Update master.yml

.github/workflows/master.yml

@@ -1,11 +1,27 @@
-# 复制到 itfx: .github/workflows/deploy.yml
-# Secrets: SSH_HOST(42), SSH_USERNAME, SSH_PASSWORD
-# 集群: ./scripts/16-setup-kaniko.sh
+# 复制到 itfx 仓库: .github/workflows/deploy.yml
+#
+# 集群一次性准备（在 42 上）:
+#   GIT_TOKEN=... TCR_USERNAME=... TCR_PASSWORD=... ./scripts/16-setup-kaniko.sh
+#
+# GitHub Secrets（本仓 Settings → Secrets）:
+#   SSH_K3S_HOST      — 42 公网 IP
+#   SSH_K3S_USERNAME  — 一般 root
+#   SSH_K3S_PASSWORD  — SSH 密码
+#
+# 排障：看哪个 step 红了
+#   1 print-info     — 仅输出变量，几乎不会挂
+#   2 check-cluster  — SSH / kubeconfig / 节点不通
+#   3 check-secrets  — 未跑 16-setup-kaniko，或 Secret 名不对
+#   4 kaniko-build   — 拉代码 / Dockerfile / 推 TCR（看 Job 日志）
+#   5 set-image      — Deployment 名或容器名不对
+#   6 rollout        — 镜像拉不下来 / 探针失败 / 资源不够
+#
 name: Build and Deploy
 
 on:
   push:
     branches: [master]
+  workflow_dispatch:
 
 env:
   GIT_URL: https://github.com/xiaoliutral/itfx.git
@@ -15,31 +31,167 @@ env:
   CONTAINER_NAME: interfaceforward-api
   TCR_REGISTRY: ccr.ccs.tencentyun.com
   TCR_NAMESPACE: ideship
+  K3S_DIR: /data/k3s
+  KUBECONFIG: /etc/rancher/k3s/k3s.yaml
+  NAMESPACE: prod
 
 jobs:
   deploy:
     runs-on: ubuntu-latest
     steps:
-      - name: Kaniko build + deploy
+      # ── 0. 本机算出 TAG / 完整镜像名（不 SSH）────────────────
+      - name: "0. print build info"
+        id: info
+        run: |
+          TAG="${{ github.sha }}"
+          IMG="${{ env.TCR_REGISTRY }}/${{ env.TCR_NAMESPACE }}/${{ env.IMAGE_NAME }}:${TAG}"
+          echo "tag=${TAG}" >> "$GITHUB_OUTPUT"
+          echo "image=${IMG}" >> "$GITHUB_OUTPUT"
+          echo "======= build info ======="
+          echo "repo    = ${{ github.repository }}"
+          echo "ref     = ${{ github.ref_name }}"
+          echo "sha     = ${TAG}"
+          echo "git_url = ${{ env.GIT_URL }}"
+          echo "dockerfile = ${{ env.DOCKERFILE }}"
+          echo "image   = ${IMG}"
+          echo "deploy  = ${{ env.NAMESPACE }}/deploy/${{ env.DEPLOY_NAME }}"
+          echo "=========================="
+
+      # ── 1. SSH + 集群是否通 ─────────────────────────────────
+      - name: "1. check cluster (SSH + kubectl)"
         uses: appleboy/ssh-action@v1.0.3
         with:
           host: ${{ secrets.SSH_K3S_HOST }}
-          username: ${{ secrets.SSH_K3S_HOST }}
-          password: ${{ secrets.SSH_K3S_HOST }}
+          username: ${{ secrets.SSH_K3S_USERNAME }}
+          password: ${{ secrets.SSH_K3S_PASSWORD }}
+          script_stop: true
           script: |
-            set -e
-            export KUBECONFIG=/etc/rancher/k3s/k3s.yaml
-            cd /data/k3s
-            TAG="${{ github.sha }}"
-            IMG="${{ env.TCR_REGISTRY }}/${{ env.TCR_NAMESPACE }}/${{ env.IMAGE_NAME }}:${TAG}"
+            set -euo pipefail
+            export KUBECONFIG="${{ env.KUBECONFIG }}"
+            echo "==> whoami: $(whoami) @ $(hostname)"
+            echo "==> kubeconfig: ${KUBECONFIG}"
+            test -f "${KUBECONFIG}" || { echo "FAIL: 找不到 ${KUBECONFIG}"; exit 1; }
+            test -d "${{ env.K3S_DIR }}" || { echo "FAIL: 找不到 ${{ env.K3S_DIR }}，请把 k3s 清单同步到该目录"; exit 1; }
+            test -x "${{ env.K3S_DIR }}/scripts/ci-kaniko-build.sh" || {
+              echo "FAIL: 缺少可执行脚本 ${{ env.K3S_DIR }}/scripts/ci-kaniko-build.sh"
+              ls -la "${{ env.K3S_DIR }}/scripts/" || true
+              exit 1
+            }
+            echo "==> nodes:"
+            kubectl get nodes -o wide
+            echo "==> namespace ${{ env.NAMESPACE }}:"
+            kubectl get ns "${{ env.NAMESPACE }}"
+            echo "==> deploy/${{ env.DEPLOY_NAME }}:"
+            kubectl -n "${{ env.NAMESPACE }}" get deploy "${{ env.DEPLOY_NAME }}" -o wide
 
+      # ── 2. Kaniko 凭据是否已创建 ────────────────────────────
+      - name: "2. check kaniko secrets"
+        uses: appleboy/ssh-action@v1.0.3
+        with:
+          host: ${{ secrets.SSH_K3S_HOST }}
+          username: ${{ secrets.SSH_K3S_USERNAME }}
+          password: ${{ secrets.SSH_K3S_PASSWORD }}
+          script_stop: true
+          script: |
+            set -euo pipefail
+            export KUBECONFIG="${{ env.KUBECONFIG }}"
+            NS="${{ env.NAMESPACE }}"
+            echo "==> 检查 Secret（由 16-setup-kaniko.sh 创建）"
+            kubectl -n "$NS" get secret kaniko-git -o jsonpath='{.metadata.name}{" keys="}{.data}{"\n"}' >/dev/null \
+              && echo "OK secret/kaniko-git" \
+              || { echo "FAIL: 缺少 secret/kaniko-git — 在 42 执行: GIT_TOKEN=... ./scripts/16-setup-kaniko.sh"; exit 1; }
+            kubectl -n "$NS" get secret kaniko-docker-config >/dev/null \
+              && echo "OK secret/kaniko-docker-config" \
+              || { echo "FAIL: 缺少 secret/kaniko-docker-config — TCR 账号未写入"; exit 1; }
+            echo "==> build 节点:"
+            kubectl get nodes -l role=build -o wide
+            COUNT="$(kubectl get nodes -l role=build --no-headers 2>/dev/null | wc -l | tr -d ' ')"
+            if [[ "${COUNT}" == "0" ]]; then
+              echo "FAIL: 没有 role=build 节点，Kaniko Job 会 Pending"
+              kubectl get nodes --show-labels
+              exit 1
+            fi
+
+      # ── 3. Kaniko 构建并推 TCR（最慢，失败看 Job 日志）─────
+      - name: "3. kaniko build & push"
+        uses: appleboy/ssh-action@v1.0.3
+        with:
+          host: ${{ secrets.SSH_K3S_HOST }}
+          username: ${{ secrets.SSH_K3S_USERNAME }}
+          password: ${{ secrets.SSH_K3S_PASSWORD }}
+          command_timeout: 50m
+          script_stop: true
+          script: |
+            set -euo pipefail
+            export KUBECONFIG="${{ env.KUBECONFIG }}"
+            cd "${{ env.K3S_DIR }}"
+            TAG="${{ steps.info.outputs.tag }}"
+            echo "==> 开始 Kaniko: ${{ env.IMAGE_NAME }}:${TAG}"
+            echo "    失败时在 42 执行:"
+            echo "    kubectl -n ${{ env.NAMESPACE }} get job -l app=kaniko"
+            echo "    kubectl -n ${{ env.NAMESPACE }} logs -l job-name=... -c git"
+            echo "    kubectl -n ${{ env.NAMESPACE }} logs -l job-name=... -c kaniko"
             ./scripts/ci-kaniko-build.sh \
               "${{ env.GIT_URL }}" \
               "${{ env.DOCKERFILE }}" \
               "${{ env.IMAGE_NAME }}" \
               "${TAG}" \
               "${TAG}"
+            echo "==> 镜像已推送: ${{ steps.info.outputs.image }}"
 
-            kubectl -n prod set image "deploy/${{ env.DEPLOY_NAME }}" \
+      # ── 4. 改 Deployment 镜像 ───────────────────────────────
+      - name: "4. set deployment image"
+        uses: appleboy/ssh-action@v1.0.3
+        with:
+          host: ${{ secrets.SSH_K3S_HOST }}
+          username: ${{ secrets.SSH_K3S_USERNAME }}
+          password: ${{ secrets.SSH_K3S_PASSWORD }}
+          script_stop: true
+          script: |
+            set -euo pipefail
+            export KUBECONFIG="${{ env.KUBECONFIG }}"
+            NS="${{ env.NAMESPACE }}"
+            IMG="${{ steps.info.outputs.image }}"
+            echo "==> 当前镜像:"
+            kubectl -n "$NS" get deploy "${{ env.DEPLOY_NAME }}" \
+              -o jsonpath='{.spec.template.spec.containers[?(@.name=="${{ env.CONTAINER_NAME }}")].image}{"\n"}'
+            echo "==> 设为: ${IMG}"
+            kubectl -n "$NS" set image "deploy/${{ env.DEPLOY_NAME }}" \
               "${{ env.CONTAINER_NAME }}=${IMG}"
-            kubectl -n prod rollout status "deploy/${{ env.DEPLOY_NAME }}" --timeout=180s
+            echo "==> 已更新:"
+            kubectl -n "$NS" get deploy "${{ env.DEPLOY_NAME }}" \
+              -o jsonpath='{.spec.template.spec.containers[?(@.name=="${{ env.CONTAINER_NAME }}")].image}{"\n"}'
+
+      # ── 5. 等待滚动发布 ─────────────────────────────────────
+      - name: "5. wait rollout"
+        uses: appleboy/ssh-action@v1.0.3
+        with:
+          host: ${{ secrets.SSH_K3S_HOST }}
+          username: ${{ secrets.SSH_K3S_USERNAME }}
+          password: ${{ secrets.SSH_K3S_PASSWORD }}
+          command_timeout: 5m
+          script_stop: true
+          script: |
+            set -euo pipefail
+            export KUBECONFIG="${{ env.KUBECONFIG }}"
+            NS="${{ env.NAMESPACE }}"
+            DEP="${{ env.DEPLOY_NAME }}"
+            echo "==> rollout status..."
+            if ! kubectl -n "$NS" rollout status "deploy/${DEP}" --timeout=180s; then
+              echo "FAIL: rollout 超时/失败，诊断信息如下"
+              kubectl -n "$NS" get deploy,rs,pods -l app="${DEP}" -o wide || \
+                kubectl -n "$NS" get pods -o wide | grep -E "${DEP}|NAME" || true
+              kubectl -n "$NS" describe deploy "${DEP}" | tail -60 || true
+              POD="$(kubectl -n "$NS" get pods -l app="${DEP}" -o jsonpath='{.items[0].metadata.name}' 2>/dev/null || true)"
+              if [[ -n "${POD}" ]]; then
+                echo "==> events for ${POD}:"
+                kubectl -n "$NS" describe pod "${POD}" | tail -40 || true
+                echo "==> logs:"
+                kubectl -n "$NS" logs "${POD}" --tail=80 || true
+              fi
+              exit 1
+            fi
+            echo "==> pods:"
+            kubectl -n "$NS" get pods -o wide | grep -E "${DEP}|NAME" || \
+              kubectl -n "$NS" get pods -o wide
+            echo "OK deployed ${{ steps.info.outputs.image }}"