Streaming SQL in C++

Clink brings streaming SQL into a native C++23 runtime. A SQL statement is compiled into a continuously running operator graph rather than executed as a one-shot database query.

The SQL frontend uses the PostgreSQL grammar through libpg_query and adds streaming concepts including event-time windows and MATCH_RECOGNIZE.

Example

CREATE TABLE bids (
  auction BIGINT,
  bidder BIGINT,
  price BIGINT,
  event_time BIGINT
) WITH (
  connector='kafka',
  format='json',
  topic='bids',
  event_time_column='event_time',
  watermark_lag_ms='2000'
);

SELECT auction, SUM(price)
FROM bids
GROUP BY auction;

Clink validates state-retention requirements for operations that can retain unbounded state and supports explicit TTL controls where appropriate.

Event time and state

Streaming queries can use watermarks, tumbling/sliding/session windows, joins, aggregations, CEP and stateful operators. Checkpoints capture operator state and replayable source positions for recovery.

See the full SQL reference, time and windowing internals, and checkpointing.

Connectors from SQL

Depending on the build, SQL tables can bind directly to Kafka, Pulsar, RabbitMQ, NATS, PostgreSQL, MySQL, ClickHouse, Redis, S3/Parquet, Iceberg, Kinesis and HTTP-oriented systems. The runtime exposes a capability manifest describing what the current binary actually contains.

See Connectors.

Native embedding

SQL execution does not require a JVM service beside the application. Clink can run the pipeline in-process, expose results through its C ABI and Arrow interfaces, or submit the same pipeline to a distributed Clink cluster.

For the runtime model, see C++ stream processing engine.