Jupyter重运行Plotly Dash应用遇OSError:端口已占用的解决求助

Hey there, I've run into this exact headache when working with Dash in a VM—super frustrating when you can't just flip to a new port! Let's break down why this happens and the practical fixes that work:

Why does this happen? #

When you run Dash with debug=True, it uses an auto-reloader that spins up a secondary process to watch for code changes. Sometimes when you modify your code, this reloader fails to clean up the old process properly, leaving the port locked by a "zombie" process that's still hanging around.

Solution 1: Disable the auto-reloader (quick no-fuss fix) #

If you don't mind manually restarting the app after code edits, you can turn off the reloader entirely. Just add use_reloader=False to your run_server call:

if __name__ == '__main__':
    app.run_server(port = 8021, debug = True, use_reloader=False)

Pros: No more port lock issues ever again.
Cons: You'll have to stop and restart the app every time you tweak your code.

Solution 2: Use a more reliable reloader (keep auto-updates working) #

The default reloader can be flaky. Install the watchdog library, which provides a far more stable reloader that properly cleans up old processes when you make changes:

First install watchdog via pip:

pip install watchdog

Then update your run_server call to use it:

if __name__ == '__main__':
    app.run_server(port = 8021, debug = True, reloader_type='watchdog')

Pros: Keeps auto-reloading functional, and rarely leaves orphaned processes behind.
Cons: Requires installing one extra library (totally worth it for the peace of mind).

Solution 3: Manually kill the stuck process (emergency fix) #

If you already have a locked port right now, you can find and kill the process using it directly in your VM:

1. First, find the PID (process ID) of the process hogging port 8021:

lsof -i :8021

(If lsof isn't installed, run sudo apt install lsof for Debian/Ubuntu or sudo yum install lsof for RHEL/CentOS first.)

2. Kill the process with the PID you found (replace <PID> with the actual number from the previous step):

kill -9 <PID>

Pros: Fixes the immediate port lock problem.
Cons: It's a manual step you'll have to repeat every time the issue pops up.

Bonus: Switch to JupyterDash (if you're in a notebook environment) #

I noticed you imported JupyterDash but are using dash.Dash—if you're running this in a Jupyter notebook/lab, switching to JupyterDash might help with process management:

# Replace dash.Dash with JupyterDash in your initialization
app = JupyterDash("SimpleExample")

Then run it with your usual port setting:

if __name__ == '__main__':
    app.run_server(port=8021, debug=True)

JupyterDash handles process lifecycle differently in notebook environments, which can cut down on port lock issues entirely.

内容的提问来源于stack exchange，提问作者Manu Sharma