The shards that hold the rows, a Slap.Cluster. A partition's rows are on shard shard_for(partition). Each shard runs partition writers, each the only process that writes its own subset of the shard's partitions. Set child_options: [partition_writers: n] on the cluster (default 16).

Run one instance of this built-in cluster per VM. It can own many shards and participate in a cluster across VMs. The same VM can also run Slap.Streams.Cluster.

To configure a cluster under your application's name, define a module with use Slap.KV.Cluster, otp_app: :my_app, supervise that module in place of Slap.KV.Cluster, and pass cluster: MyApp.KVCluster to KV operations and Slap.KV.HTTP.Router. Run one KV cluster per VM; peer discovery is node-wide.

Configure it like any Slap.Cluster:

config :slap_kv, Slap.KV.Cluster,
  store: {:url, "s3://bucket/prefix", aws_endpoint: "https://rustfs:9000"},
  path: "kv",
  shards: 8

Durable writes wait for SlateDB's next WAL flush. This cluster sets flush_interval to 10 ms; SlateDB's default is 100 ms. Other settings leave the 10 ms default intact. The README describes the latency and object-store request trade-off.

Give the cluster its own :path when it shares a store with another cluster.