lib/dispatchcloud/scheduler/run_queue_test.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package scheduler
   6
   7 import (
   8         "context"
   9         "sync"
  10         "time"
  11
  12         "git.curoverse.com/arvados.git/lib/dispatchcloud/test"
  13         "git.curoverse.com/arvados.git/lib/dispatchcloud/worker"
  14         "git.curoverse.com/arvados.git/sdk/go/arvados"
  15         "git.curoverse.com/arvados.git/sdk/go/ctxlog"
  16         check "gopkg.in/check.v1"
  17 )
  18
  19 var (
  20         // arbitrary example container UUIDs
  21         uuids = func() (r []string) {
  22                 for i := 0; i < 16; i++ {
  23                         r = append(r, test.ContainerUUID(i))
  24                 }
  25                 return
  26         }()
  27 )
  28
  29 type stubQuotaError struct {
  30         error
  31 }
  32
  33 func (stubQuotaError) IsQuotaError() bool { return true }
  34
  35 type stubPool struct {
  36         notify    <-chan struct{}
  37         unalloc   map[arvados.InstanceType]int // idle+booting+unknown
  38         idle      map[arvados.InstanceType]int
  39         running   map[string]time.Time
  40         atQuota   bool
  41         canCreate int
  42         creates   []arvados.InstanceType
  43         starts    []string
  44         shutdowns int
  45         sync.Mutex
  46 }
  47
  48 func (p *stubPool) AtQuota() bool               { return p.atQuota }
  49 func (p *stubPool) Subscribe() <-chan struct{}  { return p.notify }
  50 func (p *stubPool) Unsubscribe(<-chan struct{}) {}
  51 func (p *stubPool) Running() map[string]time.Time {
  52         p.Lock()
  53         defer p.Unlock()
  54         r := map[string]time.Time{}
  55         for k, v := range p.running {
  56                 r[k] = v
  57         }
  58         return r
  59 }
  60 func (p *stubPool) Unallocated() map[arvados.InstanceType]int {
  61         p.Lock()
  62         defer p.Unlock()
  63         r := map[arvados.InstanceType]int{}
  64         for it, n := range p.unalloc {
  65                 r[it] = n
  66         }
  67         return r
  68 }
  69 func (p *stubPool) Create(it arvados.InstanceType) bool {
  70         p.Lock()
  71         defer p.Unlock()
  72         p.creates = append(p.creates, it)
  73         if p.canCreate < 1 {
  74                 return false
  75         }
  76         p.canCreate--
  77         p.unalloc[it]++
  78         return true
  79 }
  80 func (p *stubPool) KillContainer(uuid, reason string) {
  81         p.Lock()
  82         defer p.Unlock()
  83         delete(p.running, uuid)
  84 }
  85 func (p *stubPool) Shutdown(arvados.InstanceType) bool {
  86         p.shutdowns++
  87         return false
  88 }
  89 func (p *stubPool) CountWorkers() map[worker.State]int {
  90         p.Lock()
  91         defer p.Unlock()
  92         return map[worker.State]int{
  93                 worker.StateBooting: len(p.unalloc) - len(p.idle),
  94                 worker.StateIdle:    len(p.idle),
  95                 worker.StateRunning: len(p.running),
  96         }
  97 }
  98 func (p *stubPool) StartContainer(it arvados.InstanceType, ctr arvados.Container) bool {
  99         p.Lock()
 100         defer p.Unlock()
 101         p.starts = append(p.starts, ctr.UUID)
 102         if p.idle[it] == 0 {
 103                 return false
 104         }
 105         p.idle[it]--
 106         p.unalloc[it]--
 107         p.running[ctr.UUID] = time.Time{}
 108         return true
 109 }
 110
 111 func chooseType(ctr *arvados.Container) (arvados.InstanceType, error) {
 112         return test.InstanceType(ctr.RuntimeConstraints.VCPUs), nil
 113 }
 114
 115 var _ = check.Suite(&SchedulerSuite{})
 116
 117 type SchedulerSuite struct{}
 118
 119 // Assign priority=4 container to idle node. Create a new instance for
 120 // the priority=3 container. Don't try to start any priority<3
 121 // containers because priority=3 container didn't start
 122 // immediately. Don't try to create any other nodes after the failed
 123 // create.
 124 func (*SchedulerSuite) TestUseIdleWorkers(c *check.C) {
 125         ctx := ctxlog.Context(context.Background(), ctxlog.TestLogger(c))
 126         queue := test.Queue{
 127                 ChooseType: chooseType,
 128                 Containers: []arvados.Container{
 129                         {
 130                                 UUID:     test.ContainerUUID(1),
 131                                 Priority: 1,
 132                                 State:    arvados.ContainerStateLocked,
 133                                 RuntimeConstraints: arvados.RuntimeConstraints{
 134                                         VCPUs: 1,
 135                                         RAM:   1 << 30,
 136                                 },
 137                         },
 138                         {
 139                                 UUID:     test.ContainerUUID(2),
 140                                 Priority: 2,
 141                                 State:    arvados.ContainerStateLocked,
 142                                 RuntimeConstraints: arvados.RuntimeConstraints{
 143                                         VCPUs: 1,
 144                                         RAM:   1 << 30,
 145                                 },
 146                         },
 147                         {
 148                                 UUID:     test.ContainerUUID(3),
 149                                 Priority: 3,
 150                                 State:    arvados.ContainerStateLocked,
 151                                 RuntimeConstraints: arvados.RuntimeConstraints{
 152                                         VCPUs: 1,
 153                                         RAM:   1 << 30,
 154                                 },
 155                         },
 156                         {
 157                                 UUID:     test.ContainerUUID(4),
 158                                 Priority: 4,
 159                                 State:    arvados.ContainerStateLocked,
 160                                 RuntimeConstraints: arvados.RuntimeConstraints{
 161                                         VCPUs: 1,
 162                                         RAM:   1 << 30,
 163                                 },
 164                         },
 165                 },
 166         }
 167         queue.Update()
 168         pool := stubPool{
 169                 unalloc: map[arvados.InstanceType]int{
 170                         test.InstanceType(1): 1,
 171                         test.InstanceType(2): 2,
 172                 },
 173                 idle: map[arvados.InstanceType]int{
 174                         test.InstanceType(1): 1,
 175                         test.InstanceType(2): 2,
 176                 },
 177                 running:   map[string]time.Time{},
 178                 canCreate: 0,
 179         }
 180         New(ctx, &queue, &pool, time.Millisecond, time.Millisecond).runQueue()
 181         c.Check(pool.creates, check.DeepEquals, []arvados.InstanceType{test.InstanceType(1)})
 182         c.Check(pool.starts, check.DeepEquals, []string{test.ContainerUUID(4)})
 183         c.Check(pool.running, check.HasLen, 1)
 184         for uuid := range pool.running {
 185                 c.Check(uuid, check.Equals, uuids[4])
 186         }
 187 }
 188
 189 // If Create() fails, shutdown some nodes, and don't call Create()
 190 // again.  Don't call Create() at all if AtQuota() is true.
 191 func (*SchedulerSuite) TestShutdownAtQuota(c *check.C) {
 192         ctx := ctxlog.Context(context.Background(), ctxlog.TestLogger(c))
 193         for quota := 0; quota < 2; quota++ {
 194                 c.Logf("quota=%d", quota)
 195                 shouldCreate := []arvados.InstanceType{}
 196                 for i := 0; i < quota; i++ {
 197                         shouldCreate = append(shouldCreate, test.InstanceType(3))
 198                 }
 199                 queue := test.Queue{
 200                         ChooseType: chooseType,
 201                         Containers: []arvados.Container{
 202                                 {
 203                                         UUID:     test.ContainerUUID(2),
 204                                         Priority: 2,
 205                                         State:    arvados.ContainerStateLocked,
 206                                         RuntimeConstraints: arvados.RuntimeConstraints{
 207                                                 VCPUs: 2,
 208                                                 RAM:   2 << 30,
 209                                         },
 210                                 },
 211                                 {
 212                                         UUID:     test.ContainerUUID(3),
 213                                         Priority: 3,
 214                                         State:    arvados.ContainerStateLocked,
 215                                         RuntimeConstraints: arvados.RuntimeConstraints{
 216                                                 VCPUs: 3,
 217                                                 RAM:   3 << 30,
 218                                         },
 219                                 },
 220                         },
 221                 }
 222                 queue.Update()
 223                 pool := stubPool{
 224                         atQuota: quota == 0,
 225                         unalloc: map[arvados.InstanceType]int{
 226                                 test.InstanceType(2): 2,
 227                         },
 228                         idle: map[arvados.InstanceType]int{
 229                                 test.InstanceType(2): 2,
 230                         },
 231                         running:   map[string]time.Time{},
 232                         creates:   []arvados.InstanceType{},
 233                         starts:    []string{},
 234                         canCreate: 0,
 235                 }
 236                 New(ctx, &queue, &pool, time.Millisecond, time.Millisecond).runQueue()
 237                 c.Check(pool.creates, check.DeepEquals, shouldCreate)
 238                 c.Check(pool.starts, check.DeepEquals, []string{})
 239                 c.Check(pool.shutdowns, check.Not(check.Equals), 0)
 240         }
 241 }
 242
 243 // Start lower-priority containers while waiting for new/existing
 244 // workers to come up for higher-priority containers.
 245 func (*SchedulerSuite) TestStartWhileCreating(c *check.C) {
 246         ctx := ctxlog.Context(context.Background(), ctxlog.TestLogger(c))
 247         pool := stubPool{
 248                 unalloc: map[arvados.InstanceType]int{
 249                         test.InstanceType(1): 2,
 250                         test.InstanceType(2): 2,
 251                 },
 252                 idle: map[arvados.InstanceType]int{
 253                         test.InstanceType(1): 1,
 254                         test.InstanceType(2): 1,
 255                 },
 256                 running:   map[string]time.Time{},
 257                 canCreate: 4,
 258         }
 259         queue := test.Queue{
 260                 ChooseType: chooseType,
 261                 Containers: []arvados.Container{
 262                         {
 263                                 // create a new worker
 264                                 UUID:     test.ContainerUUID(1),
 265                                 Priority: 1,
 266                                 State:    arvados.ContainerStateLocked,
 267                                 RuntimeConstraints: arvados.RuntimeConstraints{
 268                                         VCPUs: 1,
 269                                         RAM:   1 << 30,
 270                                 },
 271                         },
 272                         {
 273                                 // tentatively map to unalloc worker
 274                                 UUID:     test.ContainerUUID(2),
 275                                 Priority: 2,
 276                                 State:    arvados.ContainerStateLocked,
 277                                 RuntimeConstraints: arvados.RuntimeConstraints{
 278                                         VCPUs: 1,
 279                                         RAM:   1 << 30,
 280                                 },
 281                         },
 282                         {
 283                                 // start now on idle worker
 284                                 UUID:     test.ContainerUUID(3),
 285                                 Priority: 3,
 286                                 State:    arvados.ContainerStateLocked,
 287                                 RuntimeConstraints: arvados.RuntimeConstraints{
 288                                         VCPUs: 1,
 289                                         RAM:   1 << 30,
 290                                 },
 291                         },
 292                         {
 293                                 // create a new worker
 294                                 UUID:     test.ContainerUUID(4),
 295                                 Priority: 4,
 296                                 State:    arvados.ContainerStateLocked,
 297                                 RuntimeConstraints: arvados.RuntimeConstraints{
 298                                         VCPUs: 2,
 299                                         RAM:   2 << 30,
 300                                 },
 301                         },
 302                         {
 303                                 // tentatively map to unalloc worker
 304                                 UUID:     test.ContainerUUID(5),
 305                                 Priority: 5,
 306                                 State:    arvados.ContainerStateLocked,
 307                                 RuntimeConstraints: arvados.RuntimeConstraints{
 308                                         VCPUs: 2,
 309                                         RAM:   2 << 30,
 310                                 },
 311                         },
 312                         {
 313                                 // start now on idle worker
 314                                 UUID:     test.ContainerUUID(6),
 315                                 Priority: 6,
 316                                 State:    arvados.ContainerStateLocked,
 317                                 RuntimeConstraints: arvados.RuntimeConstraints{
 318                                         VCPUs: 2,
 319                                         RAM:   2 << 30,
 320                                 },
 321                         },
 322                 },
 323         }
 324         queue.Update()
 325         New(ctx, &queue, &pool, time.Millisecond, time.Millisecond).runQueue()
 326         c.Check(pool.creates, check.DeepEquals, []arvados.InstanceType{test.InstanceType(2), test.InstanceType(1)})
 327         c.Check(pool.starts, check.DeepEquals, []string{uuids[6], uuids[5], uuids[3], uuids[2]})
 328         running := map[string]bool{}
 329         for uuid, t := range pool.running {
 330                 if t.IsZero() {
 331                         running[uuid] = false
 332                 } else {
 333                         running[uuid] = true
 334                 }
 335         }
 336         c.Check(running, check.DeepEquals, map[string]bool{uuids[3]: false, uuids[6]: false})
 337 }
 338
 339 func (*SchedulerSuite) TestKillNonexistentContainer(c *check.C) {
 340         ctx := ctxlog.Context(context.Background(), ctxlog.TestLogger(c))
 341         pool := stubPool{
 342                 unalloc: map[arvados.InstanceType]int{
 343                         test.InstanceType(2): 0,
 344                 },
 345                 idle: map[arvados.InstanceType]int{
 346                         test.InstanceType(2): 0,
 347                 },
 348                 running: map[string]time.Time{
 349                         test.ContainerUUID(2): time.Time{},
 350                 },
 351         }
 352         queue := test.Queue{
 353                 ChooseType: chooseType,
 354                 Containers: []arvados.Container{
 355                         {
 356                                 // create a new worker
 357                                 UUID:     test.ContainerUUID(1),
 358                                 Priority: 1,
 359                                 State:    arvados.ContainerStateLocked,
 360                                 RuntimeConstraints: arvados.RuntimeConstraints{
 361                                         VCPUs: 1,
 362                                         RAM:   1 << 30,
 363                                 },
 364                         },
 365                 },
 366         }
 367         queue.Update()
 368         sch := New(ctx, &queue, &pool, time.Millisecond, time.Millisecond)
 369         c.Check(pool.running, check.HasLen, 1)
 370         sch.sync()
 371         for deadline := time.Now().Add(time.Second); len(pool.Running()) > 0 && time.Now().Before(deadline); time.Sleep(time.Millisecond) {
 372         }
 373         c.Check(pool.Running(), check.HasLen, 0)
 374 }