X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/74df5a58360fe6bcb273480d3ddec507a53b6b2b..f416c6e0370c78a9aa18c2e461e0861b404a669a:/lib/dispatchcloud/dispatcher.go diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go index 2415094ac0..ae91a710e3 100644 --- a/lib/dispatchcloud/dispatcher.go +++ b/lib/dispatchcloud/dispatcher.go @@ -5,6 +5,7 @@ package dispatchcloud import ( + "context" "crypto/md5" "encoding/json" "fmt" @@ -13,14 +14,17 @@ 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/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/sshexecutor" + "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/health" + "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" "github.com/sirupsen/logrus" @@ -34,16 +38,22 @@ const ( type pool interface { scheduler.WorkerPool + CheckHealth() error 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 + Registry *prometheus.Registry InstanceSetID cloud.InstanceSetID logger logrus.FieldLogger - reg *prometheus.Registry instanceSet cloud.InstanceSet pool pool queue scheduler.ContainerQueue @@ -70,7 +80,12 @@ 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 @@ -86,7 +101,8 @@ 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 := sshexecutor.New(inst) + exr.SetTargetPort(disp.Cluster.Containers.CloudVMs.SSHPort) exr.SetSigners(disp.sshKey) return exr } @@ -101,48 +117,59 @@ 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(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 = &instanceSetProxy{instanceSet} - disp.reg = prometheus.NewRegistry() - disp.pool = worker.NewPool(disp.logger, disp.reg, disp.instanceSet, disp.newExecutor, disp.Cluster) - disp.queue = container.NewQueue(disp.logger, disp.reg, disp.typeChooser, arvClient) + disp.instanceSet = instanceSet + 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) { http.Error(w, "Management API authentication is not configured", http.StatusForbidden) }) } else { - 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{ + 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.Registry, promhttp.HandlerOpts{ ErrorLog: disp.logger, }) - mux.Handle("/metrics", metricsH) - mux.Handle("/metrics.json", metricsH) + mux.Handler("GET", "/metrics", metricsH) + mux.Handler("GET", "/metrics.json", metricsH) + mux.Handler("GET", "/_health/:check", &health.Handler{ + Token: disp.Cluster.ManagementToken, + Prefix: "/_health/", + Routes: health.Routes{"ping": disp.CheckHealth}, + }) disp.httpHandler = auth.RequireLiteralToken(disp.Cluster.ManagementToken, mux) } } @@ -152,15 +179,15 @@ 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 } - sched := scheduler.New(disp.logger, disp.queue, disp.pool, staleLockTimeout, pollInterval) + sched := scheduler.New(disp.Context, disp.queue, disp.pool, disp.Registry, staleLockTimeout, pollInterval) sched.Start() defer sched.Stop() @@ -169,12 +196,8 @@ 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 { @@ -185,13 +208,65 @@ 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) +} + +// 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 + } +} + +// 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 == "" { + 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 + } +}