[ClusterLabs] Data location

[Ken Gaillot]

On Tue, 2020-09-08 at 16:28 +0200, Seb wrote:
> Hi,
> 
> 
> I'm new to clustering. My purpose is to replace our current server
> (web, 
> mail received and sent, mailing lists, FTP) with an active/passive
> cluster 
> in order to minimize the downtime.
> 
> I've read the "Clusters from Scratch" guide, which seems like a good 
> starting point (thanks to the authors). The guide does not talk much
> of 
> the data that is used by the services, though.
> 
> For instance, what should be done with the emails received on the
> primary 
> node? For the email service to keep working when the primary node
> fails, 
> the emails should be replicated in real time on the secondary node.
> On my 
> Debian box they are stored in /var/mail. From the guide, my
> understanding 
> is that /var/mail should then be located on the partition that is
> shared 
> between nodes through DRBD. Is this the way to go? Should all of /var
> be 
> on the shared partition, perhaps through a symlink?
> 
> What about colder data, such as Apache's conf files for enabled
> websites?
> Should they be on the shared partition, or synchronised by hand?
> Should 
> all of /etc be on the shared partition?
> 
> 
> Many thanks for your guidance!
> 
> Sébastien.

I wouldn't put all of /var or /etc on shared storage since much of that
is host-specific (hostname, caches, etc.).

Whether data can be shared at the disk level (using DRBD, NFS, external
storage, etc.) depends on the service. Some services, like most
databases, have a native data replication method that must be used
because the service keeps some sort of host-specific meta-data. Others,
like Apache, don't care if their configuration and data are shared.

Another possibility is to create containers or virtual machines for
each service, then use shared storage for the images. You can use
bundles (for containers) or Pacemaker Remote guest nodes (for VMs) to
monitor the service inside.
-- 
Ken Gaillot <kgaillot at redhat.com>