Are you using something different than Phoenix’s channels? The official JS client by default sends a heartbeat every 30 seconds which keeps the connection alive. Otherwise, as José mentioned, proxies and browsers will kill the connection.

We recently had a nasty timing issue where a channel would “randomly” die. The issue was that one of the proxies presumably had a 30 second timeout, so the issue would only be reproducible in the prod environment when the heartbeat would be just a few milliseconds too late

My understanding most browsers and proxies will close the socket if there is no data after a while. You need to send a ping or similar over the wire if you want to keep it open.

What you did set an “infinite” or no timeout value for regular HTTP requests. But you wanted to change the default timeout for websocket connections, which is different to configure. You may need something like

socket "/cable", ServerWeb.YourSocket,
  websocket: [      # <- add this!
    timeout: 3000. # <- add this!
  ],
  longpoll: false

This option is documented here. As far I undertand it still maps to the low-lewel cowboy idle_timeout option that default to 60s, but here you’re controlling the timeout in the websocket connections rather than the HTTP requests.