CEPH on multipathed devices - override default behaviour

[mdwuznik]

Hello,

I'm doing a PoC on nodes with available FC arrays (multiple)
So -> 3 nodes, 3 FC arrays, volumes from each exported to nodes to be used as OSDs


on each node, e.g.:
pveceph osd create /dev/mapper/CEPH1 fails with
unable to get device info for '/dev/dm-44'

That's a known issue (or rather a feature), as described by https://forum.proxmox.com/members/aaron.72961/
in multiple threads

using FC of course adds latency, but going over FC switches is basically non-blocking in this scenario

So - is there a way to override the default behaviour of preventing the use of multipathed devices?

<I'm aware it's an edge case, but still usable...>

 

[alexskysilk]

<I'm aware it's an edge case, but still usable...>

Its not. if you really insist on using this hardware- make all disks passthrough (or single drive raid0, although that can still have issues when mapped as OSDs) and connect each storage device DIRECTLY to a SINGLE HOST. there's no point or benefit in multipathing for your usecase.