermittent handshake or reset under load |
| SUT crash or restart | Bursts of resets when the app dies |
| Aggressive timeouts | Client gives up; appears as socket errors depending on stack |
| Network path instability | VPN, NAT, or multi-hop firewalls |

## Fix (ordered)

1. Capture **error rate vs active threads** on the [dashboard](/user-manual/generating-dashboard/) and SUT metrics (connections, 5xx, restarts).
2. Reproduce with **moderate** threads; note whether failures are immediate or after idle.
3. Align **keep-alive** behaviour with the environment (HTTP client settings / connection reuse). If a proxy closes idle sockets quickly, reduce reuse or match idle timeouts.
4. Check SUT and LB **max connections** and **idle timeout** documentation for your stack.
5. On the injector: watch open files, TIME_WAIT, and CPU; split load across engines if the client is saturated ([distributed testing](/topics/distributed-testing/), [Thread Calculator](/tools/thread-calculator/)).
6. Distinguish from **ConnectException** (never connected) vs reset (peer closed an established or half-open connection).
7. Review TLS separately if resets cluster around HTTPS only ([SSLHandshakeException](/topics/errors/ssl-handshake-exception/)).
8. Ensure you are not melting the SUT with zero think time and huge threads ([best practices](/user-manual/best-practices/)).

## Related tools and topics

| Resource | Use when |
| --- | --- |
| [Throughput stuck](/topics/errors/throughput-stuck/) | Resets plus flat RPS |
| [OutOfMemoryError heap](/topics/errors/out-of-memory-heap/) | Injector death causing drops |
| [Coordinated Omission](/tools/coordinated-omission/) | Throughput below target while latency looks fine |
| [Grafana / Backend Listener](/topics/grafana-influx-backend-listener/) | Live error charts |

## Frequently asked questions

### Is connection reset always a JMeter bug?

Rarely. It usually means something on the path or server closed the TCP connection. Validate with server and load-balancer metrics.

### Why only under load?

Connection pools and OS limits appear only when many sockets are open. One-thread tests often hide them.

### Should I disable keep-alive?

Sometimes as a diagnostic. Prefer fixing idle timeouts and server capacity; disabling keep-alive increases handshakes and can worsen injector load.

## Continue Learning

→

### Next Practical Step

Correlate JMeter error timestamps with SUT connection counts and load-balancer idle timeout settings.

📖

### Related Reference

- [Non HTTP response code](/topics/errors/non-http-response-code/)
- [Best practices](/user-manual/best-practices/)
- [Distributed testing](/topics/distributed-testing/)