[Pdns-users] MASTER/SLAVE problems

[Martin Kuchar]

> Martin Kuchar wrote:
> > master say:
> > May 27 22:42:17 ns pdns[6107]: Queued notification of 
> domain 'supernews.cz'
> > to 81.0.239.51
> > May 27 22:42:18 ns pdns[6107]: Removed from notification list:
> > 'supernews.cz' to 81.0.239.51 (was acknowledged)
> > May 27 22:42:20 ns pdns[6107]: No master domains need notifications
> > 
> > slave say:
> > May 27 22:42:05 ns pdns[23496]: Received NOTIFY for 
> supernews.cz from master
> > 194.213.231.160, we are up to date: 1085674507<=2003072300
> 
> Ok, so you've switched to using automatic serials, but the 
> new serials 
> are less than the old ones. You can use 'pdns_control 
> retrieve [zone]' 
> to force it to repull the zone.


This is not the case (and not a right solution). I played with autoserial
and this domain some hours before this check. I can confirm, that this FAIL
is not about autoserial and while this happened, i have another and newest
serial in database. The 1085674507 was only in cache on master.

Now i test it again:

ns.wproduction.cz is my master
ns.wproduction.com is my slave
cache on master is long (i thing 6 hours)

14:00:00 (cca) i "dig soa supernews.cz @ns.wproduction.cz", so i send the
record to cache and get response, that serial is 2004052702
14:00:05 (cca) i "dig soa supernews.cz @ns.wproduction.com", and get
response, that serial is 2004052702
14:00:10 i update content string in soa record on master with new serial
2004052801

in log on MASTER i have:
May 28 14:00:37 ns pdns[6107]: 1 domain for which we are master needs
notifications
May 28 14:00:37 ns pdns[6107]: Queued notification of domain 'supernews.cz'
to 81.0.239.51
May 28 14:00:38 ns pdns[6107]: Removed from notification list:
'supernews.cz' to 81.0.239.51 (was acknowledged)
May 28 14:00:41 ns pdns[6107]: No master domains need notifications

in log on SLAVE i have:
May 28 14:00:37 ns pdns[23492]: Received NOTIFY for supernews.cz from master
194.213.231.160, we are up to date: 2004052702<=2004052702
-----------------
after about four minutes i change serial on master again. This time to
2004052802

in log on MASTER i have:
May 28 14:14:42 ns pdns[6107]: 1 domain for which we are master needs
notifications
May 28 14:14:42 ns pdns[6107]: Queued notification of domain 'supernews.cz'
to 81.0.239.51
May 28 14:14:43 ns pdns[6107]: Removed from notification list:
'supernews.cz' to 81.0.239.51 (was acknowledged)
May 28 14:14:45 ns pdns[6107]: No master domains need notifications

in log on SLAVE i have:
May 28 14:14:42 ns pdns[23494]: Received NOTIFY for supernews.cz from master
194.213.231.160, we are up to date: 2004052702<=2004052702
-------------------
The problem continues. I'm (almost) sure, that slave read the info about
domain from master cache, not from master database.

Now i try to send "pdns_control retrieve supernews.cz" from slave. Now the
AXFR is processed and i have in database on SLAVE right serial (2004052802)

But of course i cannot call pdns_control retrieve on all slaves after
editing some record on master. At least because not all slaves can be mine.

I thing slave server makes some check about serial after retreiving the
notification and if this check is positive (master is biggest), then invoke
the AXFR. This looks good and i thing the problem is on master side, because
IMHO master MUST clear his chache for domain before sending the notification
to slaves.

best regards,
Martin Kuchar