ci: 保留 kind 失败现场

.gitea/workflows/vm-kind-smoke.yml

@@ -1,5 +1,7 @@
+---
 name: VM kind smoke
 
+# yamllint disable-line rule:truthy
 on:
   workflow_dispatch:
 
@@ -15,17 +17,51 @@ jobs:
         run: |
           set -euo pipefail
           version=v0.33.0
+          base_url="https://kind.sigs.k8s.io/dl/${version}"
           curl --fail --location --silent --show-error \
-            --output /tmp/kind "https://kind.sigs.k8s.io/dl/${version}/kind-linux-amd64"
+            --output /tmp/kind "${base_url}/kind-linux-amd64"
           curl --fail --location --silent --show-error \
-            --output /tmp/kind.sha256sum "https://kind.sigs.k8s.io/dl/${version}/kind-linux-amd64.sha256sum"
+            --output /tmp/kind.sha256sum \
-          printf '%s  %s\n' "$(cut -d ' ' -f1 /tmp/kind.sha256sum)" /tmp/kind | sha256sum --check
+            "${base_url}/kind-linux-amd64.sha256sum"
+          checksum=$(cut -d ' ' -f1 /tmp/kind.sha256sum)
+          printf '%s  %s\n' "$checksum" /tmp/kind | sha256sum --check
           chmod 0755 /tmp/kind
 
       - name: Create and delete kind cluster
         shell: bash
         run: |
           set -euo pipefail
-          trap '/tmp/kind delete cluster --name smoke' EXIT
+          diagnose_and_cleanup() {
-          /tmp/kind create cluster --name smoke --wait 180s
+            status=$?
+            node=smoke-control-plane
+            if (( status != 0 )) && docker inspect "$node" >/dev/null 2>&1; then
+              echo '::group::kind node inspect'
+              docker inspect "$node"
+              echo '::endgroup::'
+              echo '::group::kind node logs'
+              docker logs "$node" 2>&1 || true
+              echo '::endgroup::'
+              echo '::group::kind node guest state'
+              docker exec "$node" bash -c '
+                set +e
+                echo "### pid 1"
+                ps -p 1 -o pid,ppid,user,stat,comm,args
+                cat /proc/1/status
+                echo "### cgroup"
+                cat /proc/1/cgroup
+                findmnt -R /sys/fs/cgroup
+                stat -fc "%T %a" /sys/fs/cgroup
+                echo "### systemd"
+                systemctl --no-pager --failed
+                systemctl --no-pager status \
+                  multi-user.target containerd.service kubelet.service
+                journalctl --no-pager -b -n 500
+              ' 2>&1 || true
+              echo '::endgroup::'
+            fi
+            /tmp/kind delete cluster --name smoke || true
+            exit "$status"
+          }
+          trap diagnose_and_cleanup EXIT
+          /tmp/kind create cluster --name smoke --wait 180s --retain
           /tmp/kind get clusters | grep -Fx smoke