X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/17479bd75a29c52470abe0049cb447e114eb39e9..d4da0c6adf235aa103337430fd73cf3bfba64fd9:/lib/dispatchcloud/dispatcher.go diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go index 97aacf6044..adf1028b35 100644 --- a/lib/dispatchcloud/dispatcher.go +++ b/lib/dispatchcloud/dispatcher.go @@ -5,6 +5,7 @@ package dispatchcloud import ( + "context" "crypto/md5" "encoding/json" "fmt" @@ -20,10 +21,12 @@ 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/Sirupsen/logrus" + "github.com/julienschmidt/httprouter" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" + "github.com/sirupsen/logrus" "golang.org/x/crypto/ssh" ) @@ -35,10 +38,13 @@ const ( type pool interface { scheduler.WorkerPool Instances() []worker.InstanceView + SetIdleBehavior(cloud.InstanceID, worker.IdleBehavior) error + Stop() } type dispatcher struct { Cluster *arvados.Cluster + Context context.Context InstanceSetID cloud.InstanceSetID logger logrus.FieldLogger @@ -51,6 +57,7 @@ type dispatcher struct { setupOnce sync.Once stop chan struct{} + stopped chan struct{} } // Start starts the dispatcher. Start can be called multiple times @@ -79,11 +86,13 @@ func (disp *dispatcher) Close() { case disp.stop <- struct{}{}: default: } + <-disp.stopped } // 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.SetSigners(disp.sshKey) return exr } @@ -109,36 +118,48 @@ func (disp *dispatcher) initialize() { } } disp.stop = make(chan struct{}, 1) - disp.logger = logrus.StandardLogger() + disp.stopped = make(chan struct{}) + disp.logger = ctxlog.FromContext(disp.Context) - if key, err := ssh.ParsePrivateKey(disp.Cluster.Dispatch.PrivateKey); err != nil { + if key, err := ssh.ParsePrivateKey([]byte(disp.Cluster.Dispatch.PrivateKey)); err != nil { disp.logger.Fatalf("error parsing configured Dispatch.PrivateKey: %s", err) } else { disp.sshKey = key } - instanceSet, err := newInstanceSet(disp.Cluster, disp.InstanceSetID) + instanceSet, err := newInstanceSet(disp.Cluster, disp.InstanceSetID, disp.logger) if err != nil { disp.logger.Fatalf("error initializing driver: %s", err) } - disp.instanceSet = &instanceSetProxy{instanceSet} + disp.instanceSet = instanceSet disp.reg = prometheus.NewRegistry() - disp.pool = worker.NewPool(disp.logger, disp.reg, disp.instanceSet, disp.newExecutor, disp.Cluster) + 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) - mux := http.NewServeMux() - mux.HandleFunc("/arvados/v1/dispatch/containers", disp.apiContainers) - mux.HandleFunc("/arvados/v1/dispatch/instances", disp.apiInstances) - metricsH := promhttp.HandlerFor(disp.reg, promhttp.HandlerOpts{ - ErrorLog: disp.logger, - }) - mux.Handle("/metrics", metricsH) - mux.Handle("/metrics.json", metricsH) - disp.httpHandler = auth.RequireLiteralToken(disp.Cluster.ManagementToken, mux) + if disp.Cluster.ManagementToken == "" { + disp.httpHandler = http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + http.Error(w, "Management API authentication is not configured", http.StatusForbidden) + }) + } else { + 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/hold", disp.apiInstanceHold) + mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/drain", disp.apiInstanceDrain) + mux.HandlerFunc("POST", "/arvados/v1/dispatch/instances/run", disp.apiInstanceRun) + metricsH := promhttp.HandlerFor(disp.reg, promhttp.HandlerOpts{ + ErrorLog: disp.logger, + }) + mux.Handler("GET", "/metrics", metricsH) + mux.Handler("GET", "/metrics.json", metricsH) + disp.httpHandler = auth.RequireLiteralToken(disp.Cluster.ManagementToken, mux) + } } func (disp *dispatcher) run() { + defer close(disp.stopped) defer disp.instanceSet.Stop() + defer disp.pool.Stop() staleLockTimeout := time.Duration(disp.Cluster.Dispatch.StaleLockTimeout) if staleLockTimeout == 0 { @@ -148,7 +169,7 @@ 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() @@ -157,12 +178,8 @@ func (disp *dispatcher) run() { // Management API: all active and queued containers. func (disp *dispatcher) apiContainers(w http.ResponseWriter, r *http.Request) { - if r.Method != "GET" { - httpserver.Error(w, "method not allowed", http.StatusMethodNotAllowed) - return - } var resp struct { - Items []container.QueueEnt + Items []container.QueueEnt `json:"items"` } qEntries, _ := disp.queue.Entries() for _, ent := range qEntries { @@ -173,13 +190,37 @@ func (disp *dispatcher) apiContainers(w http.ResponseWriter, r *http.Request) { // Management API: all active instances (cloud VMs). func (disp *dispatcher) apiInstances(w http.ResponseWriter, r *http.Request) { - if r.Method != "GET" { - httpserver.Error(w, "method not allowed", http.StatusMethodNotAllowed) - return - } var resp struct { - Items []worker.InstanceView + Items []worker.InstanceView `json:"items"` } resp.Items = disp.pool.Instances() json.NewEncoder(w).Encode(resp) } + +// Management API: set idle behavior to "hold" for specified instance. +func (disp *dispatcher) apiInstanceHold(w http.ResponseWriter, r *http.Request) { + disp.apiInstanceIdleBehavior(w, r, worker.IdleBehaviorHold) +} + +// Management API: set idle behavior to "drain" for specified instance. +func (disp *dispatcher) apiInstanceDrain(w http.ResponseWriter, r *http.Request) { + disp.apiInstanceIdleBehavior(w, r, worker.IdleBehaviorDrain) +} + +// Management API: set idle behavior to "run" for specified instance. +func (disp *dispatcher) apiInstanceRun(w http.ResponseWriter, r *http.Request) { + disp.apiInstanceIdleBehavior(w, r, worker.IdleBehaviorRun) +} + +func (disp *dispatcher) apiInstanceIdleBehavior(w http.ResponseWriter, r *http.Request, want worker.IdleBehavior) { + 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 { + httpserver.Error(w, err.Error(), http.StatusNotFound) + return + } +}