Hi altogether,
thank you for sharing your thoughts.
I did some testing. The records table has around 115k rows. When I send
single updates for a zone that has ~20 rows (out of the 100k) I get
about 100 updates per second (what would be sufficient in my use case).
When the zone to be updated has 80k records I only get 1 ddns update per
second.
I activated the mysql query log and analyzed (explain) them. All
statements use the indexes and, when running them individually, they
perform well.
I will review my mysql configuration if there is optimization possible.
Maybe the query cache is to small or something else.
Cheers
Thomas
Am 25.02.2016 um 18:25 schrieb Ruben d'Arco:
On Thu, Feb 25, 2016 at 07:54:26AM -0600, [email protected] wrote:
On Thu, Feb 25, 2016 at 12:01:19PM +0100, Thomas Mieslinger wrote:
Hi,
today I wanted to migrate my ddns master from bind to pdns. 3 DHCP
Servers sent a about 50 updates per second to the pdns 3.4.8 and
only one or two updates per second could be successfully commited to
the database.
All failed transactions rolled back after trying to edit the SOA record.
I'm using 5.1.73-log MySQL Community Server.
I'm using REPEATABLE-READ transaction-isolation and
distributor-threads=2
receiver-threads=10
as pdns performance settings.
I tried to disable SOA-EDITs with domainmetadata
SOA-EDIT-DNSUPDATE | SOA-EDIT
SOA-EDIT | NONE
but that didn't change anything.
What DDNS performance do you get from your pdns instances with which
settings?
Thanks Thomas
Hi Thomas,
You will need to investigate your DB performance. Turn on query logging
and slow query logging. Also check system I/O stats to see if you have
a bottleneck there. We are just getting started on adding a DDNS component
to our network so I do not have any firsthand experience with it and the
queries it uses. Good luck in your hunt for the bottleneck.
Regards,
Ken
Just to add to this, there is a lock for every update message. This is to avoid
2 update messages from causing some transactional issues.
There's also a comment above that lock that the lock might should be placed per
zone, allowing multiple zones to be updated at the same time.
When i did some testing on this, it seemed that this was the limiting factor.
It's just not fast if 1 update needs to wait for the other to be completed.
But, at the time, i didn't check if this could be done better.
You mentioned that you have 3 dhcp servers - i'm assuming they all update the
same zone?
Regards,
Ruben
_______________________________________________
Pdns-users mailing list
[email protected]
http://mailman.powerdns.com/mailman/listinfo/pdns-users
_______________________________________________
Pdns-users mailing list
[email protected]
http://mailman.powerdns.com/mailman/listinfo/pdns-users
