deployment.md

Deployment

To run bovine as on a webserver, the following information might be useful. I currently have no idea how resource intensive Bovine is. Deployment is currently causing the most load on the server. Of course, I'm not popular, so this might not be a good measure.

I put the source code into /opt/bovine as user bovine in group bovine

Systemd unit

[Unit]
Description=Bovine
After=network.target

[Service]
User=bovine
Group=bovine
Restart=always
Type=simple
WorkingDirectory=/opt/bovine/bovine_blog
Environment="DOMAIN=$MY_DOMAIN"
Environment="ACCESS_TOKEN=$MY_TOKEN"
LimitAS=infinity
LimitRSS=infinity
LimitCORE=infinity
LimitNOFILE=65536
ExecStart=/usr/local/bin/poetry run hypercorn bovine_blog:app --bind unix:/run/bovine.sock -m 777

[Install]
WantedBy=multi-user.target

You have to update $MY_DOMAIN and $MY_TOKEN. The token is the one used in buffalo to access the client to server endpoints. Under ubuntu place this file in /etc/systemd/system/ and then reload systemd via systemctl daemon-reload.

The settings starting with Limit are not fine tuned. In particular, the number of allowed file connections is needed in order to handle the many GET requests from fellow Fediverse servers, when somebody announces your content.

The default value of the number of allowed open files can be viewed on your system with ulimit -n. Having too low number of open files leads to errors such as

[2023-02-07 19:16:10,518] ERROR    asyncio      socket.accept() out of system resource
socket: <asyncio.TransportSocket fd=3, family=AddressFamily.AF_UNIX, type=SocketKind.SOCK_STREAM, proto=0, laddr=/tmp/bovine.sock>
Traceback (most recent call last):
  File "/usr/lib/python3.10/asyncio/selector_events.py", line 159, in _accept_connection
    conn, addr = sock.accept()
  File "/usr/lib/python3.10/socket.py", line 293, in accept
OSError: [Errno 24] Too many open files

Nginx

Nginx can then be configured via in sites-available

	location / {
		proxy_set_header Host $http_host;
		proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
		proxy_set_header X-Forwarded-Proto $scheme;
		proxy_set_header Upgrade $http_upgrade;
		proxy_redirect off;
		proxy_buffering off;
		proxy_pass http://unix:/run/bovine.sock;
	}

Furthermore, one needs to do the usual gymnastics to get SSL working.

Finally, one needs to edit the nginx config, i.e. /etc/nginx/nginx.conf and increase the number of worker connections, e.g.

events {
        worker_connections 4096;
}

This is again necessary to handle the many get requests that occur if somebody announces your content. The corresponding error message is

2023/02/11 13:21:46 [alert] 650#650: 768 worker_connections are not enough

Furhermore, the error message

2023/02/20 11:58:43 [alert] 54804#54804: *7206  socket() failed (24: Too many open files) while connecting to upstream,

can be hopefully be fixed by adding

worker_rlimit_nofile 8192;

to nginx.conf (this one belongs outside the event block).

FIXME: Should I add proxy_buffering?