[Pdns-users] pdns cannot handle large pdns notify

[Brian Candler]

On 24/12/2019 02:25, Jackson Yap wrote:
>
> We realised when we do a pdns_control notify '*' , many axfrs cannot 
> be transferred to our pdns slave. We have about 500+ zones in total.
>
> Only when we do specific domain notify like pdns_control notify 
> 'example.com <http://example.com>’, it will work properly.
>
> My question is, what settings could have provided mass pdns notify 
> from working properly?
>
I'd suggest you try using tcpdump to look at what's happening. Some 
thoughts:

1. Notifies are sent over UDP.  Maybe when you try to send 500 UDP 
packets simultaneously, some of them are being thrown away (e.g. some 
buffer is overflowing).  tcpdump at the receiving server(s) too if possible.

2. Maybe when there are too many incoming connections, some are being 
rejected (look for R[ST] as the response). See 
https://docs.powerdns.com/authoritative/settings.html#max-tcp-connections

3. Maybe your backend database is slow and can't handle that many 
concurrent AXFRs

As for solutions: maybe you're better avoiding doing pdns_control notify 
"*".  One way would be to spread your notifies over a minute or two.  
Another would be to use native replication between your backend 
databases, which avoids the need for AXFR completely.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20191224/397f0a81/attachment.htm>