Search <book_title>...

Storage Foundation Cluster File System High Availability 7.2 Administrator's Guide - Solaris

Last Published: 2018-03-06

Product(s): InfoScale & Storage Foundation (7.2)

Platform: Solaris

Example of setting the cluster node preference value for master failover

This example describes a case where the cluster environment includes three types of nodes. You can define master preferences for each type of node, so that Cluster Volume Manager (CVM) fails over the master role to the best candidate.

Consider the three types of nodes as the following pools:

Pool1 (Node1, Node2, Node3)
These nodes have high capacity (storage and processing) and have full storage visibility. An I/O intensive application is running at these nodes of the cluster.
You are OK to choose a node from this pool over nodes from the other pool even if it loses 30% of its disks.
Pool2 (Node4, Node5)
The cluster has a few low capacity (storage and processing) nodes. Internal (in-house) applications use these nodes to do post-processing over the data. Some of these nodes are also used for snapshots and backups. You want to choose a node from this pool:
If there are no nodes to choose from Pool1 OR
If all the nodes in Pool1 have lost access to substantial disks reducing their preference values below the preference values of nodes from Pool2.
Pool3 (Node6, Node7)
These nodes run applications that do not need to work on all the volumes. These nodes may have a restricted view to the storage (limited storage visibility). CVM internally offsets the preference values of nodes in this pool, because they do not see all of the storage. To reduce the likelihood that one of these nodes becomes the master node, you can assign negative preference values.

If you do not define any custom preferences, CVM determines the preference as an offset depending on disk availability. Suppose at time A, the current master, Node1, leaves the cluster. CVM calculates the preferences, and chooses Node5 since it has the best connectivity. If Node5 leaves, CVM selects Node4. Nodes from Pool2 are preferred over nodes in Pool1.

Pool	Node	CVM offset at time A
Pool1	Node1 (Current master)	0
Pool1	Node2	-30
Pool1	Node3	-25
Pool2	Node4.	-20
Pool2	Node5	0
Pool3	Node6	-50
Pool3	Node7	-50

In this example, you want to specify higher preference values to nodes in Pool1 and lower preferences to the nodes in Pool3. The following diagram shows possible preference values to the nodes in the cluster:

To set the cluster node preference

For each node in Pool1, set the preference to 30.

# hares -modify cvm_clus CVMNodePreference \
"node1=30, node2=30, node3=30"

For each node in Pool3, set the preference to -50.

# hares -modify cvm_clus CVMNodePreference "node6=-50, node7=-50"

After you set the preference values as above, the behavior reflects the desired behavior in case of failover. If Node1 fails, the other nodes in Pool1 are the most likely candidates to take over the master. Node3 has lost 25 percent of its disks, but it is still preferred over the nodes in the other pools. Although Node5 has good connectivity, the nodes in Pool1 are preferred over Node5.

Node	CVM offset at time A	Custom preference	Total preference
Node1 (current master)	0	30	30
Node2	-29	30	1
Node3	-25	30	5
Node4.	-20	0	-20
Node5	0	0	0
Node6	-50	-50	-100
.Node7	-50	-50	-100