[Samba] about samba failover

[me at electronico.nc]

Le 29/06/2014 06:50, Davor Vusir a écrit :
> 2014-06-28 20:16 GMT+02:00 steve <steve at steve-ss.com>:
>> On Sat, 2014-06-28 at 20:02 +0200, Davor Vusir wrote:
>>
>>> Never the less, domain based DFS works. Thanks to Garming, if I recall
>>> correctly.
>> Hi
>> That's exactly what we want. Two domain file servers carry the same
>> share. One goes down, there's still another left with the same share.
>>
>> Can you point us at a howto?
>> Thanks,
>> Steve
>>
>>
> Hi!
>
> I haven't found a howto. If you can't Explore to
> \\example.org\netlogon, I suggest you revert to Ubuntu 12.04 for
> starters. :)
>
> According to the documentation
> (https://www.samba.org/samba/docs/man/Samba-HOWTO-Collection/msdfs.html),
> it is not possible in my opinion. At least not with automatic
> failover. Because there is not enough information embedded in Sambas
> implementation that informs the (Windows) client which server is to
> prioritize and which is secondary and for how long this information is
> valid. You can create links to several shares but you can't determine
> which server the client will connect to (if it does not connect
> alphabetacally(?)). If client A connects to server A the first time,
> and client B connects to server B, you have to implement two-way
> synchronization. There is no way to determine if the opposite occurs
> at the next logon. The documentation fails to tell.
>
> The only thing you know is that domainbased DFS is possible. The
> availability solution is for you to decide.
>
> I think it boils down to a solution with one (1) link to either a cold
> stand-by server or a hot stand-by server(cluster) of some sort.
>
> If you choose the cold stand-by solution, you'll have to 'relink' to
> the stand-by server manually in case of server crash. Does rsync copy
> open files? How often is enough? Every five minutes? Or is inotify
> (http://www.kutukupret.com/2011/06/28/postfix-one-way-maildir-replication-backup-using-inotify-and-rsync/)
> good enough?
>
> If you choose the hot stand-by solution, I think DRBD will be satisfactory.
This is what my original question was about : RAID1 DRDB over network
http://blog.remibergsma.com/2012/09/09/building-a-redundant-pair-of-linux-storage-servers-using-drbd-and-heartbeat/
I actually use this with 2 xen servers 10.10.20.2 and 10.10.20.3
It creates a virtual ip 10.10.20.1 that holds the NFS whatever server is 
on (with server1 prefered).
Servers are connected through a direct ethernet link via Gigabit NICs 
(that is used for the HA communication) and a 20G RAID1 array is 
actually used to store common files between servers (dhcpd.conf, bind9 
files) for basic HA functions.
I was thinking to use another network RAID1 array for Samba files but 
haven't tried it right now.
I'll let you know, but I now feel confident about the data integrity 
(RAID1).
Nicolas
>
> With the first solution you will loose data/information. The second
> guaranties consistency.
>
> The trick is one link to one server. High avaliability not included.
>
> Regards
> Davor