Graph Language and GeaFlow DSL for Integrated Graph‑Table Analytics in Ant Financial

This talk introduces graph language concepts and their applications, describing graph models (property graphs, attribute graphs) and comparing their expressive power and performance advantages over relational tables.

It reviews major graph query languages such as Gremlin and ISO/GQL, highlighting their pattern‑matching semantics and the lack of a unified standard, and explains why GeaFlow adopts both Gremlin and ISO/GQL.

The speaker outlines typical use cases in social networks, collaborative recommendation, and financial risk control, emphasizing the need for integrated graph‑table processing.

GeaFlow is presented as Ant's real‑time graph computation engine, detailing its evolution, architecture (SQL plus, unified parser, logical/physical plans, graph optimizer, runtime) and the DSL that unifies graph construction, SQL, and Gremlin/ISO‑GQL queries.

Key design aspects of the DSL include syntax for CREATE GRAPH and INSERT statements, support for both batch and streaming execution, and the ability to express complex graph‑table workflows in a single language.

Performance optimizations are discussed, covering distributed execution of complex graph queries via an enhanced DAG model with jump operators and sub‑DAGs, full‑link binary data formats to reduce serialization overhead, and graph‑table optimizer techniques such as predicate push‑down and column pruning.

Future plans involve extending DSL syntax, adopting native and vectorized execution, tighter integration with data lake and Kafka ecosystems, and improving CBO path‑matching optimization.

The Q&A section clarifies that GeaFlow’s SQL is based on Apache Calcite, ISO/GQL is a subset of the ISO standard, and that the engine focuses on low‑latency, non‑transactional graph analytics rather than strict ACID guarantees.