Ping at scheduled intervals

[aaugustin]

We could add a ping_interval argument to the protocol.

When set, the protocol would start a task to send a ping at regular intervals.

This can help to keep connections open, especially across proxies with short timeouts.

This task would be a simple as:

while self.open:
    yield from asyncio.sleep(self.ping_interval)
    yield from self.ping()

As a bonus, we could kill the connection if no pong is received in one ping_interval.

[bmoscon]

I'd love to see this - I did something very similar for a project that uses this library, so its definitely useful

[RemiCardona]

On our project, we already have something really similar, but with automatic reconnection. I fail to see how I could reuse what is proposed here without reimplementing the whole method/function to fit out own needs.

Don't get me wrong, I'm not opposed to websockets having support for something like this, I just wanted to say that, in my case, I most likely wouldn't be able to use it as proposed. However, I definitely see use for this.

Cheers

[cjerdonek]

As an alternative, since it's so simple, it might be useful to start out by documenting the pattern. Also, for people calling a websockets server from a browser, the same pattern using Javascript would be similarly useful.

[RemiCardona]

While the ping mechanism is officially part of the protocol spec, there's no official API to access it from the JS/browser side. It's up to browsers to figure if/when to initiate pings or pongs to check the connection status.

https://developer.mozilla.org/en-US/docs/Web/API/WebSocket

[cjerdonek]

Yes, I was just suggesting that, if sample code is provided to show how to ping at regular intervals using Python to keep a connection alive, then it could also be useful to provide such sample code for Javascript -- even if there is no "canonical" solution due to the lack of an API.

[aaugustin]

@RemiCardona To signal to the application that the connection is lost, we can provide a Future and complete it when a ping times out. Then it's up to the application to close and reopen the connection (only on the client side).

I think we don't enforce enough structure to handle this automatically.

I don't believe we can inject an exception either — see #338 for a discussion of cancelling the connection handler (on the server side) when the connection is close, a more basic situation than the one we're discussing here.

[vgoklani]

Hey - has this been implemented? Please add :) Thanks!

[aaugustin]

This issue is open, which means this feature hasn't been implemented yet.

[aaugustin]

I proposed an implementation of this feature in #449.

It only deals with sending pings regularly, not with automatic reconnection — that's #414. However, if automatic reconnection is implemented, this feature will make it more reliable because it will detect connection failures more eagerly.

[aaugustin]

I would very much appreciate feedback on my PR, especially if you can test it in realistic circumstances and report whether it behaves cleanly.

[vgoklani]

testing now :) how do I change the ping message?

[aaugustin]

The ping message is a random 4-byte value. This allows matching pongs to pings.

[aaugustin]

(You can send a ping with a specific payload if you want to with the ping API, but that's another story.)

[aaugustin]

I added tests and improved the pull request. I think it's ready to merge.

I'm still looking for a code review and/or reports of whether it works well.