High Availability

Cluster Manager

Apart from the DRBD software for replicating data between two nodes, a cluster manager is needed to create a working high availability cluster. The most prominent one is Pacemaker.


DRBD ships with seamless integration to Pacemaker. Allowing Pacemaker to start, stop, promote and demote DRBD. With the fence-peer mechanism, DRBD utilizes Pacemaker's CIB to help avoid split-brain situations.


Last, but not least, it should be noted that using the Pacemaker cluster manager is not a requirement. Starting with version 9.0, DRBD has an auto-promote feature. That allows you arbitrarily use other cluster managers that are able to mount a file system on a shared storage device.

How DRBD recovers from outages

After a node outage

After an outage of a node, DRBD® automatically resynchronizes any out of date data. It does this in the background without any intervention required, or interfering with the services running on top of it.

After an outage of the replication network

Restoring service after the temporary failure of the replication network is a comon example of how the automatic recovery mechanism just described works. DRBD will reestablish the connection and do the necessary resynchronization automatically.

After an outage of a storage subsystem

DRBD can mask the failure of a disk on the active node, i.e., the service can continue to run there, without needing to migrate the service. If the disk can be replaced without shutting down the machine, it can be reattached to DRBD. DRBD resynchronizes the data as needed to the replacement disk.

After an outage of all network links

Split brain is a situation where, due to the (temporary) failure of all network links between cluster nodes, both nodes switched to the primary role while disconnected. DRBD supports various automatic and manual recovery options in the event of split brain.

More Features

DRBD is OpenSource

The kernel driver is part of the upstream kernel as of 2009

Consistency groups

DRBD obeyes write fidelity for volumes with potentially different characteristics

Diskless operation

Continue operation after disk failure. DRBD can, transparently, use other nodes in the cluster as NAS

Different Transports

DRBD supports TCP/IP by default and RDMA available as a high perfomance alternative

Use Cases


MySQL, PostgreSQL, ...

File servers

NFS, CIFS, ...

Storage targets

iSCSI, FC, ...

Custom applications


For the SysAdmin

For each service you will have a dedicated DRBD resource. That gives you a block device (/dev/drbdX).

On top of that will be a file system (e.g. XFS or ext4). This file-system gets mounted by Pacemaker (e.g. to /mnt/ha-mysql).

Then, you configure your service (e.g. mysql instance) to put its data there.

Finally, you hand over starting and stopping of the service to Pacemaker.

Last updated: Fri Feb 19 11:06:35 2016 Last change: Thu Feb 18 11:27:24 2016 by root via cibadmin on node1
Stack: heartbeat
Current DC: node1 (version 1.1.14-8d93feb) - partition with quorum
2 nodes and 6 resources configured

Node node1 online
p_filesystem_test (ocf::heartbeat:Filesystem): Started
p_ip_test (ocf::heartbeat:IPaddr2): Started
Node node2 online

Inactive resources:

Migration Summary:
* Node node1:
* Node node2:

DRBD Road Map

OpenNebula Driver

OpenNebula Driver

Enables OpenNebula users to base clouds on DRBD9/DRBD Manage

Status: Release Candidate 1

Microsoft Windows Support

Microsoft Windows Support

DRBD for MS Windows

Release planned with Version 9.0.2

PeerDirect RDMA

PeerDirect RDMA

PeerDirect allows a write request to be sent from an InfiniBand HCA directly to an NVMe device

Planning stage

DRBD on Power

DRBD on Power

Support Power8, enabling access to LINBIT's products

Released – with IBM in Germany – November 12

OpenStack Driver

OpenStack Driver

Enables OpenStack users to base their clouds on DRBD9/DRBD Manage

Released - Tokyo Open Stack Summit - October 16

RDMA / InfiniBand

RDMA / InfiniBand

Multi-path support: Aggregates bandwidth of configured paths; increases replication link availability

Release planned for November 2016