You Gotta Push If You Wanna Pull

Pull queries are the default in data systems but face challenges around performance, data shape, format, and location. Materialized views solve these by precomputing and storing query results optimized for specific access patterns. The key insight is that maintaining multiple derived views from a single canonical source is safe and practical. Push queries enable incremental view maintenance (IVM), updating only the affected portions of derived views as source data changes rather than recomputing everything from scratch. Combining push queries (for efficient incremental updates) with pull queries (for human-friendly on-demand access) gives the best of both worlds. Tools like Flink SQL, Materialize, RisingWave, Feldera, and pg_ivm implement this pattern, and the IVM space is growing rapidly.