Skip to content

Multi-cluster fleet (shared sink)

Run one operator per cluster, all exporting to the same backend with a distinct spec.cluster value. No hub tier — Postgres, Git, or event sinks merge by cluster id.

Prerequisites

  • Two (or more) clusters with network reachability to the shared sink (Postgres, Git remote, or NATS/Kafka broker)
  • Helm chart mode: single (default) on each cluster

Postgres fleet

Each cluster installs the operator with the same DSN and a unique cluster label:

# kollect-doc: ignore Helm values, not a kollect CR
# cluster-a — values fragment
mode: single
# …

Apply a database sink (see postgres-state-store.md) with:

apiVersion: kollect.dev/v1alpha1
kind: KollectDatabaseSink
metadata:
  name: fleet-postgres
  namespace: platform
spec:
  type: postgres
  cluster: cluster-a   # unique per installation
  postgres:
    databaseRef:                    # Secret holding the DSN; never inline credentials
      name: inventory-postgres-dsn
      namespace: kollect-system
    schema: public
    table: inventory_items

Repeat on cluster B with spec.cluster: cluster-b. Rows merge in one table; primary key includes cluster (ADR-0501).

Git fleet

Use pathTemplate on a snapshot sink:

# kollect-doc: fragment KollectSnapshotSink
spec:
  type: git
  endpoint: https://github.com/org/inventory.git
  cluster: cluster-a
  pathTemplate: clusters/{cluster}/inventory.json
  git:
    branch: main

Each cluster commits under its own path; CI can aggregate clusters/* if a single commit is required.

Inventory collection

Per cluster, use the same pattern as deployment-inventory.md:

  • KollectProfile for the GVK schema
  • KollectTarget selecting workloads
  • KollectInventory referencing the fleet sink

Samples: config/samples/e2e/team-inventory.yaml in the repository (single-cluster shape; add sink refs for your backend).