Glossary¶
Plain-language definitions of the terms used throughout this guide. Where a term has a chapter that explains it in depth, follow the link.
- amd64 (x86-64)
- The CPU architecture of Intel/AMD processors. All our nodes use it; mixing it
with ARM (
cax) would break CloudNativePG. - Base backup
- A full, point-in-time copy of the database files. Combined with archived WAL, it enables point-in-time recovery. See PostgreSQL HA.
- Barman Cloud Plugin
- The CloudNativePG add-on that performs backups and restores to S3-compatible
object storage. Replaces the deprecated in-tree
barmanObjectStore. - CNI (Container Network Interface)
- The plugin that provides pod networking (Flannel, Cilium, Calico). Determines whether NetworkPolicies are enforced.
- CNPG-I
- The CloudNativePG plugin interface; how the Barman Cloud Plugin talks to the operator over mTLS.
- Cluster (CNPG)
- The custom resource (
postgresql.cnpg.io) describing your PostgreSQL deployment: instances, storage, config, backups. The centerpiece manifest. - Controller / reconciliation
- A control loop that observes actual state, compares it to desired state, and acts to close the gap. The core mechanic of Kubernetes. See the Kubernetes primer.
- CRD (Custom Resource Definition)
- Teaches the Kubernetes API a new kind of object (e.g.
Cluster). - dataDurability
- A CNPG synchronous-replication setting (1.25+).
required= strict (pause writes if the sync standby is gone, RPO=0);preferred= self-healing. - Failover
- Unplanned promotion of a standby after the primary fails.
- ImageCatalog / ClusterImageCatalog
- Maps a PostgreSQL major version to a specific operand image. Cluster-wide
variant is
ClusterImageCatalog. - Kured
- The daemon that coordinates node reboots for OS updates. On MicroOS this makes failovers a routine event.
- Longhorn
- Distributed block storage for Kubernetes. Provides snapshots and S3 backups — the reason we chose it over Hetzner's snapshot-less CSI.
- Operand image
- The container image that actually runs PostgreSQL (e.g.
postgresql:18.3-system-trixie). - Operator
- A controller with application-specific knowledge. CloudNativePG is the PostgreSQL operator. See Why CloudNativePG.
- PITR (Point-in-Time Recovery)
- Restoring to any past moment by replaying archived WAL onto a base backup.
- Pooler / PgBouncer
- A connection pooler that multiplexes many client connections onto few server connections. See Connecting & pooling.
- Primary / Standby
- The read-write instance and its read-only replicas.
- PVC / PV / StorageClass
- A storage request, the actual volume, and the recipe for provisioning volumes, respectively.
- R2 (Cloudflare)
- S3-compatible object storage with no egress fees; our backup destination.
- RPO (Recovery Point Objective)
- How much data you can afford to lose, measured in time. Synchronous replication targets RPO=0.
- Service (Kubernetes)
- A stable network name/IP that load-balances to a changing set of pods. The
operator keeps
pg-rwpointed at the primary. - Switchover
- Planned move of the primary role, with no data loss.
- Synchronous replication
- The primary waits for a standby to confirm a write before reporting commit. Trades latency for durability.
- WAL (Write-Ahead Log)
- The append-only log of changes Postgres writes before touching data files. Powers crash recovery, replication, and PITR.
- Zonal volume
- A volume confined to one cloud location; a pod using it can only run there. Why we keep all nodes in one location.