Clustering, in general, requires that all cluster nodes
in the recovery domain of an application cluster resource group reside
on the same LAN (use the same subnet addressing). Cluster resource
services supports a user configured takeover IP address when configuring
application CRGs.
Address Resolution
Protocol (ARP) is the network protocol that is used to switch
the configured application takeover IP address from one node to another
node in the recovery domain. To enable the application switchover
across subnets, you need to use the virtual IP address support and
the Routing Information Protocol (RIP) for IPv4.
The following
manual configuration steps are required to enable the switchover environment.
This set of instructions must be done on all the nodes in the
recovery domain, and repeated for the other nodes in the cluster that
will become nodes in the recovery domain for the given application
CRG.
- Select an IPv4 takeover IP address to be
used by the application CRG.
- To avoid confusion, this address should not overlap with any other
existing addresses used by the cluster nodes or routers. For example,
if choosing 19.19.19.19, ensure that 19.0.0.0 (19.19.0.0) are not
routes known by the system routing tables.
- Add the takeover interface (for example, 19.19.19.19.
Create it with a line description of *VIRTUALIP, subnet mask of 255.255.255.255
(host route), maximum transmission unit of 1500 (any number in the
range 576-16388), and autostart of *NO. This takeover address (for
example, 19.19.19.19) does must exist as a *VIRTUALIP address before
identifying it as an associated local interface in next step. It does
not, however, must be active.
- Associate the intended takeover IP address with one or
both of the IP addresses that you specify to be used by cluster communications
when you create the cluster or add a node to the cluster.
- Create the cluster and create any CRGs.
For the application CRG, specify QcstUserCfgsTakeoverIpAddr for
the Configure takeover IP address field. Do
not start any application CRGs.
- Using Configure TCP/IP applications (option
20) from the Configure TCP/IP menu, then Configure RouteD (option
2), then Change RouteD attributes (option 1), ensure that the Supply
field is set to *YES. If not, set it to *YES. Then start or restart
RouteD (RIP or RIP-2) on each cluster node.
- NETSTAT option 3 shows the RouteD using a local
port if currently running. RouteD must be running and advertising
routes (ensure that the Supply field is set to *YES) on every cluster
node in the CRG recovery domain.
- Ensure that all the commercial routers in the network that
interconnect the recovery domain LANs are accepting and advertising
host routes for RIP.
- This is not necessarily the default setting for
routers. The language varies with router manufacturer, but the RIP
interfaces settings should be set to send host routes and receive
dynamic hosts.
- This also applies to both the router interfaces that point to
the systems as well as the router-to-router interfaces.
Note: Do not use an IBM i machine as the router
in this configuration. Use a commercial router (IBM® or
otherwise) that is designed for routing purposes. IBM i routing cannot be configured to handle this function.
- Manually activate the takeover address on one of the cluster
nodes:
- Wait up to 5 minutes for RIP to propagate the routes.
- Ping the takeover address from all nodes in the CRG
recovery domain and from selected clients on the LANs who will be
using this address.
- Ensure the takeover address is ended again.
(Clustering will start the address on the specified primary
node when the CRGs are started.)
- Start the application CRGs.
- The takeover address is started by clustering on the specified,
preferred node, and RIP advertises the routes throughout the recovery
domain. RIP might take up to 5 minutes to update routes across the
domain. The RIP function is independent from the start CRG function.
Important: - If the above procedure is not followed for all cluster nodes in
the application CRG recovery domain, the cluster hangs during the
switchover process.
- Even though you do not perform a failover to replica nodes, it
is a good idea to perform the procedure on the replica nodes in the
event that they might be changed at a later date in time to become
a backup.
- If you want to use multiple virtual IP addresses, then each one
will require a separate application CRG and a separate IP address
with which to be associated. This address may be another logical IP
address on the same physical adapter or it may be another physical
adapter altogether. Also, care must be taken to prevent ambiguities
in the routing tables. This is best achieved by doing the following:
- Add a *DFTROUTE to the routing table for each virtual IP address.
- To use multiple IP address use CFGTCP (option 2).
- Set all parameters, including the next hop, the same to reach
the router of choice; however, the Preferred binding interface should
be set to the local system IP address that is associated with the
virtual IP address that is represented by this route.