Changing Cluster Topology
You can use the Command Line Admin Utility to add or remove
nodes from the search cluster or repartition the search cluster.
The Command Line
Admin Utility is located in bin\native folder
in the Search Service installation folder (for example, C:\bea\alui\ptsearchserver\6.5\bin\native\cadmin.exe).
- Run the topology command.
% cadmin topology new.nodes
- Change the cluster.nodes file.
To add new nodes to the search cluster (for failover capacity),
install a new node, and edit the cluster.nodes file to include the node as a peer on an existing partition.
Issue a “soft reset” to the cluster through the command line utility,
which causes all nodes to re-examine the cluster topology file and
thus recognize the new node. When the new node receives a soft reset,
it recognizes that it needs to catch up to the rest of the cluster
and begins the automated index recovery process from the last checkpoint.
- To repartition the cluster, edit the number of partitions
in the cluster.nodes file. You will be asked
to confirm the action and the admin utility will confirm that a checkpoint
exists before performing the repartitioning operation.
Since changing cluster topology can be a time-consuming process,
the admin utility displays its progress.
Example Output from Adding and Removing Nodes
Current topology:
<contents of current cluster.nodes file>
New topology:
<contents of new.nodes file>
Nodes to add: p0n2, p1n2, p2n2
Nodes to remove: p0n0, p1n0, p2n0
Is this correct (y/n)? y
Applying changes…
p0n2 has joined
p2n0 has left
...
Changes applied successfully
Current topology:
<contents of current cluster.nodes file>
New topology:
<contents of new.nodes file>
Nodes to add: p3n0, p3n1
Is this correct (y/n)? y
CAUTION: the requested changes require repartitioning the search collection
The most recent checkpoint is checkpoint_xxx from 2004-04-22 16:00:00
Is this correct (y/n)? y
Repartitioning from 3 partitions into 4
0%
5%
<progress messages>
100%
Repartitioning successful
Applying changes…
p0n2 has joined
p2n0 has left
...
Changes applied successfully
If the repartition fails, the search collection leaves the
cluster in its original state, if at all possible, and provides information
about the failure. The cluster.nodes file is rolled back to the previous
state after making sure that the last-known good checkpoint refers
to an un-repartitioned checkpoint directory.
