Azure Kubernetes上微服务有序调用的工作流架构方案咨询

Hey there! Let's walk through your options for building an on-demand workflow to orchestrate your 10 .NET Core 2.2 microservices (hosted on Azure Kubernetes Service, AKS) in a specific execution order. I'll break down the pros and cons of each approach to help you pick what fits best.

• Why it’s a strong fit: This is the go-to for low-code/no-code workflow orchestration. You get a visual designer to map out your sequence of service calls, add conditional branches, retry logic, and error handling without writing tons of code. It has built-in connectors for HTTP (to call your microservice endpoints), MongoDB, and most Azure services—so integrating with your existing stack is straightforward.
• Best for: When your workflow logic is relatively fixed (e.g., "call Service A → wait for success → call Service B with A's output → handle failures in either step") and you want to avoid managing infrastructure. It’s easy to trigger on-demand via HTTP requests, scheduled events, or even Azure Service Bus messages.
• Gotchas: Make sure your AKS microservices are accessible to Logic Apps—either expose them via an Ingress controller, use Azure Private Link for secure internal access, or deploy a gateway if they’re in a private cluster. Also, if your services use complex auth (like Azure AD OAuth2), Logic Apps supports configuring that, but you’ll need to set up the credentials correctly.

• Why it’s a strong fit: If you need more flexibility and custom code in your workflow, this is perfect—especially since you’re already using .NET Core. Durable Functions (an extension of Azure Functions) is built specifically for stateful orchestration: it handles sequence execution, parallel tasks, retries, and even long-running workflows out of the box. You can write orchestrator functions in C# to call your microservices, transform data between steps, and implement custom business logic that Logic Apps might not support easily.
• Best for: When your workflow requires dynamic logic (e.g., "skip Service C if a certain condition from Service A is met" or "aggregate results from multiple services before moving to the next step"). It’s also cost-effective, since you only pay for execution time, and integrates seamlessly with your existing .NET codebase.
• Gotchas: You’ll need to learn Durable Functions patterns (like Orchestrator vs. Activity functions), but the learning curve is gentle for .NET developers. Similar to Logic Apps, ensure your functions can reach your AKS services—you can use VNet integration for Azure Functions to access private clusters securely.

• Why it’s a strong fit: If you want to keep your entire workflow within the Kubernetes ecosystem (no external Azure PaaS dependencies), Argo Workflows is a powerful open-source tool. It lets you define workflows as YAML, directly scheduling your microservice containers in AKS in the order you specify. It supports complex logic like loops, conditional branches, and artifact passing between steps, and integrates with Kubernetes features like secrets, config maps, and RBAC.
• Best for: When you need deep control over how your microservices are scheduled (e.g., assigning specific node pools, resource limits) or want to avoid vendor lock-in. It’s also great if you already have Kubernetes expertise in your team.
• Gotchas: You’ll need to deploy and maintain Argo Workflows in your AKS cluster, which adds operational overhead. There’s a learning curve for writing Argo workflow YAML definitions, but the community has lots of examples to follow.

• Pick Azure Logic Apps if you want to build and deploy workflows fast with minimal code.
• Pick Durable Functions if you need custom .NET logic and flexible orchestration.
• Pick Argo Workflows if you want full Kubernetes-native control and no external PaaS dependencies.

• Ensure all your microservices are idempotent: this means calling them multiple times won’t cause unintended side effects (critical for retry logic in workflows).
• Set up monitoring: Use Azure Monitor for Logic Apps/Functions, or Prometheus + Grafana for Argo, to track workflow execution status, errors, and performance.
• Secure access: Use Azure AD for authenticating workflow-to-service calls, or a service mesh like Istio in AKS to manage service-to-service auth and traffic control.

内容的提问来源于stack exchange，提问作者Amit Singh