Show HN: Timeplus Proton 3.0 – First vectorized streaming SQL engine (github.com)

Timeplus announced Proton 3.0, an open-source, vectorized streaming SQL engine implemented as a single C++ binary that combines real-time stream processing with queryable analytical materialized views. Positioned as a lightweight alternative to Apache Flink and ksqlDB, Proton uses SIMD-optimized, columnar execution (hence “vectorized”) and embeds ClickHouse as its table store (with write-ahead logs and stream-local data structures). It supports native sources/sinks (Kafka/MSK, ClickHouse, MySQL/Postgres, S3/Iceberg), windowing, watermarks, CDC and mutable streams, incremental materialized views, alerts, tasks and Python/JS UDFs — all via SQL. No JVM, ZooKeeper or external dependencies; the binary is <500MB and runnable via Docker, Homebrew or tiny cloud instances. Proton is Apache 2.0. Technically notable: Proton reports SIMD-driven throughput like 90M events/sec and ~4 ms end-to-end latency on an M2 Max, and handles high-cardinality aggregations (1M unique keys). By unifying low-latency stream transformation with persistent analytical storage, it can pre-process and serve feature pipelines, online aggregations and anomaly detectors without separate batch ingestion steps into ClickHouse. For ML/AI teams this means faster feature updates, simpler real-time ETL, lower operational overhead (no JVM ecosystem), and fewer moving parts when building low-latency inference, monitoring, and personalization systems.