hello,
i wonder which one is better for my mail server failover and why?
Rsync vs. DRBD

best regards.

IMOHO: What you really want is either SAN shared storage, or a cluster file system.

I love DRBD but have found it unreliable during maintenance events. An upgrade that you would not think pertaining can break it easily. If you use it you need a duplicate environment where you can test every patch and upgrade. (needed in my enterprise environment anyway, but overkill and wasteful for any smaller office)

@hesisaboury:

This is very much a subjective question, and is dependent entirely on your environment. If money is available for it, I agree that a LUN (on a well-performing SAN) would be ideal. For an active-passive cluster, you wouldn't necessarily even need a clustered filesystem. For active-active, of course, you would.

I have no experience with DRBD yet, but listen carefully to wpeckham's comments. If you're going to deploy DRBD for a critical application, then be sure you have a proper test environment and procedures for vetting all system updates before doing them in production.

You probably do not want rsync(1) as a solution, unless 1) your mail server is low volume; and 2) you can accept some level of lost messages. That is an entirely different category than block-level storage (SAN LUN) or block-level replication (DRBD), BTW, so your dichotomy of "rsync vs. DRBD" probably needs more analysis and thought.

thank you guys for your guides,
i want to run failover with DRBD+pacemaker (debian Lenny) but after two weeks i haven't succeed yet, i dont know if i continue or stop it and use rsync+pacemaker..
my mail server is very important and i dont want see it's failur..


thank u...

Rsync is one of my goto tools. Nice fast secure. It is NOT, however, any kind of realtime solution.

I have made the combination of OpenVZ, heartbeat, and DRBD work for a high availability server, but it was tricky, fragile, and complex. There are purposes I might recommend it for, but mail is not one.

A cluster or distributed (shared) file system is perfect for what you want to do. the replication is real-time and no storage failover is required only assumption of the mail server network address. While a cluster would allow you to use two or three server and do load-share as well as instant failover, it might be overly complex for your needs. The shared filesystem and intelligent failover are the only parts you need to make this elegant.

A few years ago this would have been difficult, today there are several filesystems to choose from and dozens of how-to documents on the web. Look up some of the clusters or distributed file systems, pick a couple of the newer, and google for documents related.