Nerd Level Tech logo
|
Lesson 22 of 24

GitOps & Deployment Strategies

ArgoCD for Model Deployment

5 min read

English Content

What is ArgoCD?

ArgoCD is a declarative, GitOps continuous delivery tool for Kubernetes. It continuously monitors Git repositories and automatically applies changes to your cluster, ensuring your deployed applications match the desired state in Git.

Key features for ML:

  • Declarative model deployment
  • Automatic sync and self-healing
  • Rollback with Git history
  • Multi-cluster support
  • Web UI for visibility

Installing ArgoCD

# Create namespace
kubectl create namespace argocd

# Install ArgoCD
kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

# Access the UI (port forward)
kubectl port-forward svc/argocd-server -n argocd 8080:443

# Get initial admin password
kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d

# Login with CLI
argocd login localhost:8080

Creating an Application

Define an ArgoCD Application for your model:

# apps/model-server-app.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server
  namespace: argocd
  finalizers:
    - resources-finalizer.argocd.argoproj.io
spec:
  project: default

  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/production

  destination:
    server: https://kubernetes.default.svc
    namespace: ml-production

  syncPolicy:
    automated:
      prune: true
      selfHeal: true
    syncOptions:
      - CreateNamespace=true

  # Health checks for ML workloads
  ignoreDifferences:
    - group: apps
      kind: Deployment
      jsonPointers:
        - /spec/replicas  # Ignore HPA-managed replicas

Sync Strategies

Manual Sync: Review before applying

syncPolicy: {}  # No automated sync

Automated Sync: Apply changes automatically

syncPolicy:
  automated:
    prune: true      # Delete removed resources
    selfHeal: true   # Revert manual changes
    allowEmpty: false

Sync with Waves: Control deployment order

# In your Kubernetes manifests
metadata:
  annotations:
    argocd.argoproj.io/sync-wave: "1"  # Deploy order

Multi-Environment Setup

# apps/staging.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server-staging
  namespace: argocd
spec:
  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/staging
  destination:
    server: https://kubernetes.default.svc
    namespace: ml-staging
  syncPolicy:
    automated:
      selfHeal: true
      prune: true

# apps/production.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server-production
  namespace: argocd
spec:
  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/production
  destination:
    server: https://kubernetes.default.svc
    namespace: ml-production
  syncPolicy:
    automated:
      selfHeal: true
      prune: true

ApplicationSet for Multiple Models

Deploy multiple models with a single template:

# apps/model-appset.yaml
apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata:
  name: ml-models
  namespace: argocd
spec:
  generators:
    - list:
        elements:
          - model: sentiment-classifier
            version: v1.2.0
            replicas: "3"
          - model: fraud-detector
            version: v2.1.0
            replicas: "5"
          - model: recommendation-engine
            version: v1.0.0
            replicas: "2"
  template:
    metadata:
      name: '{{model}}'
    spec:
      project: default
      source:
        repoURL: https://github.com/myorg/ml-deployments.git
        targetRevision: main
        path: models/{{model}}
        helm:
          parameters:
            - name: image.tag
              value: '{{version}}'
            - name: replicas
              value: '{{replicas}}'
      destination:
        server: https://kubernetes.default.svc
        namespace: ml-production
      syncPolicy:
        automated:
          selfHeal: true

Health Checks for ML Services

# Custom health check for model server
apiVersion: v1
kind: ConfigMap
metadata:
  name: argocd-cm
  namespace: argocd
data:
  resource.customizations.health.argoproj.io_Application: |
    hs = {}
    hs.status = "Progressing"
    hs.message = ""
    if obj.status ~= nil then
      if obj.status.health ~= nil then
        hs.status = obj.status.health.status
        if obj.status.health.message ~= nil then
          hs.message = obj.status.health.message
        end
      end
    end
    return hs

Rollback Procedure

# View application history
argocd app history model-server

# Rollback to specific revision
argocd app rollback model-server 5

# Or via Git
git revert HEAD
git push origin main
# ArgoCD automatically syncs to reverted state

Key Takeaways

ArgoCD Feature ML Use Case
Application CRD Define model deployment
Automated sync Deploy on Git changes
Self-heal Prevent configuration drift
ApplicationSet Multi-model deployments
Sync waves Control deployment order

المحتوى العربي

ما هو ArgoCD؟

ArgoCD هو أداة تسليم مستمر تصريحية GitOps لـ Kubernetes. يراقب باستمرار مستودعات Git ويطبق التغييرات تلقائياً على الكلستر الخاص بك، مما يضمن تطابق تطبيقاتك المنشورة مع الحالة المطلوبة في Git.

الميزات الرئيسية لـ ML:

  • نشر النماذج التصريحي
  • المزامنة التلقائية والشفاء الذاتي
  • التراجع مع تاريخ Git
  • دعم الكلسترات المتعددة
  • واجهة ويب للرؤية

تثبيت ArgoCD

# إنشاء namespace
kubectl create namespace argocd

# تثبيت ArgoCD
kubectl apply -n argocd -f https://raw.githubusercontent.com/argoproj/argo-cd/stable/manifests/install.yaml

# الوصول إلى الواجهة (تحويل المنفذ)
kubectl port-forward svc/argocd-server -n argocd 8080:443

# الحصول على كلمة مرور المشرف الأولية
kubectl -n argocd get secret argocd-initial-admin-secret -o jsonpath="{.data.password}" | base64 -d

# تسجيل الدخول مع CLI
argocd login localhost:8080

إنشاء تطبيق

عرّف تطبيق ArgoCD لنموذجك:

# apps/model-server-app.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server
  namespace: argocd
  finalizers:
    - resources-finalizer.argocd.argoproj.io
spec:
  project: default

  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/production

  destination:
    server: https://kubernetes.default.svc
    namespace: ml-production

  syncPolicy:
    automated:
      prune: true
      selfHeal: true
    syncOptions:
      - CreateNamespace=true

  # فحوصات الصحة لأحمال عمل ML
  ignoreDifferences:
    - group: apps
      kind: Deployment
      jsonPointers:
        - /spec/replicas  # تجاهل replicas المُدارة بـ HPA

استراتيجيات المزامنة

المزامنة اليدوية: المراجعة قبل التطبيق

syncPolicy: {}  # لا مزامنة آلية

المزامنة الآلية: تطبيق التغييرات تلقائياً

syncPolicy:
  automated:
    prune: true      # حذف الموارد المُزالة
    selfHeal: true   # التراجع عن التغييرات اليدوية
    allowEmpty: false

المزامنة مع Waves: التحكم في ترتيب النشر

# في manifests Kubernetes الخاصة بك
metadata:
  annotations:
    argocd.argoproj.io/sync-wave: "1"  # ترتيب النشر

إعداد بيئات متعددة

# apps/staging.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server-staging
  namespace: argocd
spec:
  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/staging
  destination:
    server: https://kubernetes.default.svc
    namespace: ml-staging
  syncPolicy:
    automated:
      selfHeal: true
      prune: true

# apps/production.yaml
apiVersion: argoproj.io/v1alpha1
kind: Application
metadata:
  name: model-server-production
  namespace: argocd
spec:
  source:
    repoURL: https://github.com/myorg/ml-deployments.git
    targetRevision: main
    path: overlays/production
  destination:
    server: https://kubernetes.default.svc
    namespace: ml-production
  syncPolicy:
    automated:
      selfHeal: true
      prune: true

ApplicationSet لنماذج متعددة

انشر نماذج متعددة بقالب واحد:

# apps/model-appset.yaml
apiVersion: argoproj.io/v1alpha1
kind: ApplicationSet
metadata:
  name: ml-models
  namespace: argocd
spec:
  generators:
    - list:
        elements:
          - model: sentiment-classifier
            version: v1.2.0
            replicas: "3"
          - model: fraud-detector
            version: v2.1.0
            replicas: "5"
          - model: recommendation-engine
            version: v1.0.0
            replicas: "2"
  template:
    metadata:
      name: '{{model}}'
    spec:
      project: default
      source:
        repoURL: https://github.com/myorg/ml-deployments.git
        targetRevision: main
        path: models/{{model}}
        helm:
          parameters:
            - name: image.tag
              value: '{{version}}'
            - name: replicas
              value: '{{replicas}}'
      destination:
        server: https://kubernetes.default.svc
        namespace: ml-production
      syncPolicy:
        automated:
          selfHeal: true

فحوصات الصحة لخدمات ML

# فحص صحة مخصص لخادم النموذج
apiVersion: v1
kind: ConfigMap
metadata:
  name: argocd-cm
  namespace: argocd
data:
  resource.customizations.health.argoproj.io_Application: |
    hs = {}
    hs.status = "Progressing"
    hs.message = ""
    if obj.status ~= nil then
      if obj.status.health ~= nil then
        hs.status = obj.status.health.status
        if obj.status.health.message ~= nil then
          hs.message = obj.status.health.message
        end
      end
    end
    return hs

إجراء التراجع

# عرض تاريخ التطبيق
argocd app history model-server

# التراجع إلى revision محدد
argocd app rollback model-server 5

# أو عبر Git
git revert HEAD
git push origin main
# ArgoCD يزامن تلقائياً إلى الحالة المُرجعة

النقاط الرئيسية

ميزة ArgoCD حالة استخدام ML
Application CRD تعريف نشر النموذج
المزامنة الآلية النشر عند تغييرات Git
الشفاء الذاتي منع انجراف التكوين
ApplicationSet نشر نماذج متعددة
Sync waves التحكم في ترتيب النشر

Quiz

Module 6: GitOps & Deployment Strategies

Take Quiz