Running Reflex via systemd with --env prod param only runs the frontend and not the backend.

[rushout09]

Describe the bug in systemd: env DEV work env PROD fails

running manually : both works

To Reproduce Steps to reproduce the behavior:

Create a systemd file

[Unit]
Description=Databees Website App
After=network.target

[Service]
ExecStart=/home/ubuntu/databees_website/venv/bin/reflex run --env prod
WorkingDirectory=/home/ubuntu/databees_website
User=ubuntu
Group=ubuntu
Restart=always

[Install]
WantedBy=multi-user.target

sudo systemctl daemon-reload sudo systemctl start test.service

Expected behavior ubuntu@ip-172-31-41-251:~$ sudo lsof -i -P -n | grep LISTE systemd-r 325 systemd-resolve 14u IPv4 15785 0t0 TCP 127.0.0.53:53 (LISTEN) sshd 565 root 3u IPv4 17438 0t0 TCP :22 (LISTEN) sshd 565 root 4u IPv6 17449 0t0 TCP :22 (LISTEN) reflex 6894 ubuntu 13u IPv4 68388 0t0 TCP :8000 (LISTEN) python3 6927 ubuntu 13u IPv4 68388 0t0 TCP :8000 (LISTEN) next-serv 6957 ubuntu 20u IPv6 69656 0t0 TCP *:3000 (LISTEN)

Screenshots If applicable, add screenshots to help explain your problem.

Specifics (please complete the following information):

• Python Version: 3.10.8
• Reflex Version: reflex==0.3.6 reflex-hosting-cli==0.1.3
• OS: ubuntu 22
• Browser (Optional):

Additional context Add any other context about the problem here. https://discord.com/channels/1029853095527727165/1184109557946003576

[jackie-pc]

Thanks for the report.

When backend prod runs, it calls out to the program gunicorn. However gunicorn program is only available once the virtual env is activated by source venv/bin/activate. Simply calling venv/bin/reflex is insufficient as that takes care only of libraries within the Python process.

This exposes a bug in the way to propagate errors though - it should have been straightforward from logs ("gunicorn not found"). #2296 above will address this.

[rushout09]

@jackie-pc is there a way to properly initialize venv using systemd?

[jackie-pc]

You would initialize like how you normally would. source ...../activate. E.g.

source ...../activate && reflex run. You can try that in ExecStart: (not sure about exact Systemd support for &&). Putting it in a custom run_from_systemd.sh script that contains the same shell snippet would definitely work. You would call run_from_systemd.sh from ExecStart

[rushout09]

got it. thanks will try this

[martincpt]

I spent a day trying all possible solutions, and I can confirm that the only way to make this work is to create a custom Bash script that activates the environment first. Systemd does not support &&, cannot execute source, and cannot activate the virtual environment via poetry shell in ExecPreStart.

As mentioned, you must create a Bash script like this:

#!/bin/bash

# Navigate to the script's directory to ensure relative paths work correctly
cd "$(dirname "$0")"

# Activate the virtual environment using a relative path
source .venv/bin/activate

# Run the Reflex command with the 'prod' environment using a relative path
.venv/bin/reflex run --env prod

And then the systemd file like this:

[Unit]
Description=reflex daemon
After=network.target

[Service]
User=root
Group=www-data
WorkingDirectory=/your/app/path
ExecStart=/your/app/path/run_reflex.sh
Restart=always

[Install]
WantedBy=multi-user.target

I'm not 100% sure, but another solution could be to include the absolute path to the Gunicorn or Uvicorn binary when executing them via Reflex. However, this would require updating the codebase.

Cheers.