EADDRINUSE: address already in use :::3000 / 0.0.0.0:8080

Your server tried to open a port and the operating system said someone is already sitting there. Nine times out of ten, that someone is an old copy of your own server.

What the error means

Only one program can listen on a given address and port at a time. When a second one tries, the OS refuses:

Error: listen EADDRINUSE: address already in use :::3000
Error: listen EADDRINUSE: address already in use 0.0.0.0:8080
OSError: [Errno 98] Address already in use          # Python / Linux
OSError: [Errno 48] Address already in use          # Python / macOS
bind: address already in use                        # Go, C, Docker
WSAEADDRINUSE (10048)                               # Windows

It is the opposite of localhost refused to connect, where nothing is listening at all. DNS never got involved — if the browser said the name itself was unknown, that is ERR_NAME_NOT_RESOLVED instead.

Find what holds the port

Read the process name before you stop it. node, python, your app name → leftover. postgres, AirPlay, a system service → pick a different port. On recent macOS, AirPlay Receiver owns ports 5000 and 7000 (Flask's classic clash).

IPv4 vs IPv6 (::: vs 0.0.0.0)

Node's default server.listen(3000) often shows as :::3000. That can block a second process that tries 0.0.0.0:3000 (or the reverse) depending on net.ipv6.bindv6only. If two tools "shouldn't" collide but do, check both families with ss -ltnp.

SO_REUSEADDR myths

• What it actually does: lets a new socket bind while old connections are still in TIME_WAIT. Most frameworks (Node, Go's Listen, Python's HTTP servers) already set it.
• What it does not do: let two live listeners share the same host:port. If another process is in LISTEN, you still get EADDRINUSE — raising SO_REUSEADDR will not help.
• SO_REUSEPORT (Linux/BSD) is different: multiple sockets can listen together for load-balancing. Node's cluster / some proxies use it. It is not a "just make the error go away" flag.

# Plain Python (frameworks usually do this for you)
import socket
s = socket.socket()
s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
s.bind(("0.0.0.0", 8080))

Why it keeps happening

• Old copy in another terminal — or one that survived closing the editor. Ctrl+Z only suspends and keeps the port; use Ctrl+C.
• nodemon / watch mode starting the new copy before the old one released the port. Handle SIGTERM/SIGINT and call server.close().
• systemd leftover — you started the app by hand and the unit is still enabled. systemctl stop … then start only one way.
• Multiple Node workers — cluster.fork() or PM2 instances all told to bind the same port without SO_REUSEPORT / a proper master. Either let the master listen and workers share, or give each worker a unique port behind a proxy.
• Docker port maps — two Compose services with "3000:3000", or a host process plus a published container. docker ps --filter publish=3000 shows the container side; ss -ltnp shows the host listener Docker created.
• Two apps configured for 3000 (frontend + API) or two Compose services publishing the same host port.

Pick a free port

Most tools take a port option: PORT=3001 npm start, vite --port 5174, flask run --port 5001, python -m http.server 8001. Listening on port 0 asks the OS for any free port (handy in tests). Ports below 1024 need admin rights — that is EACCES, not EADDRINUSE.

Not sure what a mystery PID is? Paste the process name into What is this?.

Related

Reachable locally but not from outside? · Common port numbers (3000, 5000, 8080...) · localhost refused to connect · ECONNREFUSED ::1 · ssh Connection timed out · ERR_NAME_NOT_RESOLVED · JavaScript heap out of memory · EMFILE too many open files

Questions

Published 4 November 2026 · updated 9 November 2026.