[Pdns-users] Possible bug observed in PowerDNS Recursor 3.2.1

[Nuno Nunes]

Hello all,


I've gone through the last few months of the ML, up until the
announcement of the release of 3.2.1, and didn't find any reference to
this bug I'm apparently seeing, so I'm reporting this to you all for
help.

I work at an ISP where we have a number of servers running PowerDNS
Resolver 3.2.1 as our customer-facing resolvers.

We have had this setup for a few months now and sometimes a weird thing
happens (and no, I can't reproduce it in any deterministic way and it
only happens sometimes): when the TTL for a record of a given zone
expires and a new request comes in for it, some of the caches on the
farm go out and get the new information, but some others just seem to
ignore the TTL and stick with the old data forever.
This is most notable when a zone changes name servers and the owner of
the zone comes complaining to us that we still have the old data, even
after the appropriate amount of time has elapsed for it to have been
refreshed (and on these cases we typically observe this behaviour on NS
records, but we have observed it on A records also, for example).
Now we have had this happen at least three times over the last months
and we've tried to narrow it down to a specific set of circumstances,
but we haven't been able to really find a pattern.
What we do know is that every time this happens, some of the servers
behave correctly (TTL expires => get new data) and others don't. And
when that happens not even `rec_control wipe-cache` will work.
The servers are all identical (same HW, same OS and same SW).

Has anyone else observed something like this before? Is it a known bug
and I just failed to find it being discussed? More importantly: is there
a fix for this behaviour?


Thanks,
Nuno Nunes


-- 
Nuno Nunes (nuno.nunes at optimus.pt)
Tel: 351931003485 | Fax: 351931023485
Edifício Optimus
Av. D. João II - Lt. 1.06.2.4
1990-095 Lisboa
Portugal