[Samba] Building two redundant servers without clustering

[Mitch (WebCob)]

Hi M

> 1.: Use drdb to build a RAID1 across the two host's filesystems. If  one
> host fails, the RAID runs in degraded mode but it runs - or does it
> crawl anyway because drdb is slow?
[Mitch says:] I've never used this, and a quick google doesn't give me
anything useful - what's the home page?

> 2.:  Use rsync. If I remember correctly, rsync was not supposed to run
> permanently as a daemon to keep two filesystems in sync(?). I could live
> with that, but how big is the overhead if I ran rsync every 5 or 10
> minutes?

[Mitch says:] This is something I've been playing with. Wasn't having any
problems until the filesets got largish (>40GB) - now I see performance
issues.... one key is to write your sync script with a lock - you don't want
cron starting a second one while the first is still running - at some point,
I'd like to investigate some performance enhancements to rsync - either to
pull a file list from fam or something to eliminate deep traversal of
unmodified dirs (may involve a file system hack).

> I want to achieve a trouble free passive fallover. I one host fails,
> people might have to login again and they even might have to wait up to
> 30 minutes but then it has to work and they have to get all their files.
> 
> This is a public school and data is not worth real money most of the
> time, but once in year there are final exams written and if the server
> breaks down the whole exam has to be redesigned - that could bring me
> into the news.
> 
[Mitch says:] Personally (SCSI bigots aside ;-) I don't have any issues with
SATA drives for a low demand server - sure ultra fast scsi would be quicker,
but a large mirror ALWAYS beats smaller drives in raid 5 or other more
controller dependant setups. There are even a few controllers that do the
raid as a "hardware" sata raid - Adaptec has one - 3ware too I think. They
have proper recovery tools and so on for FreeBSD and Linuxes I think.

m/