Ceph Block Device mirroring

Edit online

RADOS Block Device (RBD) mirroring is a process of asynchronous replication of Ceph Block Device images between two or more Ceph storage clusters. By locating a Ceph storage cluster in different geographic locations, RBD Mirroring can help you recover from a site disaster. Journal-based Ceph block device mirroring ensures point-in-time consistent replicas of all changes to an image, including reads and writes, block device resizing, snapshots, clones and flattening.

RBD mirroring uses exclusive locks and the journaling feature to record all modifications to an image in the order in which they occur. This ensures that a crash-consistent mirror of an image is available.
Important: The CRUSH hierarchies supporting primary and secondary pools that mirror block device images must have the same capacity and performance characteristics, and must have adequate bandwidth to ensure mirroring without excess latency. For example, if you have X MB/s average write throughput to images in the primary storage cluster, the network must support N * X throughput in the network connection to the secondary site plus a safety factor of Y% to mirror N images.

The rbd-mirror daemon is responsible for synchronizing images from one Ceph storage cluster to another Ceph storage cluster by pulling changes from the remote primary image and writes those changes to the local, non-primary image. The rbd-mirror daemon can run either on a single Ceph storage cluster for one-way mirroring or on two Ceph storage clusters for two-way mirroring that participate in the mirroring relationship.

For RBD mirroring to work, either using one-way or two-way replication, a couple of assumptions are made:

  • A pool with the same name exists on both storage clusters.

  • A pool contains journal-enabled images you want to mirror.

Important: In one-way or two-way replication, each instance of rbd-mirror must be able to connect to the other Ceph storage cluster simultaneously. Additionally, the network must have sufficient bandwidth between the two data center sites to handle mirroring.

One-way replication

One-way mirroring implies that a primary image or pool of images in one storage cluster gets replicated to a secondary storage cluster. One-way mirroring also supports replicating to multiple secondary storage clusters.

On the secondary storage cluster, the image is the non-primary replicate; that is, Ceph clients cannot write to the image. When data is mirrored from a primary storage cluster to a secondary storage cluster, the rbd-mirror runs ONLY on the secondary storage cluster.

For one-way mirroring to work, a couple of assumptions are made:

  • You have two Ceph storage clusters and you want to replicate images from a primary storage cluster to a secondary storage cluster.

  • The secondary storage cluster has a Ceph client node attached to it running the rbd-mirror daemon. The rbd-mirror daemon will connect to the primary storage cluster to sync images to the secondary storage cluster.

Figure 1. One-way mirroring
One-way mirroring

Two-way replication

Two-way replication adds an rbd-mirror daemon on the primary cluster so images can be demoted on it and promoted on the secondary cluster. Changes can then be made to the images on the secondary cluster and they will be replicated in the reverse direction, from secondary to primary. Both clusters must have rbd-mirror running to allow promoting and demoting images on either cluster. Currently, two-way replication is only supported between two sites.

For two-way mirroring to work, a couple of assumptions are made:

  • You have two storage clusters and you want to be able to replicate images between them in either direction.

  • Both storage clusters have a client node attached to them running the rbd-mirror daemon. The rbd-mirror daemon running on the secondary storage cluster will connect to the primary storage cluster to synchronize images to secondary, and the rbd-mirror daemon running on the primary storage cluster will connect to the secondary storage cluster to synchronize images to primary.

Figure 2. Two-way mirroring
Two-way mirroring

Mirroring modes

Mirroring is configured on a per-pool basis with mirror peering storage clusters. Ceph supports two mirroring modes, depending on the type of images in the pool.

Pool Mode
All images in a pool with the journaling feature enabled are mirrored.

Image Mode
Only a specific subset of images within a pool are mirrored. You must enable mirroring for each image separately.

Image states

Whether or not an image can be modified depends on its state:

  • Images in the primary state can be modified.

  • Images in the non-primary state cannot be modified.

Images are automatically promoted to primary when mirroring is first enabled on an image. The promotion can happen:

  • Implicitly by enabling mirroring in pool mode.

  • Explicitly by enabling mirroring of a specific image.

It is possible to demote primary images and promote non-primary images.