Installation, Configuration, and Cluster Provisioning
This domain tests provisioning a node correctly the first time: package or tarball install, cassandra.yaml settings, JVM and OS tuning, ports, and safe cluster admission.
Install and identity
- Install from tarball, Debian/RPM, or a supported image; run as a non-root cassandra user. Disable swap, configure NTP or chrony, raise file descriptors and process limits, and use XFS or ext4 with suitable mount options.
- cluster_name must match across all nodes in one cluster. A new independent cluster needs a unique name; changing it later is not a rolling operation.
- seeds is a contact list, not a master list. Use a small, stable set, ideally one per rack or DC, and do not make every node a seed.
- Set listen_address and broadcast_address for internode traffic, and rpc_address and broadcast_rpc_address for clients. endpoint_snitch must match real topology, such as GossipingPropertyFileSnitch with cassandra-rackdc.properties, or Ec2Snitch in AWS.
- Ports: 7000 internode, 9042 CQL, 7199 JMX. Open only what is needed and secure JMX and SSL.
Worked example
To add a node to an existing DC, copy the same cluster_name, partitioner, and snitch configuration; set num_tokens or allocate_tokens_for_local_replication_factor consistently; point seeds at stable existing nodes; set the correct rack in cassandra-rackdc.properties; start the node and let it bootstrap stream its ranges. Do not bootstrap multiple nodes simultaneously.
Traps
Mismatched cluster names or partitioners; using public IPs where private gossip is intended; leaving default num_tokens without considering vnode density; forgetting broadcast_rpc_address behind NAT; and changing snitch or topology without repair or rebuild.