Loom

Loom is a lightweight streaming queue/router for large messages (e.g. 1–256MiB) over QUIC or HTTP/3. It is designed for “Kafka/MQ-like” partition routing, but optimized for big binary payloads by chunking and applying backpressure.

What it does

• Producers stream a message as: message header (routing key + optional declared size) + N chunks + end-of-message.
• The server routes each message to a consumer chosen by partitioned rendezvous hashing of the routing key.
• Limits and behavior are controlled by loom.yaml:
  • router.max_message_bytes (default 256MiB)
  • router.max_chunk_bytes (default 64KiB)
  • backlog and backpressure behavior for full partitions / chunk pressure
  • per-message chunk buffering is derived from max_chunk_bytes (target ~1MiB)

Run with Docker

docker build -t loom:dev .

docker run --rm \
  -p 4242:4242/udp \
  -p 9090:9090 \
  -v "$PWD/loom.yaml:/config/loom.yaml:ro" \
  loom:dev /config/loom.yaml

Clients

Implement the Loom wire protocol over QUIC (or HTTP/3). See PROTOCOL.md.

Notes

• This is a streaming router, not a durable log; if no consumers are connected, messages are discarded.
• For production, configure TLS cert/key and set insecure_skip_verify: false for clients.
• Yes I wrote this README with ChatGPT. Bite me.