stability of persistent websocket connections

[yemaney]

Deploying an app that accepts websocket connections. A client can send messages to the app through the websocket connection. For every message a client sends the app would query a database and send the results back to the websocket in batches of 10000 rows.

Since the data can have over a million rows. A persistent websocket connection needs to be maintained between the client and app, in order for all the batches to be recieved.

Any suggestions, on a client maintaining such a persistent connection? Does this pattern make sense? Is this a good usecase for python websockets?

[aaugustin]

It should work.

[RemiCardona]

confirming that it does: I had IoT devices in remote locations with open connections for weeks.

I do recommend using some sort of keep-alive mechanism so that your app and/or client can figure out if the underlying transport is dead in a timely manner (and reconnect if needs be).

A websocket ping loop in a background asyncio task is very effective and simple to implement.

Another mechanism is the TCP keep-alive option. On Unix systems:

import socket

ws_sock = ws_obj.reader._transport.get_extra_info('socket')
ws_sock.setsockopt(socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1)
ws_sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_KEEPINTVL, keepalive_intvl)
ws_sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_KEEPCNT, keepalive_probes)
ws_sock.setsockopt(socket.IPPROTO_TCP, socket.TCP_KEEPIDLE, keepalive_time)

(note this is for an older version of websockets, YMMV)

Pros and cons for the ping loop:

• pure python, works on all platforms where websockets works
• goes through all the layers (esp reverse proxies) from the client to the server

• adds a task and a little of housekeeping

Pros and cons for the TCP keep-alive:

• all the work is done by the OS, it's essentially free
• very little data is exchanged, an order of magnitude less than the WS ping (costs may be an issue on mobile networks)

• OS specific, code above does not work on windows (though it should have an equivalent feature)
• does not go through all the layers, it stops at the first reverse proxy (enabling keep-alive on the server side is usually pointless if you have nginx or varnish handling TLS on behalf of your app)

HTH!

[aaugustin]

The ping loop is built-in and enabled by default in websockets these days.

There are some traps in this area, which are documented here.