Inside Alibaba’s Tesla: Data‑Driven Ops for 100k+ Big Data Nodes

Introduction

The talk, originally presented at the 9th China Database Technology Conference (DTCC) in March 2019, introduces Tesla, Alibaba’s data‑driven SRE platform that standardizes daily operations for both offline and real‑time big‑data systems.

Big Data SRE

Alibaba’s big‑data SRE team builds a unified SRE middle‑platform (Tesla) that supports over 100,000 nodes across the company’s data infrastructure, applying software‑engineering principles to operations.

Tesla Operations Solution

Tesla consists of a unified operations portal (ticketing, vertical search) and four core platforms—process, configuration, job, and data—providing capabilities such as ticket management, automated change release, unified configuration, task scheduling, intelligent monitoring, anomaly detection, and self‑healing.

DataOps – Data‑Driven Operations

DataOps is defined as the three‑stage loop of perception, decision, and execution based on operational data. It parallels autonomous driving: data collection, analysis, and automated actions form a closed‑loop AIOps pipeline.

Practical Cases

Full‑Link Diagnosis

A diagnostic tool captures end‑to‑end metrics for MaxCompute jobs, automatically tracing failures across stages and presenting visual reports.

Hardware Self‑Healing

For a fleet of >100k physical machines, Tesla collects hardware metrics, streams them to Blink, analyses anomalies, and triggers automated repair actions via the workflow platform.

Data Value Transformation

By building a unified data warehouse (OneData) on top of Alibaba’s data middle‑platform, the team provides services such as anomaly detection, fault auto‑recovery, visual workflows, and knowledge‑graph‑driven vertical search.

AIOps Journey

AIOps is positioned as DataOps plus AI. The roadmap mirrors autonomous driving levels L0‑L5, progressing from manual ops to fully autonomous, AI‑enhanced operations. Examples include ChatOps assistants that answer queries about machine status and trigger automated migrations.

Conclusion

The presentation recaps the evolution from DevOps to DataOps and finally to AIOps, emphasizing that data‑driven operations, knowledge graphs, and automated decision‑making are essential for managing Alibaba’s massive big‑data environment.