I have 20 OSDs (HDDs) in my Ceph cluster and want to allocate 15 OSDs to one pool and 5 OSDs to another pool without causing downtime, as this is a production environment. Can someone provide guidance on how to achieve this safely and efficiently, considering my limited experience with Ceph?
Hello, I would recommend testing this in a virtual pve enviroment first! At first, you need to assign a different device class to the 5 different osds. like hdd-2 for example. Then create two custom crush rules for the setup, as the default crush-rule takes all hosts/osds in consideration. After creating two new crush-rules assign them to the pool(s).
- remove the old device-class from the 5 osds:
ceph osd crush rm-device-class osd.x - set new device class to the 5 osds:
ceph osd crush set-device-class hdd-2 osd.x - show current crush-rules:
ceph osd crush rule ls - create a new crush-rule:
ceph osd crush rule create-replicated replicated-hdd default host hdd - create the second crush-rule:
ceph osd crush rule create-replicated replicated-hdd-2 default host hdd-2 - recheck crush-rules:
ceph osd crush rule list - set crush-rool to pool:
ceph osd pool set POOL-NAME crush_rule replicated-hdd - set crush-rool to pool-2:
ceph osd pool set POOL2-NAME crush_rule replicated-hdd-2