Sync: add ai finops and deployment notes
This commit is contained in:
31
wiki/concepts/DarkLaunching.md
Normal file
31
wiki/concepts/DarkLaunching.md
Normal file
@@ -0,0 +1,31 @@
|
||||
---
|
||||
title: "DarkLaunching"
|
||||
type: concept
|
||||
tags: ["deployment", "release-management", "feature-rollout"]
|
||||
sources: ["engineering-autonomous-optimization-architect"]
|
||||
last_updated: 2026-04-26
|
||||
---
|
||||
|
||||
## Aliases
|
||||
- Dark Launch
|
||||
- 暗启动
|
||||
- 灰度发布
|
||||
- Feature Flag Deployment
|
||||
|
||||
## Definition
|
||||
暗启动是 [[AutonomousOptimizationArchitect]] 的模型引入策略——在不完全暴露给用户的前提下,将新模型部署到生产环境,通过 [[ShadowTraffic]] 验证其性能。分为三个阶段:影子测试(不返回用户)→ 灰度流量(5% 用户)→ 全量切换。
|
||||
|
||||
## Mechanism
|
||||
1. **Phase 1 - Shadow Deployment**:新模型接收影子流量,完全不影响用户
|
||||
2. **Phase 2 - Canary**:5% 真实流量切换到新模型,监控错误率和用户满意度
|
||||
3. **Phase 3 - Full Rollout**:新模型通过所有检查后,全量替换旧模型
|
||||
|
||||
## Key Properties
|
||||
- **风险可控**:任何阶段发现问题均可立即回滚
|
||||
- **数据驱动**:每个阶段都有明确的量化指标门槛
|
||||
- **与 CI/CD 集成**:暗启动可作为自动化发布流水线的组成部分
|
||||
|
||||
## Connections
|
||||
- [[AutonomousOptimizationArchitect]] — 使用暗启动作为新模型引入框架
|
||||
- [[ShadowTraffic]] — 暗启动 Phase 1 的核心实现方式
|
||||
- [[CircuitBreaker]] — 提供暗启动失败时的自动保护机制
|
||||
Reference in New Issue
Block a user