New cluster - step by step

[Alessandro 123]

As I'm still planning my new cluster (yes, the same cluster that i'm planning from this spring) I would like to create this step by step, let me try to explain.

Currently I have just 1 server with hardware RAID.
I'll remove the hardware raid card and I replace that with a simple HBA, by using ZFS (mirror? RAIDZ2?)

This server *must* be part of a new cluster (that i'll create by adding new nodes in next months) with a shared storage (currently unavailable)

Which is the best workflow to archieve this without downtime (or with minimal downtime) ?
Can I start with a single server with local disks, create a new cluster with the single server and when the storage is ready, move the VM to the shared storage ?

The move operation would be made live or I have to shutdown the VM ?

This is what I though:

1) spin up the new server with local ZFS storage
2) create a new cluster with only this server
3) spin up a second server with local ZFS storage
3) add the new server to the same cluster

(with this, i'll be able to live migrate VM without shared storage with the upgoing live-migration feature)

4) sping up a third server with local ZFS
5) add the third server to the cluster
6) add a shared storage to the cluster (should I do this on each machine?)
7) migrate all VMs to the shared storage

and now ? How can I enable HA and automatic failover for all VMs ?

 

[Alessandro 123]

One question: my current server doesn't have a slot for SSD and all HDD slots would be used for SAS 15K spinning disks.
So, I can't put the ZIL to an SSD, this will lead to worse performance than hardware raid with cache, right ? Any workaround ?

EDIT: I can try to use bigger SAS disks to free up 2 slots. In these 2 slots I can put 2 SSD for ZIL and L2ARC. Should I buy the fastest SSD (so >= 200GB) or even the smaller (and slower) one like 100GB is enough to be used as ZIL ? I'm talking about the Intel DC3610. 200GB are wasted for ZIL and L2ARC as ZFS would never use that amount of space.

 

[tschanness]

Hi,
for no Downtime wait for your shared Storage. When it's there add a Cluster with a second host, migrate all VMs to the shared storage, migrate all VMs to the new host and then reinstall the old server.
Remember: Don't use the same hostname in the cluster twice (e.g. give the newly installed Host another hostname then..

sC

 

[Alessandro 123]

So, are you suggesting the following:

1) spin up the new server with local ZFS storage
2) spin up the second server with local ZFS storage
3) create a cluster
4) add a shared storage to the cluster
5) migrate all VM to the shared storage
6) migrate all VM to the second host
7) reinstall the first host (I can use the same hostname used initially, as it wasn't part of any cluster)
8) rebalance VM acress the 2 hosts
9) add the third node

But why my initial solutions wasn't ok ?

 

[tschanness]

Leave #1 then: yes.
Keep your existing host as is, this will keep your Downtime as low as possible.

Your initial solution also said to reinstall the first host before clustering, that's why I made my suggestion.

As for SSDs: The "slower" 100gb should be enough.

sC

 

[Alessandro 123]

I've not wrote to reinstall the first host after clustering.
I would like to add the clustering from scratch, directly on the first host without the second node. Is this possible ?