Truenas HA - Thoughts/considerations

[flowdash]

Learning how Truenas scale can be used in a HA fasion and wanted to see the thoughts of others. My plan is to run 1 storage device and then have two head servers that can utilize the same storage device. This is my plan:


In my mind it should work base on this diagram from truenas but wanted to test the waters before buying the storage node and HBAs
Thank you in advance!

 

[CyklonDX]

I dislike it. It relies on SC2000 to actually being HA; but if backplane fails on sc2000, both host fail at the same time.
Sure its better than server but not great...

You can do better with 4+ servers, and actually make it zfs HA, while profiting from any performance uplifts given from all boxes.
1) You set up 2 servers with lets say HA clustering on RHEL using PCS.
2) You create raidzX depending on consistency you want to have on your storage boxes.
3) you create a zfs mirror and using NFS shares over network.
(its described here how-to, i tested this approach and can confirm it works fine.)

ZFS Cluster: A Network-Based ZFS Implementation

I always like to experimenting the idea of building a ZFS cluster, i.e., it has the robust of the ZFS with the cluster capacity. So I came up a test environment with this prototype. Keep in mind th…

icesquare.com

This way either goes offline your main server, or your storage servers the pool will always be accessible As long as 1 rhel ha box, and 1 storage box is alive.

If you plan on using it it in production i would advise 4x10G links *2x10G per each main host)
(depending on your requirements maybe 2x25G links)

Working in LACP mode; 802.3ad or different balancing modes available for your network.

 

[ano]

Learning how Truenas scale can be used in a HA fasion and wanted to see the thoughts of others. My plan is to run 1 storage device and then have two head servers that can utilize the same storage device. This is my plan:
View attachment 28307

In my mind it should work base on this diagram from truenas but wanted to test the waters before buying the storage node and HBAs
View attachment 28308Thank you in advance!

we do this with supermicro & osnexus quantastor, it works out great, even do 3 external enclosures, so we can have enclosure redundancy for 6 drives per z2

also enables you do run active/active and have 2 pools, 1 mounted on each system for more IOPS, each host has 2x100gbps in LACP, and I can do the full 400gbps

jovian does something similar to what @CyklonDX describes, and its been quite horrible for production

 

[CyklonDX]

jovian does something similar to what @CyklonDX describes, and its been quite horrible for production

I do see potential issues with failover server link to storage boxes; but could be solved if storage boxes connecting 2nd nic (to alt box) run on separate subnet so routing doesn't get rekt; this way both hosts storage boxes wouldn't be aware of each other and could switch easily without problem; also a UI would be useful to clear zpool bad/down state from HA box, so it reconnects/resilvers the pool. Potential network issues, like dropped packets could cause a lot of headache - might have to readjust tcp retransmission rate to do all attempts under a sec; so if your network ain't great for whatever the reason you may be either going down every now and then, or create excess of traffic trying to accomadate crappy network.

Obviously that would need to be tried, and tested. But i don't think it would cause any issues once setup is completed, and tested out. It not something i would want to use for Windows; only KVM/oVirt.

 

[ericloewe]

Learning how Truenas scale can be used in a HA fasion

It cannot, at least not on hardware not supplied by iX. And it does not work with separate systems, rather two integrated hosts in the same chassis.

So your proposal is a complete non-starter.