dht-ping crashes when attempting to ping a non-Bittorrent server address about torrent HOT 4 CLOSED

Oh dear, this looks non-trivial.

from torrent.

The first fix was bogus, see the later commit. Let me know how it goes.

from torrent.

Nice fix.
Do you think transaction.userOnResponse() should be called even if the query fails? Or was it just the easiest fix you thought about?

It seems to me that when the ping fails, userOnResponse() is called by transaction.close() which is called a couple of times (!) by transaction.timeout() which is called by transaction.timerCallback() the timer for which is assigned by transaction.startTimer() which is assigned by jitterDuration... this stuff looks pretty convoluted.

What was the logic there? What is the purpose of jitterDuration?

from torrent.

I don't actually recall why it works like that, I think it kind of evolved that way. I don't like how complex the response handling is, but it seemed justified at the time.

If you can think of a clearer way, I welcome it. It mainly went this path due to a need to reduce the number of goroutines. When I first wrote the DHT package, the server could easily jump to 200k goroutines, which wasn't compatible with the race detector.

jitterDuration implements slightly randomized timeouts, this is common in many network protocols to reduce the chance of network storms. That is, if a bunch of timeouts all trigger at the same time, a lot of outbound packets are sent too quickly. This occurs when you write a lot of packets over something like UDP, and then set identical timeouts to resend them all.

from torrent.