Re: Re: ZFS on a shared iSCSI

On Wed, Feb 07, 2024 at 02:36:18PM +0100, Andrea Brancatelli wrote:
> On Wed, Feb 7, 2024 at 12:25 PM Ben RUBSON <ben.rubson@gmx.com> wrote:
> 
> > > On 7 Feb 2024, at 11:55, Andrea Brancatelli <andrea@brancatelli.it>
> > wrote:
> > >
> > > Would it be "easily possible" to have some (2?) iSCSI LUN exposed to
> > both servers and then activate the pool on one or the other server?
> >
> > Hi Andrea,
> >
> > Yes, I have such configurations :
> > 2 servers with same JBODs configuration.
> > When zpool is mounted on server A, mirroring is done on both local disks
> > and iSCSI disks from server B.
> > When zpool is mounted on server B, mirroring is done on both local disks
> > and iSCSI disks from server A.
> > Works flawlessly.
> > And you can have one of the 2 servers offline for some time, when it will
> > be back online, sync will restart.
> > What you can't have is zpool online on both servers at the same time.
> >
> > So you can do the same with iSCSI disks only.
> > Disks will have to be mounted on one server at a time only.
> >
> > You'll have to have a rather low network / iSCSI latency, or it will fail
> > / zpool will be unusable.
> > If you have high latency, you'll have to go for ZFS send/receive.
> >
> 
> That sounds like an interesting setup, although we have an external iSCSI
> "SAN" so the iSCSI approach would be totally offloaded, without any local
> storage.
> 
> Imagine Server A having a certain LUN on the iSCSI device and Server B
> having another one.
> 
> If a server fails, the idea would be to activate the pool "on the fly" on
> the other one.
> 
> Otherwise I totally take iSCSI out of the picture (less is better), have
> local disks and zfs-snapshot stuff back and forth.

If you can afford to lose x minutes of data then this is the way.
FYI we use zrepl here, with snapshots taken every 5 minutes (sent both
locally to another bare metal server, an offsite to a remote Hetzner
instance over a wireguard link)

> 
> 
> 
> 
> 
> -- 
> *Andrea Brancatelli*

-- 
Julien Cigar
Belgian Biodiversity Platform (http://www.biodiversity.be)
PGP fingerprint: EEF9 F697 4B68 D275 7B11  6A25 B2BB 3710 A204 23C0
No trees were killed in the creation of this message.
However, many electrons were terribly inconvenienced.