[SOLVED] ceph class issue with nautilus nvme p3700

[RobFantini]

I have a class set up for nvme.

that worked fine with pve5 / luminous for our intel nvme p3700's.

now when i add a nvme :

1- pve screen give warning about raid controller: Note: Ceph is not compatible with disks backed by a hardware RAID controller. For details see the reference documentation.
I forget if it did so for ssd's connected to our IT mode LSI HBA's

2- the nvme gets added to ssd class.

any suggestions?

 

[RobFantini]

here is part of tree:

more info

# ceph device ls|grep  nvme
INTEL_SSDPEDMD020T4D_HHHL_NVMe_2000GB_CVFT5190000Q2P0EGN pve14:nvme0n1 osd.4                 
INTEL_SSDPEDMD020T4D_HHHL_NVMe_2000GB_CVFT735300072P0OGN pve10:nvme0n1 osd.1                 
INTEL_SSDPEDMD020T4D_HHHL_NVMe_2000GB_CVFT7353000B2P0OGN pve3:nvme0n1  osd.2                 
INTEL_SSDPEDMD020T4D_HHHL_NVMe_2000GB_CVFT735300122P0OGN pve2:nvme0n1  osd.0                 
INTEL_SSDPEDMD020T4D_HHHL_NVMe_2000GB_CVFT735300282P0OGN pve9:nvme0n1  osd.6                 
INTEL_SSDPEDMD020T4_CVFT629000L92P0EGN                   sys8:nvme0n1  osd.3                 
INTEL_SSDPEDMD020T4_PHFT6201004W2P0EGN                   pve15:nvme0n1 osd.5

 

[RobFantini]

solved with info from https://access.redhat.com/solutions/3341491

## Initial members in nvme class
# ceph osd tree|grep nvme
  1  nvme  1.81929         osd.1      up  1.00000 1.00000
  4  nvme  1.81929         osd.4      up  1.00000 1.00000


 # ceph osd crush set-device-class nvme osd.0
Error EBUSY: osd.0 has already bound to class 'ssd', can not reset class to 'nvme'; use 'ceph osd crush rm-device-class <id>' to remove old class first

# ceph osd crush rm-device-class osd.0
done removing class of osd(s): 0

# ceph osd crush set-device-class nvme osd.0
set osd(s) 0 to class 'nvme'

## Result
# ceph osd tree|grep nvme
  1  nvme  1.81929         osd.1      up  1.00000 1.00000
  4  nvme  1.81929         osd.4      up  1.00000 1.00000
  0  nvme  1.81929         osd.0      up  1.00000 1.00000

after that ceph has some backfilling to do. so naturally follow the one thing at a time - let ceph -s show no backfills and do the next changes.