aarcane at aarcane.org
Sat Nov 27 21:08:38 UTC 2010
While my knowledge of databases is admittedly limited, I know databases
can scale far more easily than filesystems can for increased
performance, and can even add their own layer of redundancy and failover.
On 11/27/2010 12:05 PM, Simon Hobson wrote:
> Soren Aalto wrote:
>> We did that on our campus, and it didn't seem to work. The one
>> time we ran out of disk space on one of the DHCP servers, we just
>> had machines start losing their leases & needing to be rebooted.
> As Steinar says, if you run out of disk space, things are going to
> fail in messy ways.
>> It's also a giant mess to maintain -- as any static assignments have
>> to be put in the config files on both machines.
> That's the case whichever method you use, but it's really no big deal.
> The trick is to use include statements for common parts of the config
> and then use whatever tools you prefer to sync the files from one
> machine to another. I do something similar at work with our DNS - I
> have a script that generate the configs for master and slaves, and
> then uses rsync to transfer the slave config to the slave servers.
> If you hand edit the files then yes, it will be a pain - but managing
> two configs by hand is the wrong way to do it for anything but trivial
> Christ Schlacta wrote:
>> Is there no database backend for leases? I may be wrong, but I think
>> that would greatly simplify redundancy AND failover.
> It's something that's been much talked about. The problem is going to
> be performance while maintaining RFC compliance. The RFCs **require**
> the server to commit leases to permanent storage before offering them
> to the client - in the current design that's done by appending a lease
> record to the leases file (and then telling the OS to sync the changes
> to disk) which is about the fastest operation you will get.
> Once you add in a "proper" database, then you are at the mercy of that
> database for your response times - not forgetting that in it's current
> form, the program is (I believe) single threaded.
> Dan Letkeman wrote:
>> Do you know if there is any documentation for running them in
>> failover mode?
> man dhcpd.conf
>> If I run them in a non-overlapping state, is there any way I can set
>> them up so that the first one always answers first, and the second one
>> always answers second? Sort of there, but only if needed. This way
>> its there for the brief time needed to fix the first one, and the
>> range doesn't need to be large on the second one. And my monitoring
>> system will notify me when the first dhcp server is not responding.
> I think you probably want the min-secs declaration.
More information about the dhcp-users