Cloud Storage with Red Hat Ceph Storage

RBD snapshots are read-only copies of an RBD image created at a particular time. RBD snapshots use a COW technique to reduce the amount of storage needed to maintain snapshots. Before applying a write I/O request to an RBD snapshot image, the cluster copies the original data to another area in the placement group of the object affected by the I/O operation. Snapshots do not consume any storage space when created, but grow in size as the objects that they contain change. RBD images support incremental snapshots.

Figure 6.4: RBD snapshots creation

The snapshot COW procedure operates at the object level, regardless of the size of the write I/O request made to the RBD image. If you write a single byte to an RBD image that has a snapshot, then Ceph copies the entire affected object from the RBD image into the snapshot area.

Figure 6.5: Writing to an RBD snapshot

Deleting an RBD image fails if snapshots exist for the image. Use the rbd snap purge command to delete the snapshots.

Use the rbd snap create command to create a snapshot of a Ceph block device.

[root@node ~]# rbd snap create pool/image@firstsnap
Creating snap: 100% complete...done.

Use the rbd snap ls command to list the block device snapshots.

[root@node ~]# rbd snap ls pool/image
SNAPID  NAME        SIZE     PROTECTED  TIMESTAMP
     4  firstsnap   128 MiB             Thu Oct 14 16:55:01 2021
     7  secondsnap  128 MiB             Thu Oct 14 17:48:59 2021

Use the rbd snap rollback command to roll back a block device snapshot, overwriting the current version of the image with data from the snapshot.

[root@node ~]# rbd snap rollback pool/image@firstsnap
SNAPID  NAME        SIZE     PROTECTED  TIMESTAMP
     4  firstsnap   128 MiB             Thu Oct 14 16:55:01 2021
     7  secondsnap  128 MiB             Thu Oct 14 17:48:59 2021

Use the rbd snap rm command to delete a snapshot for Ceph block devices.

[root@node ~]# rbd snap rm pool/image@secondsnap
Removing snap: 100% complete...done.

RBD clones are read/write copies of an RBD image that use a protected RBD snapshot as a base. An RBD clone can also be flattened, which converts it into an RBD image independent of its source. The cloning process has three steps:

The newly created clone behaves just like a regular RBD image. Clones support COW and COR, with COW as the default. COW copies the parent snapshot data into the clone before applying a write I/O request to the clone.

Figure 6.6: RBD clone write operation

You can also enable COR support for RBD clones. Data that is the same for the parent RBD snapshot and the clone is read directly from the parent. This can make reads more expensive if the parent's OSDs have high latency relative to the client. COR copies objects to the clone when they are first read.

If you enable COR, Ceph copies the data from the parent snapshot into the clone before processing a read I/O request, if the data is not already present in the clone. Activate the COR feature by running the ceph config set client rbd_clone_copy_on_read true command or the ceph config set global rbd_clone_copy_on_read true command for the client or global setting. The original data is not overwritten.

Figure 6.7: RBD clone read operation

The clone COW and COR procedures operate at the object level, regardless of the I/O request size. To read or write a single byte of the RBD clone, Ceph copies the entire object from the parent image or snapshot into the clone.

Use the rbd command to manage RBD clones.

When flattening a clone, Ceph copies all missing data from the parent into the clone and then removes the reference to the parent. The clone becomes an independent RBD image and is no longer the child of a protected snapshot.