Database guides

ScyllaDB

Updated 2026-08-01 · 2 min read

Cassandra-compatible and written in C++ for latency. Same CQL, same driver in AddisDB, same schema introspection.

ScyllaDB is a C++ reimplementation of Cassandra that keeps the CQL protocol and data model. AddisDB serves it with the same driver, so keyspaces, tables, partition keys and clustering keys all show up exactly as they do for Cassandra.

Who it is for

Scylla exists to get more out of each machine. A shard-per-core architecture and its own I/O scheduler mean predictable low latency at high throughput, which typically translates into a much smaller cluster for the same workload.

Because each core owns a shard, drivers that understand Scylla can route a request to the exact core holding the data — no internal hop, no cross-core locking. There is also no JVM, so there is no garbage-collection pause to explain away in your p99.

Choose it when you want Cassandra’s model and ecosystem but your Cassandra cluster has grown expensive, or when tail latency is a product requirement rather than a metric.

Set up the cluster

  1. Local: docker run --name scylla -p 9042:9042 -d scylladb/scylla.
  2. Managed: ScyllaDB Cloud. Create a cluster, add your IP to the allowlist, and copy the node address and credentials.
  3. Create a keyspace before creating tables — the CQL is identical to Cassandra’s.
  4. Migrating from Cassandra? Your existing CQL schema applies unchanged.
  5. On a laptop, add --smp 1 --memory 1G so a single container does not try to claim every core.
# A polite local Scylla
docker run --name scylla -p 9042:9042 -d \
  scylladb/scylla --smp 1 --memory 1G --overprovisioned 1

Connect from AddisDB

  1. New Connection → ScyllaDB under Wide-column. Port prefills to 9042 and the username to cassandra.
  2. Enter host, credentials and the Keyspace you want as your default.
  3. Test, then Save.

What AddisDB gives you

  • Keyspaces and tables with column types and primary keys read from the system schema.
  • CQL results in the grid, with the JSON tab for collections and user-defined types.
  • Read-only enforcement for anything that is not a read statement.
  • The Chart view, ⌘K search, and AI chat grounded in your schema.
  • Size estimates and the clients view in the Live Monitor.

Everything written in the Cassandra guide about modelling — partition keys, bucketed time series, ALLOW FILTERING as a warning rather than a solution — applies here unchanged. Scylla is faster, not more forgiving.