🏗️ Argo Master Guide: A Complete Analysis of 4 Templates (From ClusterAnalysis to ContainerSet)

Hello! This guide is prepared for those who want to build an advanced cloud-native environment using Argo Workflows and Argo Rollouts.

The Argo ecosystem has various Templates. Each template defines a step in a workflow or verifies the safety of a deployment, performing its unique role. Today, we will delve into four of these templates that are critically important in practice. 🚀

If you only knew Argo as a “container execution tool,” the templates introduced today will broaden your perspective. Shall we explore automated analysis, data processing, and complex container execution one by one? 💡

---

1. 📊 ClusterAnalysisTemplate: A Smart Validator at the Cluster Level

ClusterAnalysisTemplate is a powerful template primarily used in Argo Rollouts. As its name suggests, it defines reusable analysis logic at the ‘cluster-wide’ level.

🧐 Key Features

• Reusability: Can be shared across multiple Rollout resources throughout the cluster, not tied to a specific namespace.
• Metric-based: Queries data from external monitoring systems like Prometheus, Datadog, and New Relic to determine the success of a deployment.
• Automatic Rollback: If the analysis result is deemed a failure (Error/Failure), Argo Rollouts automatically stops the deployment and rolls back to the previous version.

💻 YAML Example and Explanation

YAML

apiVersion: argoproj.io/v1alpha1
kind: ClusterAnalysisTemplate
metadata:
  name: global-error-rate-check  # Name to be called from anywhere in the cluster
spec:
  metrics:
  - name: error-rate
    interval: 1m             # Check metrics every 1 minute
    successCondition: result[0] < 0.01  # Success if error rate is less than 1%
    failureLimit: 3          # Final failure if it fails more than 3 times
    provider:
      prometheus:
        address: http://prometheus.monitoring.svc.cluster.local
        query: |
          sum(rate(http_requests_total{status=~"5.*"}[1m])) 
          / 
          sum(rate(http_requests_total[1m]))

Tip: Unlike AnalysisTemplate, this template with the Cluster prefix is very useful for creating common monitoring rules with administrator privileges.

---

2. 🔢 Data Template: The Core of Data-Driven Decision Making

In Argo Workflows, a Data Template is used to read data from external sources (S3, HTTP artifacts, etc.) and utilize it as variables within a workflow or to determine control flow.

🧐 Key Features

• Dynamic Generation: By reading data and combining it with withParam, you can dynamically loop as many times as there are data items.
• Filtering: Excellent at extracting specific values (using JSONPath, etc.) from read JSON data.

💻 YAML Example and Explanation

YAML

- name: data-fetch-example
  data:
    source:
      artifact:
        s3:
          endpoint: s3.amazonaws.com
          bucket: my-bucket
          key: user_list.json  # File containing data to be analyzed
    transformation:
      # Extract only the IDs of 'active' users from JSON data as a list
      jsonPath: "{$.users[?(@.status == 'active')].id}"

---

3. 📜 Script Template: Python, Bash, Anything You Want

The Script Template is one of the most popular and powerful templates. It is similar to a general Container template, but differs in that you can write the source code directly (Inline) within the YAML.

🧐 Key Features

• Simple Logic Implementation: You don’t need to build a separate Docker image every time; you can load a standard image and modify only the internal script to execute the logic.
• Result Capture: It’s very easy to save the standard output (Stdout) of a script as a variable and pass it to the next step.

💻 YAML Example and Explanation

YAML

- name: generate-report-script
  script:
    image: python:3.9-slim    # Execution environment image
    command: [python]        # Interpreter to use
    source: |                # Source code to execute (Inline)
      import json
      import sys
      
      # Perform complex calculation logic
      data = {"status": "success", "score": 95}
      
      # If the result is output to Stdout, Argo captures it as a return value
      print(json.dumps(data))

Caution: If the logic becomes too long, YAML readability decreases, so it’s better to include complex code in an image. ⚠️

---

4. 📦 Container Set Template (Graph): Orchestration of Composite Containers

The Container Set Template allows you to run multiple containers within a single Pod and define their dependencies.

🧐 Key Features

• More than just a Sidecar: Beyond simply running together, you can create a ‘Pod internal workflow’ where container B runs only after container A finishes.
• Resource Efficiency: Processing multiple tasks on a single Pod node results in lower network latency and efficient resource management.

💻 YAML Example and Explanation

YAML

- name: complex-job-set
  containerSet:
    containers:
    - name: setup
      image: alpine
      command: [sh, -c, "echo 'Initializing...' > /shared/init.txt"]
      volumeMounts:
      - name: workdir
        mountPath: /shared
    
    - name: main-process
      image: my-app:latest
      dependencies: [setup]  # Runs only if the setup container succeeds
      volumeMounts:
      - name: workdir
        mountPath: /shared
        
    - name: reporter
      image: curlimages/curl
      dependencies: [main-process] # Runs after main-process
      command: [curl, -X, POST, "http://notify.me"]

---

💡 Summary and Selection Guide

Template Type	Primary Use Case	One-liner
—	—	—
ClusterAnalysis	Deployment stability verification (Rollouts)	“Company-wide common deployment acceptance criteria”
Data	Dynamic data processing and loop generation	“A brain that moves according to data”
Script	Simple logic and script execution	“A Swiss Army knife used immediately without building”
Container Set	Control multiple containers within a Pod	“Collaboration of multiple workers under one roof”

—

🏁 Conclusion

The Argo ecosystem is vast, but by understanding and combining the characteristics of these four templates, you can implement true GitOps and workflow automation. I recommend starting with the Script Template and gradually expanding to data-driven Data Templates or ClusterAnalysisTemplates for cluster-wide common analysis! 🌟