IsFqdn performance issues

[lobeck]

I have recently added the kubernetes node-cache and noticed a quite high CPU load. The cached node was actually higher than any of the load on the dns server before adding the caching layer.

So after a bit of pprof- and perf-ing around, I've noticed, that the IsFqdn call seems to be excessively expensive. Is this something that can be easily optimized? I'm not too deep into Go and the associated logic here, so feedback is welcome :D

[tmthrgd]

It's definitely showing up in your profiles, the logic is a little tricky, but I definitely see some opportunities to optimise it.

[lobeck]

Jupp, that's why I started looking, as the receiving and sending side are also somewhat expensive, but it's just plain UDP mangling.

I went deeper and did some benchmarking based on the values from the unittests, but there's nothing obvious. So, you are likely right, that it could use some optimization, especially around strings.LastIndexFunc but it's not the performance issue I was looking for. So I guess, the code is fine and "We're holding it wrong"

BTW: This is x86 on AWS, a M1 Mac is 4x as fast.

[tmthrgd]

@lobeck Would you mind trying out #1453 and seeing if that makes a difference to your profiles?