gpunnoli-2026/argus-aiops

# Argus **AIOps 事件预测与自动化响应平台** Argus 是一个端到端的 AIOps 平台，它接收来自 Kubernetes 微服务应用的基础设施遥测数据，应用 ML 来检测异常、预测容量 耗尽，并将告警风暴关联分类为事件 —— 然后通过 Slack 交付的 human-in-the-loop、审批门控自动修复来闭环。 ## 状态 🚧 **正在积极开发中。** | 阶段 | 范围 | 状态 | |---|---|---| | 0 | 架构、repo 脚手架、Terraform/EKS 基础 | 🚧 进行中 | | 1 | 遥测与故障实验室（Prometheus, Chaos Mesh, k6） | 📋 已规划 | | 2 | 异常检测（IsolationForest + MLflow） | 📋 已规划 | | 3 | 容量预测（Prophet）与告警关联 | 📋 已规划 | | 4 | Slack 事件工作流 + 门控修复 | 📋 已规划 | | 5 | MLOps 强化（重训练、漂移门控、CI/CD） | 📋 已规划 | | 6 | 多云可移植性与优化 | 📋 已规划 | ## 架构 有关完整的高层级和详细架构，请参阅 [docs/architecture.md](docs/architecture.md)，有关分阶段的构建计划，请参阅 [docs/plan.md](docs/plan.md)。 ``` Chaos fault injected → Online Boutique degrades → Prometheus metrics / Alertmanager alerts → ML services: anomaly score, capacity forecast, alert correlation → One classified incident posted to Slack with recommended runbook → [Approve] → RBAC-scoped remediation (scale/restart/rollback), audited → Grafana shows recovery ``` ## 技术栈 Kubernetes (EKS) · Terraform · Helm · Prometheus/Alertmanager/Grafana · Chaos Mesh · k6 · Python · scikit-learn · Prophet · MLflow · Evidently · FastAPI · Slack (Socket Mode) · GitHub Actions ## 快速开始 ``` make up # provision EKS + deploy platform make demo # inject fault, watch the incident flow make down # tear everything down (always run this) ``` ## 仓库结构 ``` terraform/ Infrastructure as code (aws/ now; gcp/, azure/ planned) helm/ Platform umbrella chart + per-target values services/ FastAPI microservices (detection, correlation, orchestration, remediation) ml/ Training pipelines, evaluation, drift checks chaos/ Chaos Mesh experiment library (labeled ground truth) loadgen/ k6 load profiles observability/ Dashboards, recording & alerting rules docs/ Architecture, build plan, runbooks, design decisions ``` ## License [Apache-2.0](LICENSE)

标签：AIOps, Apex, API集成, 可观测性, 异常检测, 智能运维, 机器学习, 自动修复, 自定义请求头, 逆向工具