Before performing IPMP procedures on a cluster, consider the following guidelines.
When configuring a scalable service resource (SCALABLE=TRUE in the resource type registration file for the resource type) that uses the SUNW.SharedAddress network resource, PNM can be configured to monitor IPMP group status on all IPMP groups on the cluster nodes in addition to the one the SUNW.SharedAddress is configured to use. This configuration allows the service to be restarted and failed over if any of the IPMP groups on the cluster nodes has failed in order to maximize service availability for network clients that are co-located on the same subnets as the cluster nodes. For example:
# echo ssm_monitor_all > /etc/cluster/pnm/pnm.conf
Reboot the node.
Each public network adapter that is used for data-service traffic must belong to an IPMP group. If a public network adapter is not used for data-service traffic, you do not have to configure it in an IPMP group.
The local-mac-address? variable must have a value of true for Ethernet adapters.
You can use probe-based IPMP groups or link-based IPMP groups in a cluster. A probe-based IPMP group tests the target IP address and provides the most protection by recognizing more conditions that might compromise availability.
If you are using iSCSI storage as a quorum device, ensure that the probe-based IPMP device is configured correctly. If the iSCSI network is a private network containing only the cluster nodes and iSCSI storage device and there are no other hosts present on the iSCSI network, then the probe-based IPMP mechanism can break when all but one of the cluster nodes goes down. The problem occurs because there are no other hosts on the iSCSI network for IPMP to probe, so IPMP treats this as a network failure when only one node remains in the cluster. IPMP offlines the iSCSI network adapter, and then the remaining node loses access to the iSCSI storage and thus the quorum device, To resolve this problem, you could add a router to the iSCSI network so that other hosts outside the cluster respond to the probes and prevent IPMP from offlining the network adapter. Alternatively, you could configure IPMP with link-based failover instead of probe-based failover.
Unless there are one or more non-link local IPv6 public network interfaces in the public network configuration, the scinstall utility automatically configures a multiple-adapter IPMP group for each set of public-network adapters in the cluster that uses the same subnet. These groups are link-based with transitive probes. Test addresses can be added if probe-based failure detection is required.
Test IP addresses for all adapters in the same multipathing group must belong to a single IP subnet.
Test IP addresses must not be used by normal applications because they are not highly available.
No restrictions are placed on multipathing group naming. However, when configuring a resource group, the netiflist naming convention is any multipathing name followed by either the nodeID number or the node name. For example, given a multipathing group named sc_ipmp0 , the netiflist naming could be either sc_ipmp0@1 or sc_ipmp0@phys-schost-1, where the adapter is on the node phys-schost-1, which has the nodeID of 1.
Do not unconfigure (unplumb) or bring down an adapter of an IP Network Multipathing group without first switching over the IP addresses from the adapter to be removed to an alternate adapter in the group, using the if_mpadm (1M) command.
Do not unplumb or remove a network interface from the IPMP group where the Oracle Solaris Cluster HA IP address is plumbed. This IP address can belong to the Logical Host resource or the Shared Address resource. However, if you unplumb the active interface using the ifconfig command, Oracle Solaris Cluster now recognizes this event. It fails over the resource group to some other healthy node if the IPMP group has become unusable in the process. Oracle Solaris Cluster could also restart the resource group on the same node if the IPMP group is valid but an HA IP address is missing. The IPMP group becomes unusable for several reasons: loss of IPv4 connectivity, loss of IPv6 connectivity, or both. For more information, see the if_mpadm (1M) man page.
Avoid rewiring adapters to different subnets without first removing them from their respective multipathing groups.
Logical adapter operations can be done on an adapter even if monitoring is on for the multipathing group.
You must maintain at least one public network connection for each node in the cluster. The cluster is inaccessible without a public network connection.
To view the status of IP Network Multipathing groups on a cluster, use the ipmpstat -g command.
For more information about IP Network Multipathing, see the appropriate documentation in the Oracle Solaris OS system administration documentation set.
|
For cluster software installation procedures, see the Oracle Solaris Cluster Software Installation Guide . For procedures about servicing public networking hardware components, see the Oracle Solaris Cluster 4.2 Hardware Administration Manual .