X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/79bce4a71a58118a9003882e0ca9bbfb9d2957a9..52a5451c01fda74c567fa43ba4af23688a398410:/lib/dispatchcloud/scheduler/scheduler.go

diff --git a/lib/dispatchcloud/scheduler/scheduler.go b/lib/dispatchcloud/scheduler/scheduler.go
index c3e67dd11f..b1f8ea2223 100644
--- a/lib/dispatchcloud/scheduler/scheduler.go
+++ b/lib/dispatchcloud/scheduler/scheduler.go
@@ -31,6 +31,7 @@ import (
 // shuts down idle workers, in case they are consuming quota.
 type Scheduler struct {
 	logger              logrus.FieldLogger
+	client              *arvados.Client
 	queue               ContainerQueue
 	pool                WorkerPool
 	reg                 *prometheus.Registry
@@ -45,18 +46,26 @@ type Scheduler struct {
 	stop    chan struct{}
 	stopped chan struct{}
 
+	last503time        time.Time // last time API responded 503
+	maxConcurrency     int       // dynamic container limit (0 = unlimited), see runQueue()
+	supervisorFraction float64   // maximum fraction of "supervisor" containers (these are containers who's main job is to launch other containers, e.g. workflow runners)
+	maxInstances       int       // maximum number of instances the pool will bring up (0 = unlimited)
+
 	mContainersAllocatedNotStarted   prometheus.Gauge
 	mContainersNotAllocatedOverQuota prometheus.Gauge
 	mLongestWaitTimeSinceQueue       prometheus.Gauge
+	mLast503Time                     prometheus.Gauge
+	mMaxContainerConcurrency         prometheus.Gauge
 }
 
 // New returns a new unstarted Scheduler.
 //
 // Any given queue and pool should not be used by more than one
 // scheduler at a time.
-func New(ctx context.Context, queue ContainerQueue, pool WorkerPool, reg *prometheus.Registry, staleLockTimeout, queueUpdateInterval time.Duration) *Scheduler {
+func New(ctx context.Context, client *arvados.Client, queue ContainerQueue, pool WorkerPool, reg *prometheus.Registry, staleLockTimeout, queueUpdateInterval time.Duration, maxInstances int, supervisorFraction float64) *Scheduler {
 	sch := &Scheduler{
 		logger:              ctxlog.FromContext(ctx),
+		client:              client,
 		queue:               queue,
 		pool:                pool,
 		reg:                 reg,
@@ -66,6 +75,9 @@ func New(ctx context.Context, queue ContainerQueue, pool WorkerPool, reg *promet
 		stop:                make(chan struct{}),
 		stopped:             make(chan struct{}),
 		uuidOp:              map[string]string{},
+		maxConcurrency:      maxInstances, // initial value -- will be dynamically adjusted
+		supervisorFraction:  supervisorFraction,
+		maxInstances:        maxInstances,
 	}
 	sch.registerMetrics(reg)
 	return sch
@@ -96,6 +108,20 @@ func (sch *Scheduler) registerMetrics(reg *prometheus.Registry) {
 		Help:      "Current longest wait time of any container since queuing, and before the start of crunch-run.",
 	})
 	reg.MustRegister(sch.mLongestWaitTimeSinceQueue)
+	sch.mLast503Time = prometheus.NewGauge(prometheus.GaugeOpts{
+		Namespace: "arvados",
+		Subsystem: "dispatchcloud",
+		Name:      "last_503_time",
+		Help:      "Time of most recent 503 error received from API.",
+	})
+	reg.MustRegister(sch.mLast503Time)
+	sch.mMaxContainerConcurrency = prometheus.NewGauge(prometheus.GaugeOpts{
+		Namespace: "arvados",
+		Subsystem: "dispatchcloud",
+		Name:      "max_concurrent_containers",
+		Help:      "Dynamically assigned limit on number of containers scheduled concurrency, set after receiving 503 errors from API.",
+	})
+	reg.MustRegister(sch.mMaxContainerConcurrency)
 }
 
 func (sch *Scheduler) updateMetrics() {
@@ -149,14 +175,23 @@ func (sch *Scheduler) run() {
 	}
 
 	// Keep the queue up to date.
-	poll := time.NewTicker(sch.queueUpdateInterval)
-	defer poll.Stop()
 	go func() {
-		for range poll.C {
+		for {
+			starttime := time.Now()
 			err := sch.queue.Update()
 			if err != nil {
 				sch.logger.Errorf("error updating queue: %s", err)
 			}
+			// If the previous update took a long time,
+			// that probably means the server is
+			// overloaded, so wait that long before doing
+			// another. Otherwise, wait for the configured
+			// poll interval.
+			delay := time.Since(starttime)
+			if delay < sch.queueUpdateInterval {
+				delay = sch.queueUpdateInterval
+			}
+			time.Sleep(delay)
 		}
 	}()