slap

slap provides a Mix-run HTTP server for evaluating, testing, and benchmarking Durable Streams and KV. It serves either or both services from one listener. It does not serve Files or Yjs, and it does not authenticate requests.

Applications normally depend on the embeddable service packages instead, so that they own their supervision tree and HTTP routing; see below. Adding slap as a dependency does not start a listener: run mix slap.server, or supervise {Slap.Server, opts}. The Mix task is not available in a release, so a release that needs this listener supervises {Slap.Server, opts}.

The supported setup has one Slap.Streams.Cluster and one Slap.KV.Cluster per VM. A cluster can have many shards and span several VMs.

Embed a service in your application

Choose the package for the service your application needs:

The embeddable packages do not start an HTTP listener. The Streams and KV HTTP routers are Plugs that can be mounted in your application's HTTP pipeline or served by a listener you supervise. Put your authentication and authorization in front of them. The linked package READMEs show the required children and in-process APIs.

Install the Mix-run server

If another Mix project needs this server, add slap to its mix.exs:

defp deps do
  [
    {:slap, "~> 0.1.0"}
  ]
end

Run mix deps.get. API documentation is on HexDocs.

Try the standalone server

With Elixir 1.18 or later, from slap/ in a checkout of the repository, run mix deps.get, then one of these commands:

mix slap.server --streams --store memory
mix slap.server --kv --store memory
mix slap.server --streams --kv --store memory

Use iex -S mix slap.server --streams --store memory to keep an IEx prompt while the server runs.

The listener binds to 127.0.0.1 by default. Use --ip to listen on another address. The memory store loses its data when the server stops. To try S3, create the bucket, configure AWS_REGION and credentials in the server's environment, then run:

mix slap.server --streams --kv --store s3:s3://my-bucket/my-app

Streams

With Streams running, create a JSON stream, then append to it. The JSON extension stores each element of an appended array as a separate message.

curl -X PUT -H 'Content-Type: application/json' http://localhost:4437/v1/stream/events
curl -X POST -H 'Content-Type: application/json' \
  --data '[{"type":"created","id":1},{"type":"updated","id":1}]' \
  http://localhost:4437/v1/stream/events

Read the stream. The JSON messages are returned as an array.

curl 'http://localhost:4437/v1/stream/events?offset=-1'

Fork the stream at its current end, then append to the fork. The source still contains only the first two messages.

curl -X PUT -H 'Content-Type: application/json' \
  -H 'Stream-Forked-From: /v1/stream/events' \
  http://localhost:4437/v1/stream/events-fork
curl -X POST -H 'Content-Type: application/json' --data '{"type":"deleted","id":1}' \
  http://localhost:4437/v1/stream/events-fork
curl 'http://localhost:4437/v1/stream/events-fork?offset=-1'
curl 'http://localhost:4437/v1/stream/events?offset=-1'

KV

With KV running, set three rows, get one, then scan a range of keys in the people partition. The range stops before c.

curl -X PUT --data-binary 'Ada' http://localhost:4437/v1/kv/people/a
curl -X PUT --data-binary 'Grace' http://localhost:4437/v1/kv/people/b
curl -X PUT --data-binary 'Katherine' http://localhost:4437/v1/kv/people/c
curl http://localhost:4437/v1/kv/people/a
curl 'http://localhost:4437/v1/kv/people?gte=a&lt=c'

The scan returns JSON with keys and values encoded as unpadded base64url.

mix help slap.server lists the options: the store, the listener address, each service's shard count and flush interval, the long-poll and SSE timeouts, a pid file, and the placement (local, object-lease, distributed or static) for running as one node of a cluster.

object-lease requires an S3 store; --peers requires a placement other than local.

In a store, the streams' shards are under streams/ and KV's under kv/, each with its own leases when the placement uses them.

To supervise the standalone listener in a test or benchmark harness, add {Slap.Server, opts} as a child. It starts the selected clusters and a Bandit listener together:

children = [
  {Slap.Server,
    store: {:url, "s3://bucket/db"},
    streams: [shards: 8],
    kv: [shards: 4],
    port: 4437}
]

Development and testing

This project also holds the tests that run the server as an OS process: the official Durable Streams conformance suite and benchmarks, the kill -9 crash tests, the cluster tests and the streams HTTP load benchmark.

Durable Streams conformance and benchmarks

durable_streams/ runs version 0.3.7 of @durable-streams/server-conformance-tests against a running server:

mix slap.server --streams --store memory --long-poll-timeout 500 &    # the reference servers use 500 ms for this suite
cd durable_streams && pnpm install && pnpm test

The official suite covers stream operations, including forks, on the in-memory and RustFS stores in CI. The suite skips subscriptions, which this server does not provide.

pnpm bench in durable_streams/ runs the official HTTP benchmarks, @durable-streams/benchmarks 0.2.7: append-to-long-poll round-trip latency, and append throughput for 100-byte and 1 MB messages. Both suites read the server's URL from DURABLE_STREAMS_URL (default http://127.0.0.1:4437).

scripts/streams_bench.sh runs the benchmarks against mix slap.server and against the official Caddy server (a pinned release, downloaded and checked), one after another, and durable_streams/summary.js compares the results. From the repository root:

just streams-bench                                  # slap-local caddy-file
just streams-bench slap-memory caddy-memory slap-local caddy-file

slap-local and caddy-file both reply to an append once it is on disk, so they are the comparable pair. slap-memory still waits for a SlateDB flush, which caddy-memory does not. The memory store keeps the objects SlateDB writes in RAM, about twice the bytes appended, and deleting the streams does not free it within minutes. The 1 MB benchmark appends a fixed time's worth of data, 1.5 GB in a typical run, so slap-memory needs about 4 GB of free memory, more on a faster machine. When started from the Actions tab, the slap_streams_bench.yml workflow runs slap-local, slap-s3 (on RustFS) and caddy-file; a run on main also charts the results.

Crash test

scripts/crash_test.exs runs mix slap.server as its own OS process. It:

1. has 16 writers append to their own JSON streams (half with idempotent producers)
2. kills the server with kill -9 at a random moment between 0.3 and 3 s
3. restarts the server on the same store
4. lets the producers retry the appends that were in flight, and all writers append more
5. reads every stream back

Every acknowledged append must be present exactly once, at the offset its acknowledgement gave, so no offset is reused after the crash. Each writer's messages must be in order, with no message stored twice.

mix run scripts/crash_test.exs --runs 20                         # local directory
mix run scripts/crash_test.exs --runs 20 --store s3:s3://bucket/crash

CI runs the crash test on local storage and RustFS. An append in flight when the process dies may or may not be stored.

Streams load

bench/streams_http_load.exs measures stream append throughput and latency against a running server. Choose shard count and flush_interval from the store's sustained PUT rate: each active shard can write one WAL object per flush interval. Run the benchmark on the intended store before sizing a deployment.

Cluster tests

scripts/cluster.sh start DIR runs a local cluster of three mix slap.server nodes (separate OS processes, one store, the distributed placement). The Durable Streams conformance suite runs through one of them, so most requests, long-polls and SSE streams are served by another node.

scripts/cluster_crash_test.exs runs 16 writers (half with idempotent producers) through the nodes of a 3-node cluster and injects a fault into one node at a random moment, then checks every stream as the single-node crash test does: every acknowledged append exactly once at its acknowledged offset, no duplicates, each writer's messages in order.

• kill - SIGKILL the node; the others take its shards (when its leases expire, or at once with distributed); it is restarted and takes its share back.
• pause - SIGSTOP the node past its leases (or net_ticktime): the others take its shards, fencing it in SlateDB; on SIGCONT it finds it lost them and gives them up.

The test measures how long each shard is unavailable after a kill. With distributed placement, failover waits for node failure detection and the settle interval. With object-lease, it waits for the old lease to expire and for the shard to reopen. CI runs the Durable Streams conformance suite through the cluster and injects both faults under each placement.

License

Slap is released under the terms of the Apache License 2.0.

Copyright (c) 2026, Michael Russo.