feat: 隔离 VM 集成测试标签

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

@@ -5,7 +5,7 @@ on:
 
 jobs:
   kind:
-    runs-on: [self-hosted, vm]
+    runs-on: [self-hosted, vm-dev]
     steps:
       - name: Verify Docker
         run: docker info

README.md

@@ -15,6 +15,10 @@ runs-on: [self-hosted, vm]
 只执行一个 job，并在 job 结束后连同本地状态一起销毁。完整的设计约束见
 [`docs/design-principles.md`](docs/design-principles.md)。
 
+集成期间可将 `VM_RUNNER_LABEL=vm-dev`，只接取显式使用
+`runs-on: [self-hosted, vm-dev]` 的测试任务；生产 `vm` job 将保持在 Gitea pending，
+不会在 backend 修复过程中继续涌入。
+
 目标 Go controller 组件：
 
 - `scheduler`：以常驻 Gitea RunnerService 身份直接领取 task，并把完整 assignment

cmd/gitea-dynamic-runner/controller.go

@@ -49,7 +49,7 @@ type controllerConfig struct {
 	PodNamespace, PodImage, PodServiceAccount, SPIRECluster, SPIREClass string
 	SPIREAgentID                                                        string
 	PodExecutorUID, PodCapacity                                         int
-	OpenSandboxURL, OpenSandboxAPIKey, OpenSandboxPool                  string
+	OpenSandboxURL, OpenSandboxAPIKey, OpenSandboxPool, VMRunnerLabel   string
 	VMTimeout, VMCapacity                                               int
 }
 
@@ -131,7 +131,7 @@ func runController(ctx context.Context) error {
 		labels = append(labels, string(taskassignment.BackendPod))
 	}
 	if slices.Contains(config.Components, controller.VMWorker) {
-		labels = append(labels, string(taskassignment.BackendVM))
+		labels = append(labels, config.VMRunnerLabel)
 	}
 	poller := taskscheduler.Poller{
 		Client: giteaClient,
@@ -345,7 +345,7 @@ func loadControllerConfig() (controllerConfig, error) {
 		FacadeListen: env("RUNNER_FACADE_LISTEN", ":8443"), FacadeURL: os.Getenv("RUNNER_FACADE_URL"), FacadeSPIFFEID: os.Getenv("RUNNER_FACADE_SPIFFE_ID"), CapabilityKey: []byte(capabilityKey),
 		PodNamespace: env("POD_NAMESPACE", "gitea-actions"), PodImage: os.Getenv("POD_EXECUTOR_IMAGE"), PodServiceAccount: env("POD_SERVICE_ACCOUNT", "gitea-task-executor"),
 		SPIRECluster: env("SPIRE_CLUSTER", "homelab"), SPIREClass: env("SPIRE_CLASS", "spire-mgmt-spire"), SPIREAgentID: os.Getenv("SPIRE_AGENT_ID"), PodExecutorUID: envInt("POD_EXECUTOR_UID", 2000), PodCapacity: envInt("POD_CAPACITY", 4),
-		OpenSandboxURL: os.Getenv("OPENSANDBOX_API"), OpenSandboxPool: env("OPENSANDBOX_POOL", "ci-vm"), VMTimeout: envInt("VM_TIMEOUT_SECONDS", 14400), VMCapacity: envInt("VM_CAPACITY", 1),
+		OpenSandboxURL: os.Getenv("OPENSANDBOX_API"), OpenSandboxPool: env("OPENSANDBOX_POOL", "ci-vm"), VMRunnerLabel: env("VM_RUNNER_LABEL", "vm"), VMTimeout: envInt("VM_TIMEOUT_SECONDS", 14400), VMCapacity: envInt("VM_CAPACITY", 1),
 	}
 	if config.WorkloadAPIAddr == "" || config.FacadeURL == "" || config.FacadeSPIFFEID == "" {
 		return controllerConfig{}, errors.New("SPIFFE_ENDPOINT_SOCKET, RUNNER_FACADE_URL, and RUNNER_FACADE_SPIFFE_ID are required")
@@ -357,6 +357,9 @@ func loadControllerConfig() (controllerConfig, error) {
 		return controllerConfig{}, errors.New("SPIRE_AGENT_ID is required for pod-worker")
 	}
 	if slices.Contains(selection, controller.VMWorker) {
+		if config.VMRunnerLabel != "vm" && config.VMRunnerLabel != "vm-dev" {
+			return controllerConfig{}, errors.New("VM_RUNNER_LABEL must be vm or vm-dev")
+		}
 		if config.OpenSandboxURL == "" {
 			return controllerConfig{}, errors.New("OPENSANDBOX_API is required for vm-worker")
 		}

cmd/gitea-dynamic-runner/controller_test.go

@@ -56,6 +56,7 @@ func TestLoadControllerConfigRequiresOpenSandboxSecretOnlyForVM(t *testing.T) {
 	t.Setenv("RUNNER_FACADE_URL", "https://facade:8443")
 	t.Setenv("RUNNER_FACADE_SPIFFE_ID", "spiffe://ddupan.top/controller")
 	t.Setenv("OPENSANDBOX_API", "http://opensandbox.internal")
+	t.Setenv("VM_RUNNER_LABEL", "vm-dev")
 	if _, err := loadControllerConfig(); err == nil {
 		t.Fatal("expected missing OpenSandbox API key file error")
 	}

internal/assignmentqueue/jetstream.go

@@ -167,7 +167,7 @@ func OpenConsumer(ctx context.Context, manager consumerManager, stream, subjectB
 		AckPolicy:     jetstream.AckExplicitPolicy,
 		AckWait:       5 * time.Minute,
 		MaxAckPending: capacity,
-		MaxDeliver:    20,
+		MaxDeliver:    1000,
 	})
 	if err != nil {
 		return nil, fmt.Errorf("open %s assignment consumer: %w", backend, err)

internal/assignmentqueue/jetstream_test.go

@@ -205,7 +205,7 @@ func TestOpenConsumerUsesIndependentDurablePerBackend(t *testing.T) {
 	if _, err := OpenConsumer(context.Background(), manager, "CI_RUNNER", "ci.assignment", taskassignment.BackendPod, 4); err != nil {
 		t.Fatal(err)
 	}
-	if manager.config.Durable != "pod" || manager.config.FilterSubject != "ci.assignment.pod" || manager.config.AckPolicy != jetstream.AckExplicitPolicy || manager.config.MaxAckPending != 4 {
+	if manager.config.Durable != "pod" || manager.config.FilterSubject != "ci.assignment.pod" || manager.config.AckPolicy != jetstream.AckExplicitPolicy || manager.config.MaxAckPending != 4 || manager.config.MaxDeliver != 1000 {
 		t.Fatalf("config = %#v", manager.config)
 	}
 }

internal/opensandboxbackend/backend_test.go

@@ -90,7 +90,7 @@ func TestCreateUsesPoolAndPersistsRecoveryMetadata(t *testing.T) {
 	if err != nil {
 		t.Fatal(err)
 	}
-	if lifecycle.created.Extensions["poolRef"] != "ci-vm" || lifecycle.created.Metadata[assignmentMetadata] != assignment().ID || lifecycle.created.Env["CI_SPIFFE_ID"] != assignment().Identity.SPIFFEID || executor.Name != "sandbox-42" {
+	if lifecycle.created.Extensions["poolRef"] != "ci-vm" || lifecycle.created.Metadata[assignmentMetadata] != assignment().ID || lifecycle.created.Metadata["ci.ddupan.top/runner"] != "true" || lifecycle.created.Env["CI_SPIFFE_ID"] != assignment().Identity.SPIFFEID || executor.Name != "sandbox-42" {
 		t.Fatalf("request=%#v executor=%#v", lifecycle.created, executor)
 	}
 	if lifecycle.created.Env["CI_RUNNER_CAPABILITY"] != "capability" {

internal/taskassignment/assignment.go

@@ -102,7 +102,7 @@ func backendFromTask(task *runnerv1.Task) (Backend, error) {
 		return "", fmt.Errorf("task runs-on labels must include self-hosted: %v", labels)
 	}
 	hasPod := slices.Contains(labels, string(BackendPod))
-	hasVM := slices.Contains(labels, string(BackendVM))
+	hasVM := slices.Contains(labels, string(BackendVM)) || slices.Contains(labels, "vm-dev")
 	if hasPod == hasVM {
 		return "", fmt.Errorf("task runs-on labels must select exactly one of pod or vm: %v", labels)
 	}

internal/taskassignment/assignment_test.go

@@ -28,6 +28,7 @@ func TestNewSelectsBackendFromRunsOn(t *testing.T) {
 	}{
 		{"[self-hosted, pod]", BackendPod},
 		{"[self-hosted, vm]", BackendVM},
+		{"[self-hosted, vm-dev]", BackendVM},
 	} {
 		assignment, err := New(task(t, test.labels), "ddupan.top")
 		if err != nil {

internal/taskworker/worker.go

@@ -178,6 +178,7 @@ func (w Worker) launchSpec(assignment taskassignment.Assignment) (LaunchSpec, er
 func BackendMetadata(assignment taskassignment.Assignment) Metadata {
 	return Metadata{
 		Labels: map[string]string{
+			"ci.ddupan.top/runner":        "true",
 			"ci.ddupan.top/assignment-id": assignment.ID,
 			"ci.ddupan.top/task-id":       strconv.FormatInt(assignment.Task.GetId(), 10),
 			"ci.ddupan.top/backend":       string(assignment.Backend),