Tencent OLA t‑Metric Metric Platform: Headless BI Practices and Architecture

Tencent OLA is an enterprise‑level data‑asset platform built on DataOps principles, offering a one‑stop solution for metric creation, management, discovery, and usage, with the t‑Metric metric middle‑platform as its core component.

The presentation covers six main parts: the construction goals of the OLA platform, the relationship between t‑Metric and Headless BI, configuration‑driven metric production, a unified metric query service, the t‑Metric ecosystem, and a Q&A session.

1. OLA Platform Goals – Provide trustworthy data assets, improve data value and application efficiency, and build a full‑link data lineage graph on top of tools such as datamesh, us, tdw, and venus.

2. t‑Metric and Headless BI – Headless BI introduces a Metric‑Store to centralize metric definitions, eliminate duplicate metric implementations across front‑end and back‑end systems, and provide unified query services with features like materialized acceleration and metric catalogs.

3. Configuration‑Driven Metric Production – Users define metrics via a low‑code, configuration‑driven interface that maps business concepts (measure, dimension, filters, period) to SQL, supports Hive, MySQL, ClickHouse, and StarRocks, and performs duplicate and conflict checks.

4. Core Capabilities (Build, Manage, Discover, Use)

Build: Produce metrics through standardized configurations, supporting low‑code generation of materialization tasks.

Manage: Handle metric permissions, lifecycle, approvals, and change notifications.

Discover: Provide multi‑dimensional search (keyword, dimension, table, level, business theme).

Use: Offer a unified RESTful API and a metric query language (MQL) for consumption.

5. t‑Metric Architecture – Consists of storage/OLAP engines, a semantic layer for unified metric definitions, a materialization layer for acceleration and SLA guarantees, and a query layer that routes requests to the most cost‑effective engine.

6. Unified Metric Query Service – Supports both RESTful API and MQL, performs syntax/semantic validation, generates ASTs, selects optimal execution plans, leverages caching (single‑request and granular second‑level caches), and returns results with optional second‑stage aggregation.

7. Metric Ecosystem – Integrates with PCG platforms such as DataTalk, intelligent decision systems, attribution analysis, and experiment platforms; provides automated attribution of metric anomalies and lineage information.

8. Q&A Highlights

Real‑time analysis depends on the underlying storage engine; materialized acceleration yields sub‑second P95 latency.

Dimension explosion is mitigated by selecting relevant dimension combinations for materialization (cube pruning).

Metric trust is ensured through certification workflows and conflict detection.

Future plans include large‑model (ChatGPT‑style) conversational interfaces that rely on high‑quality metric metadata.

The session concludes with thanks to the audience.