Graph Storage¶
Where the products keep the graph decides whether it survives a restart and whether you can run a
cluster. services.persistor chooses one of three:
| Choice | Where the graph lives | Survives a restart | Use it for |
|---|---|---|---|
cassandra |
A local Cassandra container, one keyspace per product | Yes | What quinectl init writes; required for a cluster |
rocks-db |
Files in your project directory | Yes | A single member without a Cassandra container |
in-memory |
The product's memory | No | Quick experiments you'll throw away |
Cassandra and RocksDB each have a short form and a long form. The long form shows every setting at its default.
Cassandra¶
quine:
edition: enterprise
licenseKey: ${QE_LICENSE_KEY}
customerId: ${QE_CUSTOMER_ID}
services:
persistor: cassandra
auth: keycloak
proxy: none
metrics: none
kafka: none
quine:
edition: enterprise
licenseKey: ${QE_LICENSE_KEY}
customerId: ${QE_CUSTOMER_ID}
services:
persistor:
cassandra:
image: cassandra:4.1
cqlPort: 9042
heap: 1G
auth: keycloak
proxy: none
metrics: none
kafka: none
Set these before Cassandra first starts.
Changing image or cqlPort empties every graph
If you change image or cqlPort later, the next up replaces the Cassandra container without
asking. The graphs of every project on this machine are empty afterwards.
heap applies only when the container is created. Changing it later has no effect.
quinectl starts a Cassandra container, quinectl-cassandra, on CQL port 9042 and waits until it
answers before launching anything. Each product stores its graph in its own keyspace: quine for
Quine Enterprise and novelty for Novelty, unless you set quine.keyspace or novelty.keyspace.
The container is shared by every project on the machine, and keeps running when you run
quinectl down, so the graph is still there next time. quinectl down --rm, run after quinectl down,
removes the container, and the next up starts with an empty graph. Give each project its own
keyspace; see Multi-tenancy.
RocksDB¶
quine:
edition: enterprise
licenseKey: ${QE_LICENSE_KEY}
customerId: ${QE_CUSTOMER_ID}
services:
persistor: rocks-db
auth: keycloak
proxy: none
metrics: none
kafka: none
quine:
edition: enterprise
licenseKey: ${QE_LICENSE_KEY}
customerId: ${QE_CUSTOMER_ID}
services:
persistor:
rocksDb:
filepath: quine.db
createParentDir: false
writeAheadLog: true
syncAllWrites: false
auth: keycloak
proxy: none
metrics: none
kafka: none
No storage container runs. Each product writes its graph to a directory in your project:
quine.db/ for Quine Enterprise and novelty.db/ for Novelty. The graph is kept when the product
stops: after quinectl down and quinectl up, it is still there.
- The long form's key is
rocksDb, notrocks-db. filepathputs the graph somewhere else, in a project that runs one product. With both products, quinectl refuses it.- With
runtime: docker, a missing directory is created for you.
RocksDB is for a single member. A cluster with more than one member needs Cassandra, and quinectl refuses the combination before starting anything.
In-memory¶
Keep the graph in the product's memory:
quine:
edition: enterprise
licenseKey: ${QE_LICENSE_KEY}
customerId: ${QE_CUSTOMER_ID}
services:
persistor: in-memory
auth: keycloak
proxy: none
metrics: none
kafka: none
Nothing is written anywhere. The graph is gone when the product stops.
Starting over with an empty graph¶
quinectl cass-reset deletes this project's keyspaces on Cassandra, both Quine Enterprise's and
Novelty's, so the next up starts with an empty graph. Other projects' keyspaces on the same Cassandra
are left alone. Two projects on the same keyspace share one graph, so while another project directory
declares this project's keyspace, cass-reset refuses, names that directory, and deletes nothing. Give
one of them its own keyspace first. It asks before deleting; --yes skips the question, and is required
in a script.
Stop the products, reset, and start again:
quinectl down
quinectl cass-reset
quinectl up
For RocksDB, stop the products and delete their RocksDB directories. In-memory storage starts empty every time.
Next steps¶
- To run several projects on one Cassandra, each with its own keyspace, see Multi-tenancy.
- To remove the Cassandra container and the other shared containers, see Stopping and removing.