How OAX Revolutionizes Open Analysis in Kuaishou’s Data Platform

OAX Overview

OAX (Open Analysis eXpressions) is Kuaishou’s open analysis language that builds on the metric middle‑platform, providing a unified query engine for BI, topic and custom products.

Design Background

Kuaishou previously had siloed analysis products, each with its own query service and language, leading to high maintenance cost, lack of standards and low integration efficiency. The solution is a unified, open architecture.

Design Principles

规范 (Standardization) : Define modeling, syntax and access protocols.

统一 (Unification) : Use a single query service and language for all analysis products.

开放 (Openness) : Allow external products to access the service and support extensible operators.

OAX Language Model

The model consists of five layers: data source, data table, data model, dataset, and data application.

Data source layer – manages raw sources.

Data table layer – imports physical tables, accelerates them, and materializes heavy views.

Data model layer – defines relationships (relational or automated modeling).

Dataset layer – abstracts models and adds computed columns.

Data application layer – external systems submit OAX queries via a unified access protocol.

OAX Syntax Design

Data Types

Four primitive types: text, numeric, date, boolean.

Compute Capabilities

Basic compute – numeric, string, date, type, logical, aggregate, metric, advanced functions.

Dynamic granularity – EXCLUDE, INCLUDE, FIXED to change grain during calculation.

Table compute – window, running, offset functions.

Five Analysis Elements

Dimension, Metric, Dataset, Time range, Filter condition (and optional aggregation). Example: GDP of each city in Hubei for 2020‑2021 with proportion.

Access Protocol

Requests are expressed with OAX syntax, results are returned in Apache Arrow‑wrapped Dataframe, and calls can be made via RPC, ADBC (JDBC‑like) or HTTP.

OAX Runtime Design

Initial Logic Construction

Transforms the five analysis elements into a logical plan containing high‑level operators and dataset references.

Query Orchestration

Expands high‑level operators (e.g., proportion) into sub‑queries and unfolds datasets into physical tables using model information.

Query Optimization

Rule‑based and cost‑based optimizations (predicate push‑down, join ordering).

Computation routing – decide whether to push the whole plan to the engine or execute part in memory.

Engine‑specific plan rewrites (e.g., ClickHouse local query).

Pipeline Execution

Optimized physical plan is submitted; results may be loaded into memory and further processed with DataFusion before returning to the user.

Summary & Benefits

OAX unifies Kuaishou’s previously siloed analysis services, improves development and usage efficiency, and establishes an open ecosystem that supports diverse BI needs.