Migration from RAID to Ceph without outage

[SudsMaker]

OK, so I’m fairly new to Proxmox and Ceph.
I need some advice on how to migrate to Ceph.

Current Layout:
SERVER01: (Same exact hardware as SERVER02)

• Production - hosting 20-30 VM’s
• Storage Hardware RAID
  • 2 - SSD’s setup as RAID1
  • 6 - SSD’s setup as RAID5
  • This storage is hosting the VM’s
  • Needs to be rebuild with No RAID (so it is like SERVER02)

SERVER02: (Same exact hardware as SERVER01)

• Not in use yet
• Storage Hardware
  • 8 - SSD’s
  • All independent (No RAID)

SERVER03: (Older hardware but same storage capacity)

• Not in use yet
• Storage Hardware
  • 8 - SSD’s
  • All independent (No RAID)

Questions:
I have been told to join SERVER02 & SERVER03 to the same cluster hosted by SERVER01 and migrate but I can confused by that.

• If I put SERVER02 & SERVER03 in the same cluster how can I migrate the VM’s off of SERVER01 to SERVER02 without an outage?
• SERVER01 will need to be rebuild without RAID just like SERVER02
• Since SERVER01 is in a RAID5 and running VM’s I can’t create anything Ceph I would assume, as there is nothing to assign to an OSD.
• I guess my biggest challenge is how to migrate all the VM’s off a server setup as a RAID to a server that is setup with Ceph without an outage for the VM’s.

Thank you for your input in advance.

 

[Folke]

Hi,

If I put SERVER02 & SERVER03 in the same cluster how can I migrate the VM’s off of SERVER01 to SERVER02 without an outage?

When your node is part of a cluster, you will have a migrate option next to the shutdown button of a vm. This can do a so-called live migration, which means that the current disk and ram contents will be transferred to the new node in the background and when done, the execution will just be continued on the second node, without any downtime.

Since SERVER01 is in a RAID5 and running VM’s I can’t create anything Ceph I would assume, as there is nothing to assign to an OSD.

You can set up Ceph on the second and third node first and migrate the VM's to it, before restructuring your first nodes' storage.

 

[SudsMaker]

Excellent, thank you!