X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/cb4efac6793d18892dde09c631895cb98c3df470..11597243529c762114fc34907f772c331d6f6ef5:/lib/dispatchcloud/dispatcher.go

diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go
index 71ff9c784e..f0aa83c2e0 100644
--- a/lib/dispatchcloud/dispatcher.go
+++ b/lib/dispatchcloud/dispatcher.go
@@ -48,10 +48,10 @@ type dispatcher struct {
 	Context       context.Context
 	ArvClient     *arvados.Client
 	AuthToken     string
+	Registry      *prometheus.Registry
 	InstanceSetID cloud.InstanceSetID
 
 	logger      logrus.FieldLogger
-	reg         *prometheus.Registry
 	instanceSet cloud.InstanceSet
 	pool        pool
 	queue       scheduler.ContainerQueue
@@ -95,7 +95,7 @@ func (disp *dispatcher) Close() {
 // Make a worker.Executor for the given instance.
 func (disp *dispatcher) newExecutor(inst cloud.Instance) worker.Executor {
 	exr := ssh_executor.New(inst)
-	exr.SetTargetPort(disp.Cluster.CloudVMs.SSHPort)
+	exr.SetTargetPort(disp.Cluster.Containers.CloudVMs.SSHPort)
 	exr.SetSigners(disp.sshKey)
 	return exr
 }
@@ -126,20 +126,19 @@ func (disp *dispatcher) initialize() {
 	disp.stop = make(chan struct{}, 1)
 	disp.stopped = make(chan struct{})
 
-	if key, err := ssh.ParsePrivateKey([]byte(disp.Cluster.Dispatch.PrivateKey)); err != nil {
-		disp.logger.Fatalf("error parsing configured Dispatch.PrivateKey: %s", err)
+	if key, err := ssh.ParsePrivateKey([]byte(disp.Cluster.Containers.DispatchPrivateKey)); err != nil {
+		disp.logger.Fatalf("error parsing configured Containers.DispatchPrivateKey: %s", err)
 	} else {
 		disp.sshKey = key
 	}
 
-	instanceSet, err := newInstanceSet(disp.Cluster, disp.InstanceSetID, disp.logger)
+	instanceSet, err := newInstanceSet(disp.Cluster, disp.InstanceSetID, disp.logger, disp.Registry)
 	if err != nil {
 		disp.logger.Fatalf("error initializing driver: %s", err)
 	}
 	disp.instanceSet = instanceSet
-	disp.reg = prometheus.NewRegistry()
-	disp.pool = worker.NewPool(disp.logger, disp.ArvClient, disp.reg, disp.instanceSet, disp.newExecutor, disp.sshKey.PublicKey(), disp.Cluster)
-	disp.queue = container.NewQueue(disp.logger, disp.reg, disp.typeChooser, disp.ArvClient)
+	disp.pool = worker.NewPool(disp.logger, disp.ArvClient, disp.Registry, disp.InstanceSetID, disp.instanceSet, disp.newExecutor, disp.sshKey.PublicKey(), disp.Cluster)
+	disp.queue = container.NewQueue(disp.logger, disp.Registry, disp.typeChooser, disp.ArvClient)
 
 	if disp.Cluster.ManagementToken == "" {
 		disp.httpHandler = http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
@@ -148,12 +147,13 @@ func (disp *dispatcher) initialize() {
 	} else {
 		mux := httprouter.New()
 		mux.HandlerFunc("GET", "/arvados/v1/dispatch/containers", disp.apiContainers)
+		mux.HandlerFunc("POST", "/arvados/v1/dispatch/containers/kill", disp.apiInstanceKill)
 		mux.HandlerFunc("GET", "/arvados/v1/dispatch/instances", disp.apiInstances)
 		mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/hold", disp.apiInstanceHold)
 		mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/drain", disp.apiInstanceDrain)
 		mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/run", disp.apiInstanceRun)
 		mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/kill", disp.apiInstanceKill)
-		metricsH := promhttp.HandlerFor(disp.reg, promhttp.HandlerOpts{
+		metricsH := promhttp.HandlerFor(disp.Registry, promhttp.HandlerOpts{
 			ErrorLog: disp.logger,
 		})
 		mux.Handler("GET", "/metrics", metricsH)
@@ -167,11 +167,11 @@ func (disp *dispatcher) run() {
 	defer disp.instanceSet.Stop()
 	defer disp.pool.Stop()
 
-	staleLockTimeout := time.Duration(disp.Cluster.Dispatch.StaleLockTimeout)
+	staleLockTimeout := time.Duration(disp.Cluster.Containers.StaleLockTimeout)
 	if staleLockTimeout == 0 {
 		staleLockTimeout = defaultStaleLockTimeout
 	}
-	pollInterval := time.Duration(disp.Cluster.Dispatch.PollInterval)
+	pollInterval := time.Duration(disp.Cluster.Containers.CloudVMs.PollInterval)
 	if pollInterval <= 0 {
 		pollInterval = defaultPollInterval
 	}
@@ -232,6 +232,20 @@ func (disp *dispatcher) apiInstanceKill(w http.ResponseWriter, r *http.Request)
 	}
 }
 
+// Management API: send SIGTERM to specified container's crunch-run
+// process now.
+func (disp *dispatcher) apiContainerKill(w http.ResponseWriter, r *http.Request) {
+	uuid := r.FormValue("container_uuid")
+	if uuid == "" {
+		httpserver.Error(w, "container_uuid parameter not provided", http.StatusBadRequest)
+		return
+	}
+	if !disp.pool.KillContainer(uuid, "via management API: "+r.FormValue("reason")) {
+		httpserver.Error(w, "container not found", http.StatusNotFound)
+		return
+	}
+}
+
 func (disp *dispatcher) apiInstanceIdleBehavior(w http.ResponseWriter, r *http.Request, want worker.IdleBehavior) {
 	id := cloud.InstanceID(r.FormValue("instance_id"))
 	if id == "" {