- Red Hat Enterprise Linux (RHEL) with the Resilient Storage Add-On
Useful References and Guides
This guide lays out Red Hat's policies applicable to the use of the Distributed Lock Manager -
dlm - in RHEL Resilient Storage deployments. Users of
dlm in RHEL Resilient Storage clusters should adhere to these policies in order to be eligible for assistance from Red Hat Support with the appropriate product support subscriptions.
Maximum number of nodes supported by DLM: The maximum number of nodes when using DLM is 16 nodes as stated in: What is the supported number of nodes for a RHEL 7 or 8 Resilient Storage cluster?
dlm unsupported with RRP / SCTP in RHEL 6 or 7: Red Hat does not support the usage of
dlm-using components when
dlm is configured to use SCTP communications - also known as "multi-homing" - or when the cluster is configured to use Redundant Ring Protocol (RRP). Such configurations are likely to behave in unexpected ways.
rgmanagerin RHEL 6 must use
dlmin clusters where RRP is configured in order to receive support from Red Hat. RHEL 6 with
rgmanagerwill automatically enable this alternate lock manager -
cpglockd- when the cluster is configured for RRP.
Resilient Storage components that rely on
dlmare not supported by Red Hat in RRP configurations on RHEL 6 or RHEL 7. Neither
cmirrordis supported in clusters configured for RRP.
dlm only supports clusters with one interface in RHEL 8: Red Hat does not support the usage of
dlm-using components with more than one ring configured in a RHEL 8 cluster using the
knet transport. Such configurations are likely to behave in unexpected ways.
Any Resilient Storage cluster that uses
gfs2 must be configured with only a single ring in the cluster in order to receive support from Red Hat. Red Hat recommends utilizing bonding or teaming for the interfaces in that single cluster ring as the method to achieve redundancy.
Fencing required: Usage of
dlm in a cluster requires that fencing be configured with a proper storage or power device, or
sbd. Red Hat does not support clusters actively using
dlm that do not have fencing configured and a proper
STONITH/fencing method configured for each node.
- See also: Support policies: General requirements for fencing/STONITH
- In the above fencing/STONITH policies guide, an allowance is made for clusters without shared storage to use "alternative" fence methods. This is not allowed for clusters utilizing
dlmfor any component - even if shared storage is not involved.
dlm-using clusters must have a STONITH method that will power a node off or cut off a node from shared resources when there is a problem.
RHEL 7 and 8: Usage of
dlm.service: Red Hat does not support starting or managing
dlm in RHEL 7 or 8 via the
dlm.service unit that is provided by the
dlm package. The
dlm_controld daemon should be started and managed using an
ocf:pacemaker:controld resource within the cluster.
dlm not supported on
pacemaker-remote "remote-nodes": When a cluster contains remote-nodes offered via
dlm is not supported by Red Hat to be used on any such remote-node. Any
controld resource and anything dependent on it that uses
dlm should be constrained to never run on any remote-node.