Ceph » RADOS

Brad Fitzpatrick makes the following request to Zac Dover in private correspondence on 05 Jan 2023:

"I'm kinda disappointed in CRUSH's inability to express the policy that I and a bunch of the homelab small-ceph-cluster folk want: 1 copy on SSD/NVME, and 2 copies on HDD but on distinct hosts from the SSD/NVME. The CRUSH algorithm almost let you do it, but not enough. It's like there's an opcode or two missing to manage the stack a bit more. I want to take 3 hosts, then work with each in order, emiting for each, without clearing the whole stack per each host.

"The Ceph community's answer is always "eh, just buy more hosts" or "just buy more disks" or "just store more copies". None of those are very satisfying of an answer for homelab folks that find it hard to justify spending money/electricity on something that's kinda just for fun. (And maybe they're getting into Ceph at home to then bring to work later! I'm kinda in that boat, debating whether I want to sell my Tailscale coworkers on libcephsqlite which I absolutely love personally)"