X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8fc29fafb91cf64ce4ededbdd85ef9507c51f216..098a6b73cd5b7ad391ed30e95b825817d22aec24:/lib/dispatchcloud/dispatcher.go

diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go
index 3bf0ee9bd5..02b6c976ae 100644
--- a/lib/dispatchcloud/dispatcher.go
+++ b/lib/dispatchcloud/dispatcher.go
@@ -14,15 +14,15 @@ import (
 	"sync"
 	"time"
 
-	"git.curoverse.com/arvados.git/lib/cloud"
-	"git.curoverse.com/arvados.git/lib/dispatchcloud/container"
-	"git.curoverse.com/arvados.git/lib/dispatchcloud/scheduler"
-	"git.curoverse.com/arvados.git/lib/dispatchcloud/ssh_executor"
-	"git.curoverse.com/arvados.git/lib/dispatchcloud/worker"
-	"git.curoverse.com/arvados.git/sdk/go/arvados"
-	"git.curoverse.com/arvados.git/sdk/go/auth"
-	"git.curoverse.com/arvados.git/sdk/go/ctxlog"
-	"git.curoverse.com/arvados.git/sdk/go/httpserver"
+	"git.arvados.org/arvados.git/lib/cloud"
+	"git.arvados.org/arvados.git/lib/dispatchcloud/container"
+	"git.arvados.org/arvados.git/lib/dispatchcloud/scheduler"
+	"git.arvados.org/arvados.git/lib/dispatchcloud/ssh_executor"
+	"git.arvados.org/arvados.git/lib/dispatchcloud/worker"
+	"git.arvados.org/arvados.git/sdk/go/arvados"
+	"git.arvados.org/arvados.git/sdk/go/auth"
+	"git.arvados.org/arvados.git/sdk/go/ctxlog"
+	"git.arvados.org/arvados.git/sdk/go/httpserver"
 	"github.com/julienschmidt/httprouter"
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
@@ -37,6 +37,7 @@ const (
 
 type pool interface {
 	scheduler.WorkerPool
+	CheckHealth() error
 	Instances() []worker.InstanceView
 	SetIdleBehavior(cloud.InstanceID, worker.IdleBehavior) error
 	KillInstance(id cloud.InstanceID, reason string) error
@@ -48,10 +49,10 @@ 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
@@ -78,7 +79,12 @@ func (disp *dispatcher) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 // CheckHealth implements service.Handler.
 func (disp *dispatcher) CheckHealth() error {
 	disp.Start()
-	return nil
+	return disp.pool.CheckHealth()
+}
+
+// Done implements service.Handler.
+func (disp *dispatcher) Done() <-chan struct{} {
+	return disp.stopped
 }
 
 // Stop dispatching containers and release resources. Typically used
@@ -132,14 +138,13 @@ func (disp *dispatcher) initialize() {
 		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 +153,13 @@ 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.apiContainerKill)
 		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)
@@ -232,6 +238,20 @@ 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 == "" {