[Pdns-users] SLAVE refresh performance

[Klaus Darilion]

Am 12.09.2011 15:37, schrieb Peter van Dijk:
> Hello Klaus,
> 
> On Sep 12, 2011, at 12:40 PM, Klaus Darilion wrote:
> 
>> Am 09.09.2011 19:40, schrieb bert hubert:
>>> 
>>> Can you try 'query-cache-ttl=200' and see if this improves
>>> things? If it does, it would help us solve the issue.
>> 
>> Which DB-queries will be cached by PDNS? Only DNS-query related 
>> DB-queries, or all DB-queries (e.g. slave-cycle-interval triggered
>> queries)?
> 
> All queries to the 'records' table (for SOA and for other resource
> types) are eligible for the query cache.

Obviously not all. For example if I trigger NOTIFYs from the master, I
see multiple SOA lookups in the DB, although DNS-SOA queries to the
slave will be answered without DB-queries directly from cache.

>> And what happens if the SLAVE updates the DB due to an incoming
>> AXFR. Will it also expire relevant cache entries?
> 
> As far as I know, the TTL (which Bert suggests to set to 200) is the
> only thing managing expiry. So, no, there would be no automatic
> expiry on a slave domain update. However, we're not suggesting
> query-cache-ttl as a fix; we are wondering whether changing it like
> that would make a difference in your observed performance, just as a
> debugging aid.

No, that does not help.

I have set query-cache-ttl=100 and slave-cycle-interval=60. When the
"slave-admin" queries would be cached as well, then I should see queries
only every 2 minutes. But I still see the 10 000 SOA DB-queries every
minute. Thus, it seems that "management" DB-queries are not affected by
the query-cache.


regards
Klaus