X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/dfb9598282b677ead60f91c14f5e96405735d42f..65b12213f740b117fb14822bce0dbb415257c355:/lib/dispatchcloud/dispatcher.go diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go index 19c8f6e0b4..3403c50c97 100644 --- a/lib/dispatchcloud/dispatcher.go +++ b/lib/dispatchcloud/dispatcher.go @@ -15,13 +15,16 @@ import ( "time" "git.arvados.org/arvados.git/lib/cloud" + "git.arvados.org/arvados.git/lib/controller/dblock" + "git.arvados.org/arvados.git/lib/ctrlctx" "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/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" @@ -52,6 +55,7 @@ type dispatcher struct { Registry *prometheus.Registry InstanceSetID cloud.InstanceSetID + dbConnector ctrlctx.DBConnector logger logrus.FieldLogger instanceSet cloud.InstanceSet pool pool @@ -82,6 +86,11 @@ func (disp *dispatcher) CheckHealth() error { 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 // in tests. func (disp *dispatcher) Close() { @@ -95,7 +104,7 @@ 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 := sshexecutor.New(inst) exr.SetTargetPort(disp.Cluster.Containers.CloudVMs.SSHPort) exr.SetSigners(disp.sshKey) return exr @@ -112,6 +121,7 @@ func (disp *dispatcher) setup() { func (disp *dispatcher) initialize() { disp.logger = ctxlog.FromContext(disp.Context) + disp.dbConnector = ctrlctx.DBConnector{PostgreSQL: disp.Cluster.PostgreSQL} disp.ArvClient.AuthToken = disp.AuthToken @@ -137,6 +147,7 @@ func (disp *dispatcher) initialize() { if err != nil { disp.logger.Fatalf("error initializing driver: %s", err) } + dblock.Dispatch.Lock(disp.Context, disp.dbConnector.GetDB) 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) @@ -148,7 +159,7 @@ 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("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) @@ -159,11 +170,17 @@ func (disp *dispatcher) initialize() { }) 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) } } func (disp *dispatcher) run() { + defer dblock.Dispatch.Unlock() defer close(disp.stopped) defer disp.instanceSet.Stop() defer disp.pool.Stop() @@ -176,7 +193,7 @@ func (disp *dispatcher) run() { if pollInterval <= 0 { pollInterval = defaultPollInterval } - sched := scheduler.New(disp.Context, disp.queue, disp.pool, staleLockTimeout, pollInterval) + sched := scheduler.New(disp.Context, disp.queue, disp.pool, disp.Registry, staleLockTimeout, pollInterval) sched.Start() defer sched.Stop()