multi-master with mysql backend

Mon Feb 14 15:52:24 UTC 2011

I'd keep two copies of the BIND config, one that has all the zones as "master", and one that has all the zones as "slave".  When the master dies, run a little script on a slave that freezes the zones, edits the SOA to make that server the MNAME and increment the serial, then thaws the zone. Swap out the config with the "master" config, and now you have a new master.

Before the broken master comes back online, swap out its config with the "slave" config.

No need for rsync or mysql, BIND replication does all the work for you.  Just be sure the updates go to the server listed in the MNAME field of the SOA.

Mike Mitchell

________________________________
From: bind-users-bounces+mike.mitchell=sas.com at lists.isc.org [bind-users-bounces+mike.mitchell=sas.com at lists.isc.org] on behalf of Bill Larson [wllarso.dns at gmail.com]
Sent: Monday, February 14, 2011 10:39 AM
To: fddi
Cc: bind-users at lists.isc.org
Subject: Re: multi-master with mysql backend

On Feb 13, 2011, at 9:06 AM, fddi wrote:

I do not know why you really don't liket this mysql solution.
OK I am talking of a DNS for HA purposes for grid computing services for exampe, so DNS
resolution must be always working at any cost.
The David solution can be OK, but I want to be sure not to have issues with serial numbers on the two servers
and the mysql solution looks safer to me. You do not have to rsync anything, just have mysql properly configured.

This list is read by many people with extreme experience with DNS and DNS operations.  Using the information that you have provided, many solutions have been provided to meet the requirements that you have stated.  I would suggest that you at least consider this other solutions and not be as adamant about using your proposed MySQL solution.

You state that you have a "HA" requirement.  Please understand that this is not an uncommon requirement for a DNS operation.  In fact, almost all DNS operations have this same requirement.  Just imagine if the root servers did not have a require for "high availability".  The same goes for the "com", "net", "org" servers ("it" also).  This is not an unusual requirement and a standard BIND DNS server provides this very well.

Now, I would also suggest that a MySQL DNS server may not be the best solution for a critical DNS operation.  Please take a look at the benchmark work done comparing BIND using various backend systems, including MySQL.  This can be found at http://bind-dlz.sourceforge.net/perf_tests.html, which is part of the work that the people that developed DLZ for BIND.  The standard BIND server could provide 16,000 queries per second.  The same BIND server using a MySQL backend could handle less than 700 qps.

BIND DLZ using MySQL is NOT what I would consider to be a high performance solution for a DNS operation.  Now, granted, these results were not using a "current" version of BIND, nor was this being run on any "high power" hardware.  Performance of BIND DLZ using a MySQL backend may have easily been improved, but so has the performance of the base system too.  A 20 to 1 performance advantage for a straight BIND solution will be hard to overcome.

So, performance isn't something that a MySQL backend provides to a DNS operation and high availability is something that all DNS server do provide, so your real requirement must be something else, such as being able to update multiple servers at the same time.  But Doug Barton has identified that using "nsupdate" to update both (all) servers also provides a solution to meet this requirement.

So, your "the mysql solution looks safer to me", may be your viewpoint which is not universally agreed upon.  You also have stated "just have mysql properly configured".  This statement is not unique to MySQL but to BIND also.  BIND also needs to be properly configured, no different than with MySQL - nothing unique here.

Now, my single belief for any DNS operation is to follow the KISS principle, "keep it simple, stupid".  A less complex system will be more reliable than a more complex system, because of having less potential points of failure.  This reliability is the single most important requirement for a DNS operation.  A DNS operation that requires running both BIND and MySQL will be inherently less reliable than one running just BIND.

The complexity of a MySQL BIND server makes for a less reliable system than one just using BIND.  The performance of a MySQL based BIND server is much less than a standard BIND server.  Managing DNS information using dynamic DNS provides a simple solution to updating the zone information.  So, what is the actual advantage of using a MySQL backend to BIND?  I'm not convinced that there is any advantage and I am sure that there are many downsides to using this.

Using MySQL for a backend to BIND is a fairly commonly proposed solution but it's actual implementation is not followed up on.  I looked at using MySQL, but the performance limitations were an absolute deal killer.  I set up a simple BIND/MySQL system and benchmarked it and duplicated the performance trends from the BIND-DLZ developers.  Maybe this has been improved, but these results have not be published so we don't know about it.

If you do implement your MySQL solution, please, please, please, keep us informed about how it works for you.  We would like to know more and are always willing to look at new technologies but aren't too accepting of hand waving.

Bill Larson

Riccardo

On 2/12/11 11:33 PM, Doug Barton wrote:
On 02/11/2011 01:51 PM, fddi wrote:
I understand you, but the advantage of having mysql backend is that
if one of the two servers dies, the other keeps running with up to
date informations, and can also be updated wit new informations. When
the  other server comes up again it will automatically sync itself
using mysql replica mechanism. if I use file backend I have to
manually sync it, and how to keep tracks of modifications ?

for this I choose mysql backend

Two questions, how often do you anticipate one of the masters failing, and how much data are you talking about? Generally the number of times a server fails is going to be pretty small, if it's not, you've got bigger problems.

If you're not talking about a huge amount of data here (and from what you've described in previous posts, you're not) then you are fairly dramatically over-architecting your solution here. Personally I think David had a great idea in regards to using nsupdate to update both masters at the same time. If you really think that one of them is going to fail often enough to justify an automated solution than scripting something that utilizes rsync shouldn't be too hard.

hth,

Doug

_______________________________________________
bind-users mailing list
bind-users at lists.isc.org<mailto:bind-users at lists.isc.org>
https://lists.isc.org/mailman/listinfo/bind-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/bind-users/attachments/20110214/bc118824/attachment.html>