As a consultant I work with Ceph using a downstream version of the product; so once in awhile I like to catch up on new features and functions that have not yet hit the downstream/supported version of the product; that process has led me to setting up my homelab (again) and using Ceph Nautilus as a base for storage.
Ceph comes with a deployment and inspection tool called ceph-volume. Much like the older ceph-deploy tool, ceph-volume will allow you to inspect, prepare, and activate object storage daemons (OSDs). The advantages of ceph-volume include support for LVM, dm-cache, and it no longer relies/interacts with udev rules.
For my use case I have installed a single Fusion IOMemory card unto each of my nodes in order to deploy OSDs with faster storage for the DB and WAL devices. It’s a very good idea to read the Bluestore configuration reference as that is default for new OSD deployments. Take careful note of the recommendations for the use of a DB and WAL device.
If there is only a small amount of fast storage available (e.g., less than a gigabyte), we recommend using it as a WAL device. If there is more, provisioning a DB device makes more sense. The BlueStore journal will always be placed on the fastest device available, so using a DB device will provide the same benefit that the WAL device would while also allowing additional metadata to be stored there (if it will fit).Bluestore Configuration Reference
In my case, due to the access to the Fusion IOMemory card, I want to create enough partitions to support 11 OSDs and make them as large as possible for the DB device (which will put the WAL device on the same partition). My fast media is 931 GB of usable storage, if I split it evenly across all eleven OSDs I should end up with partitions ~84 GB in size. I like round numbers so those partitions are now 80 GB in size and the deployment command looks something like this.
root@ganymede:~# ceph-volume lvm prepare --bluestore --dmcrypt --data /dev/sdd --block.db /dev/fioa5
Be sure to replace the –data argument with the storage device and the –block.db argument needs to point to the partition on the fast storage you wish to use for the given OSD. After that I run the activation command for all OSDs on the node.
root@ganymede:~# ceph-volume lvm activate --all
Assuming everything has gone as expected the OSDs will start up and join the cluster and you’ll get all the speedy goodness of an SSD for the write ahead log and RocksDB.