[Pdns-users] authoritative server freezes while processing NOTIFYs

bert hubert bert.hubert at netherlabs.nl
Thu Mar 3 10:00:57 UTC 2011


On Wed, Feb 16, 2011 at 10:53:05AM +0000, Richard Poole wrote:
> We have a setup with a non-public master and two publicly visible
> slaves. All three servers are using the bind backend. Sometimes we need to
> change many zones, say about 2000, at the same time, causing the master to
> send out 2000 NOTIFYs to the slaves. We find that while they're processing
> them and doing the corresponding zone transfers from the master, they
> frequently stop replying to incoming queries for several seconds at a
> time. Has anyone else seen this? Is it something we can fix by different

Hmm, this was not the original design. 

> configuration, or a consequence of having a lot of simultaneous NOTIFYs
> being processed, or a consequence of using the bind backend?

The BIND backend is somewhat special in that an incoming AXFR is first
spooled to disk, and then reloaded from that same disk.

However, the BIND backend is also fully multithreaded and should not
generally stop answering questions in that case, except for zones in fact
being reloaded, and even those should only suffer a brief interruption.

> We are running with retrieval-threads=1 and distributor-threads=1 -
> I can give the rest of our conf if anyone thinks it's relevant but most
> of it is left at defaults.

There is probably some dynamic going on between various parts of PowerDNS.
If I recall correctly you run with a truly staggering amount of zones on the
BIND backend, which may have something to do with it.

If at all possible I'd like to reproduce your problem, and will contact you
off-list.

	Bert



More information about the Pdns-users mailing list