Clustering feature in FreeNAS

[Saira]

Dear All,

How can i enable clustering feature in freenas.I mean i want to develop a plugin and load in freenas so that we can enjoy clustering feature .
Can anyone help me please.

 

[cyberjock]

There are none. ZFS is not designed for nor has provisions for clustering.

 

[jgreco]

Well, not exactly true, but it is certainly not a beginner level project, and if you need to ask, you're in far over your head.

 

[Saira]

Thanks Jgreco,
I know its not a beginner level project , actually its my Master Thesis on this topic therefore i must have to write a code etc for high availability in freenas.
Any help will be appreciated.

Thanks

 

[mjws00]

Why pick an appliance OS that specifically locks you out of the internal workings and low levels of the OS. If you need a tool for a thesis the underlying OS would be a far more suitable vehicle, IMHO. Even a product like napp-it where a GUI is added on top of a platform would be better. It is hard to justify picking a product that specifically intends to limit tinkering with the underlying system and attempts to abstract low level functions. Doesn't seem like a master's level choice.

That said. Fork it on github and go nuts. Not sure how the community is supposed to help. But I would want to look at raw Linux, or FreeBSD where there is precedent and access to internals, and a community already working on these functions.

 

[anodos]

There's been some work on running ceph on zfs.

https://wiki.ceph.com/Planning/Blueprints/Emperor/osd:_ceph_on_zfs

Of course, as is mentioned above, look into vanilla freebsd or one of the linuxes. These are not the droids you're looking for.

 

[jgreco]

There's been some work on running ceph on zfs.

https://wiki.ceph.com/Planning/Blueprints/Emperor/osd:_ceph_on_zfs

Of course, as is mentioned above, look into vanilla freebsd or one of the linuxes. These are not the droids you're looking for.

I disagree. It is perfectly possible to look at doing this sort of thing, but it isn't a beginner level project. Nexenta implements some of this using RSF-1 (from high-availability.com), and you can get some idea of the scope of this:

http://www.high-availability.com/wp-content/uploads/2013/01/RSF-1-HA-PLUGIN-ZFS-STORAGE-CLUSTER.pdf

Thanks Jgreco,
I know its not a beginner level project , actually its my Master Thesis on this topic therefore i must have to write a code etc for high availability in freenas.
Any help will be appreciated.

Thanks

So there's a few possible ways to go about this "and they're all ugly." The RSF-1 strategy provides multiple head single pool (read: SPOF) access. A strategy based on HAST could provide a solution that didn't involve a shared pool. Etc.

 

[Kewlmyst]

jgreco:
The RSF-1 strategy provides multiple head single pool (read: SPOF) access.

Multiple head, single pool has advantages ... amongst others:
Non Disruptive Upgrades, It's not always just a question of preventing failures.

Regards, Andre

 

[jgreco]

Non Disruptive Upgrades, It's not always just a question of preventing failures.

Of course it's a disruptive upgrade. It's just a matter of the specifics. A true non-disruptive upgrade is when another head unit is able to take over without any interruption. For a ZFS filer, that isn't possible. One filer has to disable the service IP alias, deactivate the services, export the pool, signal the secondary filer, which then imports the pool, activates the services, and acquires the service IP alias. This is disruptive. You have to re-establish client TCP sessions for NFS or iSCSI. It may be a minimally disruptive failover, which is good, great even, but let's be honest about what it is and not drink the marketing kool-aid.

And a strategy based on HAST can have that sort of failover occur too with the exact same disruption profile, so I guess I don't even really understand the point you're trying to make.

 

[Kewlmyst]

I was just commenting on the fact that multiple heads and single pool can be a valid choice even if it remains a spof,
one advantage I mentioned was ndu, which indeed is not achieved with ZFS+HAST, but does exist with other types
of architectures outside the Freenas world.
But you are right, this is the Freenas forum, discussions should stay related to Freenas; my mistake.

 

[MtK]

Of course it's a disruptive upgrade. It's just a matter of the specifics. A true non-disruptive upgrade is when another head unit is able to take over without any interruption. For a ZFS filer, that isn't possible. One filer has to disable the service IP alias, deactivate the services, export the pool, signal the secondary filer, which then imports the pool, activates the services, and acquires the service IP alias. This is disruptive. You have to re-establish client TCP sessions for NFS or iSCSI. It may be a minimally disruptive failover, which is good, great even, but let's be honest about what it is and not drink the marketing kool-aid.

This is achievable with Supermicro's SBB...

 

[jgreco]

This is achievable with Supermicro's SBB...

I was under the impression that the SBB was basically two nodes and shared storage. Am I wrong?

 

[MtK]

You are right.
But with the proper scripts what you suggested is achievable...

 

[jgreco]

You are right.
But with the proper scripts what you suggested is achievable...

The sequence of cutover steps I suggested is of course achievable. I hope you don't mean that a true non-disruptive upgrade is possible, because it isn't. No matter how good the scripts, there's some lag time to implement all of the details and get it right, and operations screech to a halt even if just for a few seconds. That might not be hurtful to a home user, but in an enterprise environment that can be a big hit.

 

[MtK]

Not implying any performance achievement, just the feasibility...

 

[TurkMcdirt]

Here are some attempts at doing what jgreco is suggesting using heatbeat+pacemaker, but like he said there is a lag (40sec!) during failover

http://zfs-create.blogspot.com/2013/06/building-zfs-storage-appliance-part-1.html

https://bm-stor.com/index.php/blog/Linux-cluster-with-ZFS-on-Cluster-in-a-Box/

 

[jgreco]

40 seconds is a bit much, but there is inevitably some lag in these sorts of schemes.

 

[BasitKhan]

Hi Saira,
I think below mentioned links can help you in achieving your goal.

http://qq929962616.72pines.com/2012/03/configure-highly-available-storage-on-freenas/
http://www.karl.idv.hk/?p=583
https://www.synology.com/en-global/knowledgebase/tutorials/585?lang=enu&q_id=585
https://forums.freenas.org/index.php?threads/high-availability-freenas-setup.15867/
https://forums.freebsd.org/threads/hast-and-zfs-with-carp-failover.29639/
http://www.aisecure.net/2012/02/07/hast-freebsd-zfs-with-carp-failover/