DNS resolver does not account for timing

[pouriyajamshidi]

DNS resolver does not account for timing

[pouriyajamshidi]

Hi @sonarypt,

Well, this one is a bit tricky. In order to reproduce:

starting tcpinging a host, say example.com:

tcping example.com 443 -r 2

then disconnect your internet access 😄 and watch the behavior. You will see that the retry resolve takes a few seconds to timeout and we are not adding those seconds to the total downtime.

Let me know if you need more info.

[doclai]

Uhm, I tried your scenario on my machine (Arch Linux - amd64 - go1.18.4). Previously using wifi, I disconnected my internet access by command nmcli radio wifi off. I also add line into printRetryingToResolve() that print out total down time to the terminal. Eventually, retrying to resolve does not take any extra second to timeout. Everything seems to be smooth.

[pouriyajamshidi]

yea, I also tried again and I don't see any hiccups like before. Now I wonder how it had occurred. Because it was the reason to open a ticket for it.

Let's try some more and if it does not occur again, we can close this for now.

[icemint0828]

@pouriyajamshidi
I can see the event happening with the procedure that pouriya mentioned (tcping example.com 443 -r 2 and wifi on/off).

[icemint0828]

@pouriyajamshidi
Perhaps the problem may not occur depending on when you turn on.
The way tcping is designed, if you don't turn it on just before the retry runs, it will continue to use the previous IP address.
Since the old IP is still available (the example.com IP address remains the same), I don't think the problem will occur.

Sorry for the multiple widths, but I'll put up a timed screenshot of when the problem occurs.

retrying to resolve example.com
time for retrying to resolve 809ms

func retryResolve(tcpStats *stats) {
	if tcpStats.ongoingUnsuccessfulPkts > *tcpStats.retryHostnameResolveAfter {

		connStart := getSystemTime()

		tcpStats.printRetryingToResolve()
		tcpStats.ip = resolveHostname(tcpStats)
		tcpStats.ongoingUnsuccessfulPkts = 0
		tcpStats.retriedHostnameResolves += 1
		
		connEnd := time.Since(connStart)
		rtt := connEnd.Milliseconds()
		colorLightYellow("time for retrying to resolve %dms\n", rtt)
	}
}

I guess that this event occurs several times over a long period of time, eventually resulting in a few seconds difference.
I can't take the time to get to the solution successfully, but if you need help again, please call me!

[doclai]

After virtualized a webserver for demo, by adding local domain to /etc/hosts, turning on/off the webserver, I still cannot simulate any situation when that function takes 4~5 seconds to timeout. Could you give more information for that scenario?

[pouriyajamshidi]

@t-ohtaki do you know a situation which this can be triggered?

[pouriyajamshidi]

Thanks a lot @icemint0828. Helpful as always.

I hope this clears the issue @sonarypt