Enforcing a size into cephfs.

[Gilberto Ferreira]

Hello there
I have 3 proxmox server with Ceph (quincy).
Each server I have 2 SSD, 1TB each.
In total, OSD's are apr. 5 TB.
So, I wonder if I can create a CephFS with, let's say, 50G, and use of the rest to regular pool.
Thanks for any help

 

[Gilberto Ferreira]

UPDATE
So I ended up created to partitions in /dev/sdc
/dev/sdc1 with 800G
/dev/sdc2 with 100G
Then I create two OSD using /dev/sdc1 and /dev/sdc2, but osd using /dev/sdc2 I marked as HDD.
After that I create a crush_rule:
ceph osd crush rule create-replicated HDD_REP default host hdd
After that I insert the cephfs data and metadata into HDD_REP crush_rule:
ceph osd pool set cephfs_data crush_rule HDD_REP
ceph osd pool set cephfs_metadata crush_rule HDD_REP
And then I end up with a chepfs named templates with 100G.
Toughts?

 

[alexskysilk]

thats what quotas are for.
https://docs.ceph.com/en/latest/rados/operations/pools/#set-pool-quotas

 

[Gilberto Ferreira]

My main question is: if I mark the OSD as being HDD class, would that be reason to have some performance loss? (in long term!)
Or even for using partitions in the OSD build instead of the whole disk.

 

[alexskysilk]

There are two consequences to osd object type definition-

1. It will make the OSDs available or not available for your pool based on its crush rules
2. it will apply predefined tunables, which could affect performance.