Fix 2.4.1 release date refs #19017

[arvados.git] / lib / dispatchcloud / dispatcher.go

diff --git a/lib/dispatchcloud/dispatcher.go b/lib/dispatchcloud/dispatcher.go
index ce7dc07301daf99f27a2a422ea1641c4e4d92902..ae91a710e395295f47a34cb5645f980021e79021 100644 (file)
--- a/lib/dispatchcloud/dispatcher.go
+++ b/lib/dispatchcloud/dispatcher.go
@@ -17,11 +17,12 @@ import (
        "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/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"
@@ -37,6 +38,7 @@ const (
 
 type pool interface {
        scheduler.WorkerPool
+       CheckHealth() error
        Instances() []worker.InstanceView
        SetIdleBehavior(cloud.InstanceID, worker.IdleBehavior) error
        KillInstance(id cloud.InstanceID, reason string) error
@@ -78,7 +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
@@ -94,7 +101,7 @@ 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
@@ -147,7 +154,7 @@ 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)
@@ -158,6 +165,11 @@ 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)
        }
 }
@@ -175,7 +187,7 @@ 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()