Crash and burn during performance test

[SuperFunks]

I did performance testing on CoreDNS with Unbound DNS Recursion support yesterday evening, and im gutted 🙁

The test goes fine up until ~100 queries pr. second and then the CPU usage spins out of control to 100% and hundreds og CoreDNS processes spawn in some kind of loop until the entire thing crashes in a panic condition.

I ran the test on current master with dnsperf and standard query file. First in a build with Unbound2 on Debian Stretch, then with Unbound8 on Debian Buster to the same result.

[SuperFunks]

A Resperf report for the record:

[miekg]

[ Quoting <[email protected]> in "[coredns/unbound] Crash and burn du..." ]

I did performance testing on CoreDNS with Unbound DNS Recursion support yesterday evening, and im gutted 🙁 The test goes fine up until ~100 queries pr. second and then the CPU usage spins out of control to 100% and hundreds og CoreDNS processes spawn in some kind of loop until the entire thing crashes in a panic condition. I ran the test on current master with dnsperf and standard query file. First in a build with Unbound2 on Debian Stretch, then with Unbound8 on Debian Buster to the same result.

hmm :sadface: I won't have time to debug. If I would guess it's OOMing because with cgo each goroutine takes 2MB instead of 2KB of memory

[SuperFunks]

As input here, the VM's i tried this on has 24GB available to them - and was not using up that memory at all.

[thergbway]

I have the same issue too. СoreDNS + unbound suddenly start to consume all the cpu-s. Also I see increased memory, threads, go routine and file descriptors consumption. After that my vm stops to respond on healthchecks and then it is rebooted.
I will try to use Unbound instead of CoreDns + unbound as a recursive caching resolver.

[miekg]

[ Quoting ***@***.***> in "Re: [coredns/unbound] Crash and bur..." ]

I have the same issue too. СoreDNS + unbound + cache suddenly start to consume all the cpu-s. Also I see increased memory, threads, go routine and file descriptors consumption. After that my vm stops to respond on healthchecks and then it is rebooted.

as, said previously, happy to merge something that fixes this, but don't look at me.