Runs the Slap services under one supervisor: the Streams cluster, the KV cluster, and the services built on them.

Add {Slap, opts} to a supervision tree, or call start_link/1:

children = [
  {Slap,
   store: {:url, "s3://bucket/prefix"},
   streams: [shards: 8],
   kv: [shards: 8],
   files: [],
   yjs: []}
]

Each service is started when its key is present, with a keyword list of its options ([] for the defaults). At least one cluster is required.

  • :store - required, as for Slap.SlateDB.open/2: for example :memory, {:local, dir} or {:url, "s3://bucket/prefix"}. File objects default to files/ in it; inline bodies are stored in KV.
  • :streams - starts Slap.Streams.Cluster: [shards: n, path: path, settings: settings, child_options: options]. A key left out comes from the cluster's application config (config :slap_streams, Slap.Streams.Cluster), and otherwise :shards defaults to 8 and :path to "streams". The shard count is fixed for the life of the data. See Slap.Streams.Cluster for child_options:. Snapshot logs (Slap.SnapshotLog) need only this cluster.
  • :kv - starts Slap.KV.Cluster, with the same keys (config :slap_kv, Slap.KV.Cluster; by default 8 shards under "kv"). See Slap.KV.Cluster for child_options:.
  • :files - starts Slap.Files after the KV cluster; requires :kv. Takes Slap.Files' options except :cluster. :store defaults to the store above.
  • :yjs - starts Slap.Yjs.Docs after the Streams cluster; requires :streams. Takes its options except :cluster.
  • :strategy - module or {module, opts}, a Slap.Cluster.Strategy for both clusters, to run as one node of a cluster. The nodes must be connected with distributed Erlang; see :peers. Without it, or a strategy in a cluster's application config, every shard is on this node (Slap.Cluster.Strategy.Local).
  • :peers - a fixed list of node names to stay connected to, through Slap.Cluster.Peers. Omit it on one node, or if something else connects the nodes, such as libcluster.

:settings merge with each cluster's defaults, which set flush_interval to 10 ms.

Run one Slap per VM: it starts the built-in clusters, and a VM runs at most one Streams cluster and one KV cluster. Its supervisor is registered as Slap, so a second start returns {:error, {:already_started, pid}}. An application that defines its own cluster modules (use Slap.Streams.Cluster) supervises them and the services on them directly.

Each cluster and its dependent services run in a separate :rest_for_one branch: a cluster restart also restarts its services. Repeated service failures can restart the cluster; repeated branch failures stop Slap. The supervisors use OTP's default restart limits.

Slap starts no HTTP listener. Slap.Router serves the Streams and KV HTTP APIs as one Plug.

Invalid option keys or service combinations raise ArgumentError when building the child spec or calling start_link/1. Cluster values, including :settings, are checked at startup. A top-level nil is treated as missing.

Summary

Functions

Starts a supervisor that runs the services listed in opts.

Functions

start_link(opts)

@spec start_link(keyword()) :: Supervisor.on_start()

Starts a supervisor that runs the services listed in opts.