Skip to content

Run a Local Cluster

quinectl can run a Quine Enterprise cluster on your machine: several members sharing one graph in Cassandra, plus hot spares that stand by to take over from a failed member. Use it to see how clustering, failover and cluster-wide ingest behave before you deploy a real cluster.

See it as a guided tour

The Kubernetes Deployment Tour shows the same cluster on Kubernetes, one card at a time: how members find each other, what a hot spare is, and a member being killed and a spare taking its place. Take it before you plan a production cluster.

The quinectl.yaml that quinectl init writes has a cluster: block under quine:, commented out. Uncomment it and set size and hotSpares. A complete file looks like this:

quinectl.yaml
quine:
  edition: enterprise
  licenseKey: ${QE_LICENSE_KEY}
  customerId: ${QE_CUSTOMER_ID}
  cluster:
    size: 3
    hotSpares: 2
services:
  persistor: cassandra
  auth: keycloak
  proxy: none
  metrics: none
  kafka: none
  • Enterprise only. Clustering is a Quine Enterprise feature.
  • Cassandra only. RocksDB and in-memory storage belong to one process, so every member would have its own graph. A cluster with either is rejected before anything starts.
  • Memory. Each member is a full Quine Enterprise process, so give Docker enough memory for all of them before you start.

Start it

Start the cluster:

quinectl up

Every member runs in the background. up waits until every member is running and size of them are ready, prints one row per member, and returns. Hot spares show as spare (live, standing by); they never report ready themselves, and their readiness check answers 503. Any member can be the spare, not only the last one.

Each member gets its own web port, counting up from quine.port: 8080, 8081, 8082, and so on. Members talk to each other on a cluster communication port, quine.cluster.portBase (default 25520). With the image, every member uses that same port inside its own container. With runtime: jar, members share your machine, so the port counts up from portBase.

All members share one Keycloak realm, so the sample users and quinectl token work against any member. A hot spare accepts the token but doesn't serve the graph: a query sent to it fails with 500 until it takes over. Send queries to a member that shows ready.

To reach the whole cluster at one address, add the reverse proxy, which spreads requests across the members that were ready at the last up.

Check on it

Nothing stays in the foreground, so status is how you watch the cluster:

quinectl status

status lists every member with its web address, its readiness and where its logs are. Without auth, it shows a hot spare as spare (live, standing by), as up does. With Keycloak auth, status can't ask the cluster which member is the spare, so the spare shows as live, not ready.

Stop it

Stop every member:

quinectl down

Cluster-wide Kafka ingest

With services.kafka: local, the ingest request quinectl prints for a cluster with a size of 2 or more creates a cluster-wide ingest, with the topic's partitions placed across the members, instead of an ingest on one member. See Cluster-wide ingest on the Kafka page.

Next steps