rpc: make websocket ping interval shorter than pong timeout

[zghh]

With the current ping/pong configuration, after the client establishes a websocket connection with the server, if the two do not send any message for a long time, a read error will occur on the server so that the server will disconnect the connection.

This PR adjusts the websocket ping interval according to the official example of the websocket to fix the bug.

[fjl]

Hmm. So initially, I thought, this is obviously a correct fix for an issue. But upon further investigation, I wonder if the issue is even real. We handle ping as follows:

When no message is sent for 30s (wsPingInterval), a ping frame is sent. We then wait for a response from the peer within another 30s (wsPongTimeout).

Are you sure this issue is related to these constants? Did you experience such a disconnect in practice?

[zghh]

When I make 5000-10000 connections to Geth without sending any messages, disconnecting appears in a few minutes. And it would not disconnect after changing these constants.

I think the reason is that the following two places execute concurrently. If conn.SetReadDeadline(time.Time{}) executes first after receiving the pong frame, this issue will occur probability.

go-ethereum/rpc/websocket.go

Line 288 in 99e000c

conn.SetReadDeadline(time.Time{})

go-ethereum/rpc/websocket.go

Line 354 in 99e000c

wc.conn.SetReadDeadline(time.Now().Add(wsPongTimeout))

[fjl]

Alternative patch #27733

Please try.

[fjl]

I submitted an alternative patch because I think it's a logic race. Changing the timeouts will make it less likely, but the race is still there.

[fjl]

I'm closing this in favor of the alternative patch.