Release Notes #

For some scenarios that do not involve shared storage, relying on cluster quorum together with a hardware watchdog device for fencing is sufficient as a fencing mechanism.

SBD can function in a disk-less mode which relies solely on the hardware watchdog device for detecting and recovering from node failure. This method depends on cluster quorum, and thus needs a minimum of three nodes in the cluster.

When using this fencing method, recovery of a fenced node takes at least as long as the watchdog timeout.

There are some issues in Corosync 2.3.5 which were fixed in later versions. The current Corosync release series is 2.4 which, however, has very different ABI.

In SLE HA 12 SP3, we wanted to avoid a large Corosync ABI change. Therefore, Corosync has been updated to version 2.3.6 and patches from version 2.4.2 have been backported.

In previous versions of SLE HA, if a DRBD resource had lost connection to its peer or was in standalone mode for any other reason, the DRBD resource would still run OK and the status was displayed as green in Hawk.

Hawk now detects and displays a warning when DRBD resources are in unexpected states (disconnected, error, or standalone).

The cluster resource manager has support for configuring multiple fencing devices, to allow backup fencing methods if the original method fails. This can for example be used to combine fencing via shared storage with UPS/PDU fencing.

Hawk now has an interface for creating and configuring fencing topologies. This functionality was previously only available via the command line interface.

Pacemaker can create and edit alerts, user scripts called by the cluster when interesting events occur (nodes joining or leaving, resources starting or stopping, etc.).

On the command line, these alerts can be configured using crm configure alert.

You can now also view and configure alerts in Hawk.

Clusters running in the public cloud need special fencing mechanisms to take advantage of the APIs provided by the cloud provider.

A new fencing agent has been added which relies on the Azure Resource Management (ARM) API. The necessary Azure Resource Management packages are part of the Public Cloud Module for SLES.

A system adminstrator needs tools to verify that the cluster configuration is valid and working properly, and make sure that the cluster nodes have the correct package versions. Previously, the command crm cluster health was available for this purpose.

A new wizard has been added to Hawk which performs the cluster health check and reports the cluster status information directly in the web interface.

Mismatching package versions between live and backup systems are a problem which may only noticed at failover.

A new Hawk wizard checks the software package versions of all cluster nodes, and reports any differences it discovers.

In SLE HA 12 SP3, DRBD has been updated to version 9.0.8. This version includes a number of bug fixes.

The update to ocfs2-tools 1.8.5 fixes many bugs and also brings some enhancements. New functionality includes:

It is now possible to resize the cluster-md RAID1 without stopping the multi-device.

Fail/remove one drive first, increase the size of drive, then add the drive back to the multi-device. After that has been done to all drives of the multi-device, grow the size of the multi-device by running:

mdadm --grow /dev/mdx --size=max

For more information, see https://raid.wiki.kernel.org/index.php/Growing (https://raid.wiki.kernel.org/index.php/Growing).