How Baidu’s Archer Powers One-Click, Scalable Deployments Across Environments

Overview

Business deployment (上线) is the most common operation type, involving online code changes and configuration file changes. Manual deployment is repetitive and error‑prone; Google reports 70% of production incidents are triggered by deployment changes. Reducing human error and providing a flexible, stable deployment system is a key challenge.

Basic Introduction

In the wave of operational automation, Baidu’s Noah platform released the one‑click deployment system Archer. Archer offers a complete, portable release solution covering machine initialization, service deployment, module monitoring, CT tasks, and dynamic data file distribution. It provides a command‑line tool as the entry point, making integration easy, and can serve as a lower‑level toolchain for PaaS platforms.

General Scenarios

Different business lines have varied package specifications, languages, and frameworks, leading to inconsistent deployment strategies.

Support hierarchical releases to intercept online faults promptly.

Multi‑region deployment.

Various network environments and large‑package deployments.

Increase automation efficiency and integrate with test‑release pipelines.

Service Architecture

The system consists of a command‑line tool, web service, relay service, and single‑machine agent with deployment plugins (see Figure 2). Users trigger a change via the CLI; the web side parses parameters and dispatches tasks; agents poll for tasks, invoke plugins, and handle large packages and cross‑network transfers via the relay.

Solutions

Inconsistent package specifications across business lines

Archer defines a flexible yet complete package specification; configuration file paths, service start/stop scripts, and operational commands follow a fixed standard while supporting customization, ensuring a unified package structure.

Hierarchical release and fault interception

Archer supports serial/parallel releases and pause points, allowing pauses, resumes, retries, or rollbacks at instance, data‑center, or region levels.

Multi‑region deployment

Archer offers configuration derivation, enabling different configuration values per region within a single file.

Various network environments and large‑package deployment

Archer uses a data‑relay transfer; tasks first store data on a relay machine, then target machines pull files, allowing cross‑segment transfers.

Automation efficiency and CI/CD integration

The CLI can be integrated into scripts and platforms. Archer also supports customizable single‑machine workflows via a YAML “operation_list” file, allowing users to compose common operations such as start/stop monitoring, service control, data transfer, command execution, and scheduled tasks.

Other Design Points

Each deployment is parsed by the web controller and dispatched to target machines. The agent downloads the latest plugin MD5 from a plugin cluster, updating only the plugin cluster to achieve horizontal scalability and low upgrade cost.

Summary

Baidu’s deployment service evolved from manual → web‑based → open, and is now moving toward intelligent automation. Archer, as the open‑generation product, enjoys high internal adoption and offers valuable ideas for AIOps development.