X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/97a1babd776addd419fb5c050157786bdbd6232f..9a7e2a24f5f3d261e554ac3815b7e2a4c2e24503:/lib/dispatchcloud/dispatcher.go diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go index 814dba917f..71ff9c784e 100644 --- a/lib/dispatchcloud/dispatcher.go +++ b/lib/dispatchcloud/dispatcher.go @@ -5,6 +5,7 @@ package dispatchcloud import ( + "context" "crypto/md5" "encoding/json" "fmt" @@ -20,6 +21,7 @@ import ( "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" "github.com/julienschmidt/httprouter" "github.com/prometheus/client_golang/prometheus" @@ -37,11 +39,15 @@ type pool interface { scheduler.WorkerPool Instances() []worker.InstanceView SetIdleBehavior(cloud.InstanceID, worker.IdleBehavior) error + KillInstance(id cloud.InstanceID, reason string) error Stop() } type dispatcher struct { Cluster *arvados.Cluster + Context context.Context + ArvClient *arvados.Client + AuthToken string InstanceSetID cloud.InstanceSetID logger logrus.FieldLogger @@ -104,19 +110,21 @@ func (disp *dispatcher) setup() { } func (disp *dispatcher) initialize() { - arvClient := arvados.NewClientFromEnv() + disp.logger = ctxlog.FromContext(disp.Context) + + disp.ArvClient.AuthToken = disp.AuthToken + if disp.InstanceSetID == "" { - if strings.HasPrefix(arvClient.AuthToken, "v2/") { - disp.InstanceSetID = cloud.InstanceSetID(strings.Split(arvClient.AuthToken, "/")[1]) + if strings.HasPrefix(disp.AuthToken, "v2/") { + disp.InstanceSetID = cloud.InstanceSetID(strings.Split(disp.AuthToken, "/")[1]) } else { // Use some other string unique to this token // that doesn't reveal the token itself. - disp.InstanceSetID = cloud.InstanceSetID(fmt.Sprintf("%x", md5.Sum([]byte(arvClient.AuthToken)))) + disp.InstanceSetID = cloud.InstanceSetID(fmt.Sprintf("%x", md5.Sum([]byte(disp.AuthToken)))) } } disp.stop = make(chan struct{}, 1) disp.stopped = make(chan struct{}) - disp.logger = logrus.StandardLogger() if key, err := ssh.ParsePrivateKey([]byte(disp.Cluster.Dispatch.PrivateKey)); err != nil { disp.logger.Fatalf("error parsing configured Dispatch.PrivateKey: %s", err) @@ -130,8 +138,8 @@ func (disp *dispatcher) initialize() { } disp.instanceSet = instanceSet disp.reg = prometheus.NewRegistry() - disp.pool = worker.NewPool(disp.logger, arvClient, disp.reg, disp.instanceSet, disp.newExecutor, disp.sshKey.PublicKey(), disp.Cluster) - disp.queue = container.NewQueue(disp.logger, disp.reg, disp.typeChooser, arvClient) + 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) if disp.Cluster.ManagementToken == "" { disp.httpHandler = http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { @@ -141,9 +149,10 @@ func (disp *dispatcher) initialize() { mux := httprouter.New() mux.HandlerFunc("GET", "/arvados/v1/dispatch/containers", disp.apiContainers) mux.HandlerFunc("GET", "/arvados/v1/dispatch/instances", disp.apiInstances) - mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/:instance_id/hold", disp.apiInstanceHold) - mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/:instance_id/drain", disp.apiInstanceDrain) - mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/:instance_id/run", disp.apiInstanceRun) + 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{ ErrorLog: disp.logger, }) @@ -166,7 +175,7 @@ func (disp *dispatcher) run() { if pollInterval <= 0 { pollInterval = defaultPollInterval } - sched := scheduler.New(disp.logger, disp.queue, disp.pool, staleLockTimeout, pollInterval) + sched := scheduler.New(disp.Context, disp.queue, disp.pool, staleLockTimeout, pollInterval) sched.Start() defer sched.Stop() @@ -209,11 +218,25 @@ func (disp *dispatcher) apiInstanceRun(w http.ResponseWriter, r *http.Request) { disp.apiInstanceIdleBehavior(w, r, worker.IdleBehaviorRun) } +// Management API: shutdown/destroy specified instance now. +func (disp *dispatcher) apiInstanceKill(w http.ResponseWriter, r *http.Request) { + id := cloud.InstanceID(r.FormValue("instance_id")) + if id == "" { + httpserver.Error(w, "instance_id parameter not provided", http.StatusBadRequest) + return + } + err := disp.pool.KillInstance(id, "via management API: "+r.FormValue("reason")) + if err != nil { + httpserver.Error(w, err.Error(), http.StatusNotFound) + return + } +} + func (disp *dispatcher) apiInstanceIdleBehavior(w http.ResponseWriter, r *http.Request, want worker.IdleBehavior) { - params, _ := r.Context().Value(httprouter.ParamsKey).(httprouter.Params) - id := cloud.InstanceID(params.ByName("instance_id")) - if qp := r.FormValue("instance_id"); qp != "" { - id = cloud.InstanceID(qp) + id := cloud.InstanceID(r.FormValue("instance_id")) + if id == "" { + httpserver.Error(w, "instance_id parameter not provided", http.StatusBadRequest) + return } err := disp.pool.SetIdleBehavior(id, want) if err != nil {