HAST + ZFS + NFS + CARP

Wed Aug 10 13:10:53 UTC 2016

On Sat, Jul 02, 2016 at 05:04:22PM +0200, Ben RUBSON wrote:
> 
> > On 30 Jun 2016, at 20:57, Julien Cigar <julien at perdition.city> wrote:
> > 
> > On Thu, Jun 30, 2016 at 11:32:17AM -0500, Chris Watson wrote:
> >> 
> >> 
> >> Sent from my iPhone 5
> >> 
> >>> 
> >>>> 
> >>>> Yes that's another option, so a zpool with two mirrors (local + 
> >>>> exported iSCSI) ?
> >>> 
> >>> Yes, you would then have a real time replication solution (as HAST), compared to ZFS send/receive which is not.
> >>> Depends on what you need :)
> >>> 
> >>>> 
> >>>>> ZFS would then know as soon as a disk is failing.
> >> 
> >> So as an aside, but related, for those watching this from the peanut gallery and for the benefit of the OP perhaps those that run with this setup might give some best practices and tips here in this thread on making this a good reliable setup. I can see someone reading this thread and tossing two crappy Ethernet cards in a box and then complaining it doesn't work well. 
> > 
> > It would be more than welcome indeed..! I have the feeling that HAST
> > isn't that much used (but maybe I am wrong) and it's difficult to find 
> > informations on it's reliability and concrete long-term use cases...
> > 
> > Also the pros vs cons of HAST vs iSCSI
> 
> I made further testing today.
> 
> # serverA, serverB :
> kern.iscsi.ping_timeout=5
> kern.iscsi.iscsid_timeout=5
> kern.iscsi.login_timeout=5
> kern.iscsi.fail_on_disconnection=1
> 
> # Preparation :
> - serverB : let's make 2 iSCSI targets : rem3, rem4.
> - serverB : let's start ctld.
> - serverA : let's create a mirror pool made of 4 disks : loc1, loc2, rem3, rem4.
> - serverA : pool is healthy.
> 
> # Test 1 :
> - serverA : put a lot of data into the pool ;
> - serverB : stop ctld ;
> - serverA : put a lot of data into the pool ;
> - serverB : start ctld ;
> - serverA : make all pool disks online : it works, pool is healthy.
> 
> # Test 2 :
> - serverA : put a lot of data into the pool ;
> - serverA : export the pool ;
> - serverB : import the pool : it does not work, as ctld locks the disks ! Good news, nice protection (both servers won't be able to access the same disks at the same time).
> - serverB : stop ctld ;
> - serverB : import the pool : it works, 2 disks missing ;
> - serverA : let's make 2 iSCSI targets : rem1, rem2 ;
> - serverB : make all pool disks online : it works, pool is healthy.
> 
> # Test 3 :
> - serverA : put a lot of data into the pool ;
> - serverB : stop ctld ;
> - serverA : put a lot of data into the pool ;
> - serverB : import the pool : it works, 2 disks missing ;
> - serverA : let's make 2 iSCSI targets : rem1, rem2 ;
> - serverB : make all pool disks online : it works, pool is healthy, but of course data written at step3 is lost.
> 
> # Test 4 :
> - serverA : put a lot of data into the pool ;
> - serverB : stop ctld ;
> - serverA : put a lot of data into the pool ;
> - serverA : export the pool ;
> - serverA : let's make 2 iSCSI targets : rem1, rem2 ;
> - serverB : import the pool : it works, pool is healthy, data written at step3 is here.
> 
> # Test 5 :
> - serverA : rsync a huge remote repo into the pool in the background ;
> - serverB : stop ctld ;
> - serverA : 2 disks missing, but rsync still runs flawlessly ;
> - serverB : start ctld ;
> - serverA : make all pool disks online : it works, pool is healthy.
> - serverB : ifconfig <replication_interface> down ;
> - serverA : 2 disks missing, but rsync still runs flawlessly ;
> - serverB : ifconfig <replication_interface> up ;
> - serverA : make all pool disks online : it works, pool is healthy.
> - serverB : power reset !
> - serverA : 2 disks missing, but rsync still runs flawlessly ;
> - serverB : let's wait for server to be up ;
> - serverA : make all pool disks online : it works, pool is healthy.
> 
> Quite happy with these tests actually :)

Hello,

So, after testing ZFS replication with zrep (which works more or less
perfectly) I'm busy to experiment a ZFS + iSCSI solution with two small
HP DL20 and 2 disks in each. Machines are partitionned the same 
(https://gist.github.com/silenius/d3fdcd52ab35957f37527af892615ca7) 
with a zfs root
(https://gist.github.com/silenius/f347e90ab187495cdea6e3baf64b881b)

On filer2.prod.lan I have exported the two dedicated partitions
(/dev/da0p4 and /dev/da1p4) as an iSCSI target
(https://gist.github.com/silenius/8efda8334cb16cd779efff027ff5f3bd)
which are available on filer1.prod.lan as /dev/da3 and /dev/da4
(https://gist.github.com/silenius/f6746bc02ae1a5fb7e472e5f5334238b)

Then on filer1.prod.lan I made a zpool mirror over those 4 disks
(https://gist.github.com/silenius/eecd61ad07385e16b41b05e6d2373a9a)

Interfaces are configured as the following:
https://gist.github.com/silenius/4af55df446f82319eaf072049bc9a287 with
"bge1" being the dedicated interface for iSCSI traffic, and "bge0" the
"main" interface through which $clients access the filer (it has a
floating IP 192.168.10.15). (I haven't made any network optimizations 
yet)

Primary results are encouraging too, although I haven't tested under
heavy write yet. I made more or less what Ben did above, trying to
corrupt the pool and ... without success :) I also checked manually
with: $> md5 -qs "$(find -s DIR -type f -print0|xargs -0 md5 -q)" to
check the integrity of the DIR I copied.

I tried also a basic failover scenario with
https://gist.github.com/silenius/b81e577f0f0a37bf7773ef15f7d05b5d which
seems to work atm. 

To avoid a split-brain scenario I think it is also very important that
the pool isn't automatically mounted at boot (so setting cachefile=none)

Comments ? :)

Julien

> 
> Ben
> 
> _______________________________________________
> freebsd-fs at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-fs
> To unsubscribe, send any mail to "freebsd-fs-unsubscribe at freebsd.org"

-- 
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.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20160810/84d611c5/attachment.sig>