ERR_CONNECTION_RESET: What It Means and How to Fix It
ERR_CONNECTION_RESET means the connection was cut with a TCP reset. Fixes for visitors (VPN, proxy, antivirus) and site owners (crashes, proxies, Docker).
ERR_CONNECTION_RESET means your browser had a connection to the server and something abruptly cut it off by sending a TCP reset (RST) packet. That "something" can be the server itself (an app that crashed or closed the connection early), a proxy or load balancer in between, or software on your own side such as a VPN, firewall or antivirus. If only one site fails, the problem is usually on the server's end; if many sites fail, start with your own network.
Checked against Chromium's net_error_list.h and error page strings, RFC 9293 (TCP) and the Vite server options in November 2026.
In Chrome (and Edge, Brave and other Chromium browsers) it looks like this:
This site can't be reached
The connection was reset.
Try:
Checking the connection
Checking the proxy and the firewall
Running Windows Network Diagnostics
ERR_CONNECTION_RESET
Chromium's source defines it as error -101 with a one-line comment: "A connection was reset (corresponding to a TCP RST)." Firefox shows the same situation as "The connection was reset".
How it differs from similar errors
The neighbouring errors in Chrome tell you how far the connection got:
| Error | What happened |
|---|---|
ERR_NAME_NOT_RESOLVED |
DNS lookup failed; no connection was attempted |
ERR_CONNECTION_REFUSED |
The server's machine answered, but nothing was listening on that port |
ERR_CONNECTION_TIMED_OUT |
No answer at all |
ERR_CONNECTION_RESET |
A connection existed, then was killed with a reset |
ERR_EMPTY_RESPONSE |
The server closed the connection normally without sending any data |
So a reset tells you the address and port were reachable, at least at first. (This Site Can't Be Reached covers the whole family.)
If you're a visitor
Work through these from least to most effort:
- Reload once. A single reset can be a blip, such as a server restarting during a deploy.
- Try another network. Open the site on your phone with Wi-Fi off. If it works there, the cause is on your network or computer, not the site.
- Turn off your VPN or proxy. VPNs and proxies terminate and re-open connections on your behalf, and a misbehaving one shows up as resets. In Chrome, check Settings > System > Open your computer's proxy settings.
- Pause antivirus web scanning. Security software that inspects HTTPS traffic sits in the middle of every connection and can reset ones it dislikes. Temporarily disable the "web shield" or "HTTPS scanning" feature to test, then turn it back on.
- Try a private window. This rules out an extension interfering with requests.
- Check for a network filter. Office, school and some public networks reset connections to blocked sites rather than showing a block page. If the site works everywhere except one network, that's likely the reason, and only the network's admin can change it.
If every site fails on Windows and none of the above helps, netsh winsock reset from an administrator prompt resets Windows' network stack configuration. It needs a restart, and it can break VPN clients or other software that installed network components, so treat it as a last step.
If it's your site: why servers send resets
A TCP reset means "drop this connection now". RFC 9293 describes hosts that send a RST when an application closes a connection with received data still unread, "to show that data was lost", and a reset is also the answer to traffic for a connection that doesn't exist. In practice, that maps to a short list of causes.
The app crashed or was killed mid-request
If your server process dies while handling a request (an unhandled exception, an out-of-memory kill, a container restart), the operating system closes its connections. Depending on timing, the browser sees a reset or an empty response. Check your app's logs and the kernel log at the moment of the failure:
# app logs (pick the one that fits)
docker logs --since 10m <container>
journalctl -u myapp --since "10 min ago"
# was anything OOM-killed?
journalctl -k | grep -i 'killed process'
A process that's being OOM-killed under load will produce resets that come and go. The Linux OOM Killer explains how to confirm it.
The request was too large
Upload size limits in nginx, other proxies and app frameworks can reject a request before reading the whole body. Some servers then close the connection while the browser is still sending, and the browser reports a reset instead of the 413 the server tried to send. If resets happen only on large uploads, look at body size limits first: 413 Request Entity Too Large lists where they live.
A proxy, load balancer or CDN in front is cutting connections
When there's a reverse proxy between the browser and your app, the reset may come from the proxy. Things to check:
- Timeouts. Long requests (exports, AI responses, slow queries) can exceed an idle or read timeout in the proxy or load balancer, which then drops the connection. Compare how long failing requests ran with the timeouts you've configured.
- TLS setup. A proxy that can't complete the TLS handshake, for example because there's no certificate configured for that hostname, may drop the connection instead of serving an error page. Test the handshake directly with
curl -v https://yourdomain.comand look at where it stops. - Wrong upstream. If the proxy forwards to a port where a different service is listening, or speaks HTTP to a port that expects HTTPS, the upstream may close the connection.
For the config side, see Nginx Reverse Proxy Configuration.
Firewalls and security tools on the server
Host firewalls usually drop or refuse rather than reset an established connection, but intrusion prevention tools, rate limiters and some cloud firewalls do send resets. If resets affect only some visitors (one country, one ISP, heavy users), check those rules.
MTU problems (rare)
When a VPN or tunnel shrinks the maximum packet size and the "packet too big" messages are blocked, small requests work and larger responses stall. This usually looks like a page that hangs or times out rather than a reset, so check it only after the causes above.
If it's your dev server: localhost inside Docker
A very common developer version of this error: you start a dev server inside a container, publish the port with -p 3000:3000, open http://localhost:3000, and get a reset or an empty response.
The cause is that the dev server is listening on localhost inside the container. Docker forwards traffic to the container's own network interface, not to its loopback address, so nothing receives it. Many dev servers default to localhost; Vite's docs, for example, list the default for server.host as 'localhost'.
Make the server listen on all interfaces inside the container:
vite --host 0.0.0.0
next dev -H 0.0.0.0
flask run --host=0.0.0.0
You can confirm from inside the container which address the server bound to:
docker exec <container> ss -ltn
127.0.0.1:3000 means only the container can reach it; 0.0.0.0:3000 or *:3000 is what you want. (ss may not exist in slim images; netstat -ltn is the alternative.) What Is Localhost? explains why the two addresses behave differently.
A quick way to narrow it down
Run the request from the command line, from different places:
curl -v https://yourdomain.com/path
curl: (56) Recv failure: Connection reset by peer confirms a reset. Then repeat the test closer to the app: from the server itself to the proxy, then from the proxy host straight to the app's port. The hop where the reset stops happening is the layer that's sending it. (What Is curl? covers the flags.)
In Chrome, chrome://net-export records a detailed log of a failing load if you need to show someone exactly what happened.
EasySpawn hosts your app on a managed server with SSL issued and renewed for your domain, so the TLS and proxy layer in front of the app is set up for you. Crashes and slow requests inside the app are still yours to fix, but Claude Code runs on the same server and can read the logs while you reproduce the error. See how it works or join the waitlist.
Related: "This Site Can't Be Reached": What the Error Code Tells You · ERR_CONNECTION_REFUSED: What It Means and How to Fix It · ERR_SSL_PROTOCOL_ERROR: What It Means and How to Fix It · ERR_NAME_NOT_RESOLVED: What It Means and How to Fix It
Keep reading
npm run build Fails but npm run dev Works: Why and How to Fix It
Development mode is forgiving; production builds are strict. The common reasons a build fails when dev works — TypeScript and lint errors, case-sensitive imports, missing environment variables, server-only code in the browser, prerendering errors and memory — with the fix for each.
"exec format error" in Docker: ARM vs x86 Images Explained
exec format error almost always means a container image built for one CPU architecture (ARM, like Apple Silicon Macs) is running on another (x86/amd64 servers), or vice versa. How to check, build for the right platform with --platform and buildx, and the other cause: scripts without a shebang or with Windows line endings.