lib/dispatchcloud/scheduler/scheduler.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 // Package scheduler uses a resizable worker pool to execute
   6 // containers in priority order.
   7 package scheduler
   8
   9 import (
  10         "context"
  11         "sync"
  12         "time"
  13
  14         "git.arvados.org/arvados.git/sdk/go/arvados"
  15         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  16         "github.com/prometheus/client_golang/prometheus"
  17         "github.com/sirupsen/logrus"
  18 )
  19
  20 // A Scheduler maps queued containers onto unallocated workers in
  21 // priority order, creating new workers if needed. It locks containers
  22 // that can be mapped onto existing/pending workers, and starts them
  23 // if possible.
  24 //
  25 // A Scheduler unlocks any containers that are locked but can't be
  26 // mapped. (For example, this happens when the cloud provider reaches
  27 // quota/capacity and a previously mappable container's priority is
  28 // surpassed by a newer container.)
  29 //
  30 // If it encounters errors while creating new workers, a Scheduler
  31 // shuts down idle workers, in case they are consuming quota.
  32 type Scheduler struct {
  33         logger              logrus.FieldLogger
  34         client              *arvados.Client
  35         queue               ContainerQueue
  36         pool                WorkerPool
  37         reg                 *prometheus.Registry
  38         staleLockTimeout    time.Duration
  39         queueUpdateInterval time.Duration
  40
  41         uuidOp map[string]string // operation in progress: "lock", "cancel", ...
  42         mtx    sync.Mutex
  43         wakeup *time.Timer
  44
  45         runOnce sync.Once
  46         stop    chan struct{}
  47         stopped chan struct{}
  48
  49         last503time    time.Time // last time API responded 503
  50         maxConcurrency int       // dynamic container limit (0 = unlimited), see runQueue()
  51
  52         mContainersAllocatedNotStarted   prometheus.Gauge
  53         mContainersNotAllocatedOverQuota prometheus.Gauge
  54         mLongestWaitTimeSinceQueue       prometheus.Gauge
  55         mLast503Time                     prometheus.Gauge
  56         mMaxContainerConcurrency         prometheus.Gauge
  57 }
  58
  59 // New returns a new unstarted Scheduler.
  60 //
  61 // Any given queue and pool should not be used by more than one
  62 // scheduler at a time.
  63 func New(ctx context.Context, client *arvados.Client, queue ContainerQueue, pool WorkerPool, reg *prometheus.Registry, staleLockTimeout, queueUpdateInterval time.Duration) *Scheduler {
  64         sch := &Scheduler{
  65                 logger:              ctxlog.FromContext(ctx),
  66                 client:              client,
  67                 queue:               queue,
  68                 pool:                pool,
  69                 reg:                 reg,
  70                 staleLockTimeout:    staleLockTimeout,
  71                 queueUpdateInterval: queueUpdateInterval,
  72                 wakeup:              time.NewTimer(time.Second),
  73                 stop:                make(chan struct{}),
  74                 stopped:             make(chan struct{}),
  75                 uuidOp:              map[string]string{},
  76         }
  77         sch.registerMetrics(reg)
  78         return sch
  79 }
  80
  81 func (sch *Scheduler) registerMetrics(reg *prometheus.Registry) {
  82         if reg == nil {
  83                 reg = prometheus.NewRegistry()
  84         }
  85         sch.mContainersAllocatedNotStarted = prometheus.NewGauge(prometheus.GaugeOpts{
  86                 Namespace: "arvados",
  87                 Subsystem: "dispatchcloud",
  88                 Name:      "containers_allocated_not_started",
  89                 Help:      "Number of containers allocated to a worker but not started yet (worker is booting).",
  90         })
  91         reg.MustRegister(sch.mContainersAllocatedNotStarted)
  92         sch.mContainersNotAllocatedOverQuota = prometheus.NewGauge(prometheus.GaugeOpts{
  93                 Namespace: "arvados",
  94                 Subsystem: "dispatchcloud",
  95                 Name:      "containers_not_allocated_over_quota",
  96                 Help:      "Number of containers not allocated to a worker because the system has hit a quota.",
  97         })
  98         reg.MustRegister(sch.mContainersNotAllocatedOverQuota)
  99         sch.mLongestWaitTimeSinceQueue = prometheus.NewGauge(prometheus.GaugeOpts{
 100                 Namespace: "arvados",
 101                 Subsystem: "dispatchcloud",
 102                 Name:      "containers_longest_wait_time_seconds",
 103                 Help:      "Current longest wait time of any container since queuing, and before the start of crunch-run.",
 104         })
 105         reg.MustRegister(sch.mLongestWaitTimeSinceQueue)
 106         sch.mLast503Time = prometheus.NewGauge(prometheus.GaugeOpts{
 107                 Namespace: "arvados",
 108                 Subsystem: "dispatchcloud",
 109                 Name:      "last_503_time",
 110                 Help:      "Time of most recent 503 error received from API.",
 111         })
 112         reg.MustRegister(sch.mLast503Time)
 113         sch.mMaxContainerConcurrency = prometheus.NewGauge(prometheus.GaugeOpts{
 114                 Namespace: "arvados",
 115                 Subsystem: "dispatchcloud",
 116                 Name:      "max_concurrent_containers",
 117                 Help:      "Dynamically assigned limit on number of containers scheduled concurrency, set after receiving 503 errors from API.",
 118         })
 119         reg.MustRegister(sch.mMaxContainerConcurrency)
 120 }
 121
 122 func (sch *Scheduler) updateMetrics() {
 123         earliest := time.Time{}
 124         entries, _ := sch.queue.Entries()
 125         running := sch.pool.Running()
 126         for _, ent := range entries {
 127                 if ent.Container.Priority > 0 &&
 128                         (ent.Container.State == arvados.ContainerStateQueued || ent.Container.State == arvados.ContainerStateLocked) {
 129                         // Exclude containers that are preparing to run the payload (i.e.
 130                         // ContainerStateLocked and running on a worker, most likely loading the
 131                         // payload image
 132                         if _, ok := running[ent.Container.UUID]; !ok {
 133                                 if ent.Container.CreatedAt.Before(earliest) || earliest.IsZero() {
 134                                         earliest = ent.Container.CreatedAt
 135                                 }
 136                         }
 137                 }
 138         }
 139         if !earliest.IsZero() {
 140                 sch.mLongestWaitTimeSinceQueue.Set(time.Since(earliest).Seconds())
 141         } else {
 142                 sch.mLongestWaitTimeSinceQueue.Set(0)
 143         }
 144 }
 145
 146 // Start starts the scheduler.
 147 func (sch *Scheduler) Start() {
 148         go sch.runOnce.Do(sch.run)
 149 }
 150
 151 // Stop stops the scheduler. No other method should be called after
 152 // Stop.
 153 func (sch *Scheduler) Stop() {
 154         close(sch.stop)
 155         <-sch.stopped
 156 }
 157
 158 func (sch *Scheduler) run() {
 159         defer close(sch.stopped)
 160
 161         // Ensure the queue is fetched once before attempting anything.
 162         for err := sch.queue.Update(); err != nil; err = sch.queue.Update() {
 163                 sch.logger.Errorf("error updating queue: %s", err)
 164                 d := sch.queueUpdateInterval / 10
 165                 if d < time.Second {
 166                         d = time.Second
 167                 }
 168                 sch.logger.Infof("waiting %s before retry", d)
 169                 time.Sleep(d)
 170         }
 171
 172         // Keep the queue up to date.
 173         poll := time.NewTicker(sch.queueUpdateInterval)
 174         defer poll.Stop()
 175         go func() {
 176                 for range poll.C {
 177                         err := sch.queue.Update()
 178                         if err != nil {
 179                                 sch.logger.Errorf("error updating queue: %s", err)
 180                         }
 181                 }
 182         }()
 183
 184         t0 := time.Now()
 185         sch.logger.Infof("FixStaleLocks starting.")
 186         sch.fixStaleLocks()
 187         sch.logger.Infof("FixStaleLocks finished (%s), starting scheduling.", time.Since(t0))
 188
 189         poolNotify := sch.pool.Subscribe()
 190         defer sch.pool.Unsubscribe(poolNotify)
 191
 192         queueNotify := sch.queue.Subscribe()
 193         defer sch.queue.Unsubscribe(queueNotify)
 194
 195         for {
 196                 sch.runQueue()
 197                 sch.sync()
 198                 sch.updateMetrics()
 199                 select {
 200                 case <-sch.stop:
 201                         return
 202                 case <-queueNotify:
 203                 case <-poolNotify:
 204                 case <-sch.wakeup.C:
 205                 }
 206         }
 207 }