Goal
Validate that the engine layer compiles cleanly to ClickHouse SQL, and implement it. Known unknowns:
- Schema fit. Does ClickHouse's
Map(String, String), or similar trait storage, carry the same semantics as Snowflake's VARIANT? Are JSON paths queryable at the latencies we need?
- Which engine operators map cleanly, which need ClickHouse-specific helpers, and which will have limited support?
- Cost shape. Ideally, we want comparable workload cost in ClickHouse Cloud vs Snowflake at our test sizes (10M, 100M, 900M identities).
Output: a dialect implementation in flagsmith-sql-flag-engine, a ClickHouse dialect test harness (against local ClickHouse in docker-compose), and a written tradeoff comparison vs Snowflake.
Goal
Validate that the engine layer compiles cleanly to ClickHouse SQL, and implement it. Known unknowns:
Map(String, String), or similar trait storage, carry the same semantics as Snowflake'sVARIANT? Are JSON paths queryable at the latencies we need?Output: a dialect implementation in
flagsmith-sql-flag-engine, a ClickHouse dialect test harness (against local ClickHouse in docker-compose), and a written tradeoff comparison vs Snowflake.