HotStandbyJobPython

A lightweight Python operator and CRD that provisions hot standby Kubernetes Jobs—keeping a configurable number of pre-spawned pods “warm” to reduce cold-start latency when real workload arrives.

TL;DR: Apply the CRD, deploy the operator, then create a HotStandbyJob custom resource. The operator maintains standby pods and spins up real Jobs instantly when needed.

---

✨ Features

• Hot Standby Pods: Keep N pods ready to serve to minimize first-request latency.
• On-Demand Job Launch: Converts standby capacity into real Jobs when signals/conditions are met.
• Auto Scale-Down: Optional delay window before tearing down surplus pods to avoid thrash.
• Parallelism Variant: An alternative operator path supporting parallel job launches.
• K8s-Native UX: Define behavior declaratively via a CRD and manage with kubectl.

---

📦 Repository Layout

.
├─ operate_hsj.py                 # Main operator/controller (hot-standby logic)
├─ operate_hsj_parallelism.py     # Variant focusing on parallel job orchestration
├─ crd.yaml                       # CustomResourceDefinition for HotStandbyJob
├─ example.yaml                   # Example HotStandbyJob resource
├─ deploy-operate.yaml            # Deployment/Pod for running the operator in-cluster
├─ rbac.yaml                      # ClusterRole/Role + RoleBinding for operator
├─ sa-rbac.yaml                   # ServiceAccount + RBAC combo (alt. manifest)
├─ requirements.txt               # Python dependencies
└─ Dockerfile                     # Container image for operator

（以上檔名出自專案首頁檔案清單。）

---

🚀 Quick Start

Prerequisites

• Kubernetes 1.22+（建議 1.25+）
• kubectl 已連到你的叢集 -（可選）本機開發：Python 3.10+，能安裝 requirements.txt

1) 安裝 CRD

kubectl apply -f crd.yaml

2) 佈署 Operator（叢集內執行）

# RBAC / ServiceAccount
kubectl apply -f sa-rbac.yaml

# 部署 operator
kubectl apply -f deploy-operate.yaml

如果你要以容器執行，deploy-operate.yaml 會參考由 Dockerfile 打好的 image。請依你自己的 registry 調整。

3) 建立範例 HotStandbyJob

kubectl apply -f example.yaml
kubectl get hotstandbyjobs
kubectl describe hotstandbyjob <name>

（選擇）本機執行 Operator

需要先安裝 CRD 不過不需要 RBAC 或 SA

若要在本機直連叢集測試：

pip install -r requirements.txt
kopf run -A operate_hsj_parallelism.py
# 或針對平行版：
# python operate_hsj_parallelism.py

---

🧩 Custom Resource（CRD）概念與欄位

以下為預期欄位說明，實際以 crd.yaml 與 example.yaml 為準。

HotStandbyJob（namespaced）常見欄位示意：

apiVersion: apps.paia.tech/v1alpha1
kind: HotStandbyJob
metadata:
  name: demo-hsj
spec:
  idleTarget: 20 # 預備的 Pod 數量
  minReplicas: 0
  maxReplicas: 100

  # 你的工作負載模板（通常類似 Job 的模板）
  jobTemplate:
    image: busybox
    command: ["sh", "-c", "echo hello && sleep 5"]
    parallelism: 1
    backoffLimit: 0
    ttlSecondsAfterFinished: 60

  busyProbe:
    mode: exec
    exec:
      command: [ "cat", "/tmp/busy_state" ]
      timeoutSeconds: 1
      successIsBusy: true

---

🛠️ 開發與建置

Build Image

docker build -t <your-registry>/hotstandbyjob-operator:latest .
docker push <your-registry>/hotstandbyjob-operator:latest
kubectl apply -f deploy-operate.yaml

本機測試

kopf run -A operate_hsj_parallelism.py
kubectl get pods,job,hotstandbyjob -w

---

🧪 驗證與觀察

kubectl get hotstandbyjob
kubectl get pods -w
kubectl get jobs -w
kubectl describe hotstandbyjob <name>

---

🧹 清除

kubectl delete -f example.yaml
kubectl delete -f deploy-operate.yaml
kubectl delete -f sa-rbac.yaml
kubectl delete -f crd.yaml