Cassandra K8s key terms

A Kubernetes workload controller that gives each pod a stable, ordinal identity, its own persistent volume claim, and ordered start, stop, and rolling update behavior, which is required for database pods.

A Service with clusterIP set to None that publishes per-pod DNS records instead of load balancing, letting Cassandra nodes and clients resolve individual pod addresses by name.

A namespaced request for storage that binds to a PersistentVolume and is mounted into a pod, decoupling the Cassandra data directory from the pod lifecycle.

A cluster-scoped template describing how volumes are dynamically provisioned, including the provisioner, parameters, reclaim policy, and volume binding mode used for Cassandra data disks.

A StorageClass field, either Immediate or WaitForFirstConsumer, that controls when binding occurs; WaitForFirstConsumer ensures the volume is created in the same zone as the scheduled pod.

A PersistentVolume backed by a disk attached to one specific node, offering low latency and high throughput for Cassandra but sacrificing portability if that node is lost.

The rule controlling what happens to a PersistentVolume after its claim is deleted: Retain keeps data for manual recovery, while Delete removes the backing storage automatically.

A policy that limits how many pods of a label selector may be voluntarily evicted at once, protecting Cassandra quorum during node drains and cluster maintenance.

The kubectl operation that evicts pods from a Kubernetes node before maintenance, respecting PodDisruptionBudgets and triggering Cassandra data handoff when a node is removed.

A custom controller that watches custom resources describing a Cassandra cluster and reconciles actual StatefulSets, Services, and configuration to match the declared desired state.

A Kubernetes API extension that registers a new resource type, such as a Cassandra datacenter or cluster object, so the operator's custom resources can be stored and validated by the API server.

The continuous control cycle in which an operator compares observed cluster state with the desired spec and issues create, update, or delete calls until they converge.

An API server callback that validates or mutates custom resources before persistence, used by operators to reject invalid topology changes or to inject defaults and sidecars.

A metadata entry that blocks deletion of a custom resource until the controller completes cleanup such as decommissioning nodes or releasing storage, preventing orphaned cloud resources.

A contact point listed in cassandra.yaml that new or restarting nodes query to learn cluster topology; seeds should be stable and not all seeds should restart simultaneously.

Cassandra's peer-to-peer membership and state dissemination mechanism, exchanging node, schema, and load information so every node converges on the same cluster view.

A Cassandra component that reports network topology and rack placement to the cluster, informing replica placement and request routing decisions.

Configuring Cassandra so replicas land in distinct racks or zones, ensuring a zone failure does not remove all copies of a partition's data.

A replication strategy that specifies replica counts per datacenter and, where racks are defined, distributes replicas across racks for fault tolerance.

The number of copies of each partition stored across the cluster; it determines durability and the maximum consistency level achievable when nodes are unavailable.

A per-query setting specifying how many replicas must acknowledge a read or write before returning success, trading latency and availability against data freshness.

A majority of replicas, calculated as replication factor divided by two plus one, that must respond for strong consistency in reads and writes.

The first component of a Cassandra primary key that is hashed by the partitioner to determine which node and replicas store a row, driving data distribution.

A primary key component after the partition key that defines sort order within a partition and enables efficient range slices on disk.

The hash space from negative to positive values over which Cassandra distributes partitions; each node owns token ranges, and vnodes split ownership into many smaller ranges.

A configurable number of token ranges assigned per physical node, improving data distribution and speeding streaming during bootstrap and decommission.

The process by which a newly added node streams its assigned token ranges from existing replicas before joining the ring and serving traffic.

The controlled removal of a node that streams its data to remaining replicas before leaving the ring, unlike simply deleting a pod.

A mechanism where a coordinator stores writes destined for an unavailable replica and replays them when that node returns, reducing temporary inconsistency.

The anti-entropy process that compares replicas and reconciles divergent data, typically run with nodetool repair or the Reaper tool on a scheduled basis.

The background merge of SSTables that removes tombstones and obsolete versions, reclaiming disk space and improving read performance.

An immutable sorted on-disk file holding Cassandra rows for a token range, created by flushing memtables and later merged by compaction.

The append-only durability log written before memtables, replayed on restart to recover writes not yet flushed to SSTables.

A hard-linked point-in-time copy of SSTables on a node, used for backups; in Kubernetes it is often paired with volume snapshots for whole-disk capture.

A Kubernetes API object that captures a point-in-time copy of a PersistentVolumeClaim through a CSI driver, enabling consistent cluster-wide backups when coordinated across pods.

Rebuilding a cluster or keyspace from snapshots and commitlogs, then running repair, while ensuring the restored topology matches the original token ownership.

A Kubernetes scheduling constraint that distributes pods evenly across zones or nodes, aligning Cassandra replica placement with failure domains.

A controlled sequential restart of Cassandra pods, waiting for each node to rejoin and serve traffic before proceeding, to avoid losing quorum.

Two-way certificate authentication where both client and server present certificates, commonly used between Cassandra nodes and between drivers and the cluster.

The entity that issues and signs certificates for nodes and clients; its certificate must be trusted by every participant for TLS handshakes to succeed.

A Kubernetes controller that automates issuance and renewal of TLS certificates as Secrets, simplifying rotation of Cassandra node and client certificates.

A Java file holding a node's private key and certificate chain that Cassandra presents during TLS handshakes; its password is supplied through configuration or a Secret.

A Java file containing certificates of trusted peers and authorities that Cassandra uses to verify incoming connections when TLS is enabled.

Verifying the identity of a connecting user or node, for example with the PasswordAuthenticator or a pluggable authenticator backed by an external identity system.

Deciding which operations an authenticated identity may perform, governed by permissions granted to roles on keyspaces, tables, or the whole cluster.

A named security principal that can hold permissions and be granted to users or other roles, supporting hierarchical privilege management in Cassandra.

A built-in role with full cluster privileges, created at first startup; its default credentials must be changed and protected immediately in production.

A Java agent that exposes Cassandra MBean metrics over HTTP in Prometheus format, typically scraped through pod annotations and a ServiceMonitor.

A time-series monitoring system that scrapes metrics endpoints and evaluates alerting rules, commonly deployed alongside the Cassandra operator for observability.

A custom resource from the Prometheus Operator that selects Services and defines scrape intervals and paths for metric collection.

A visualization layer over Prometheus queries showing Cassandra latency, throughput, compaction backlog, and node health for operators and on-call engineers.