PowerDNS DNSdist 2.1.0-beta2 Released

[rgacogne]

https://blog.powerdns.com/2026/03/12/powerdns-dnsdist-2.1.0-beta2-released

[paddg]

I'm wondering how expensive shuffle=true in newPacketCache() is, in terms of CPU consumption. Is it possible to estimate that, or is it easiest to just try it out?

[rgacogne]

It is not easy to estimate because the impact depends on the number of cache hits with multiple A/AAAA in the answer section. I would expect it to be quite inexpensive, though.

[paddg]

Okay, thanks! I'll try it with real-world recursive queries and see if CPU usage increases, then report back.

[paddg]

I don't see a significant increase in CPU usage with real resolver traffic. I find that amazing. Thank you so much, @karelbilek, @edmonds and everyone else involved. This feature is very useful.

[karelbilek]

@paddg the shuffling is done only in a subset of cases, where it's safe to just shuffle the bits directly (but that's actually most of the time); there is no new packet being created/allocated, it's switching the memory in the cache. So it should be also fast. But there is still some work being done. It's using Fisher-Yates algorithm which is O(N) with respect to number of elements.

[paddg]

While doing tests with this beta, I noticed a lot of these messages appear constantly in the journal. This is new and presumably not intentional.

msg="Had a downstream timeout" subsystem="backend" level="0" prio="Info" ts="1773388540.795" backend.address="192.168.1.2:534" backend.name="" backend.protocol="DoUDP"  ....

[rgacogne]

Thanks! #16979 will fix it.