Production MessagePack

Real-world usage:

Library options:

• Python: msgpack-python, umsgpack.
• Go: vmihailenco/msgpack, tinylib/msgp.
• JavaScript: msgpack-lite, @msgpack/msgpack.
• Rust: rmp, rmp-serde.
• Java: msgpack-core, msgpack-jackson.

Common pitfalls:

• Strings vs bytes: MsgPack distinguishes; pre-2013 didn't. Compatibility issues with old data.
• Integer overflow: 64-bit ints don't fit in JSON's 53-bit float. MsgPack handles native; serialization to JSON loses precision.
• NaN/Inf in floats: MsgPack handles. JSON doesn't (some libs special-case).
• Map key types: Some libraries restrict. Stringify if portability matters.
• Large messages: 32-bit length limits. Plan for streaming if approaching.

Schema validation:

• MsgPack alone has no schema. Apps must validate.
• Solutions: combine with JSON Schema, or use Protobuf if strict schema needed.

Versioning:

• No native versioning.
• App can include version field in maps: {"v": 2, "name": ...}.
• Or branch on type detection.

Performance tips:

• Reuse encoder/decoder instances (avoid allocation).
• Use pre-sized buffers.
• For known types: handcoded faster than reflection.
• Streaming for large data; avoid full-buffer allocation.

Compression:

• Already binary; gzip helps for very repeating data.
• Snappy + msgpack common for cache wire formats.
• Trade CPU for bytes.

Use cases I've seen:

• Redis values (compact storage, faster than JSON).
• Cache layer for API responses.
• Job queue payloads.
• Cross-language RPC (Tarantool, Apache Hivemall).
• IoT device → cloud data.

When NOT to use:

• Public-facing APIs (browser doesn't speak it natively).
• Human inspection (binary, not readable).
• Strict schema enforcement (use Protobuf).

Modern alternatives gaining traction:

• CBOR (RFC 8949): nearly identical, IETF-blessed.
• Cap'n Proto: zero-copy alternative if perf-critical.

For most "JSON but smaller" needs: MsgPack still solid choice.