lib/dispatchcloud/dispatcher_test.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package dispatchcloud
   6
   7 import (
   8         "encoding/json"
   9         "io/ioutil"
  10         "math/rand"
  11         "net/http"
  12         "net/http/httptest"
  13         "os"
  14         "sync"
  15         "time"
  16
  17         "git.curoverse.com/arvados.git/lib/dispatchcloud/test"
  18         "git.curoverse.com/arvados.git/sdk/go/arvados"
  19         "github.com/sirupsen/logrus"
  20         "golang.org/x/crypto/ssh"
  21         check "gopkg.in/check.v1"
  22 )
  23
  24 var _ = check.Suite(&DispatcherSuite{})
  25
  26 type DispatcherSuite struct {
  27         cluster     *arvados.Cluster
  28         instanceSet *test.LameInstanceSet
  29         stubDriver  *test.StubDriver
  30         disp        *dispatcher
  31 }
  32
  33 func (s *DispatcherSuite) SetUpSuite(c *check.C) {
  34         if os.Getenv("ARVADOS_DEBUG") != "" {
  35                 logrus.StandardLogger().SetLevel(logrus.DebugLevel)
  36         }
  37 }
  38
  39 func (s *DispatcherSuite) SetUpTest(c *check.C) {
  40         dispatchpub, _ := test.LoadTestKey(c, "test/sshkey_dispatch")
  41         dispatchprivraw, err := ioutil.ReadFile("test/sshkey_dispatch")
  42         c.Assert(err, check.IsNil)
  43
  44         _, hostpriv := test.LoadTestKey(c, "test/sshkey_vm")
  45         s.stubDriver = &test.StubDriver{
  46                 HostKey:                   hostpriv,
  47                 AuthorizedKeys:            []ssh.PublicKey{dispatchpub},
  48                 ErrorRateDestroy:          0.1,
  49                 MinTimeBetweenCreateCalls: time.Millisecond,
  50         }
  51
  52         s.cluster = &arvados.Cluster{
  53                 CloudVMs: arvados.CloudVMs{
  54                         Driver:          "test",
  55                         SyncInterval:    arvados.Duration(10 * time.Millisecond),
  56                         TimeoutIdle:     arvados.Duration(150 * time.Millisecond),
  57                         TimeoutBooting:  arvados.Duration(150 * time.Millisecond),
  58                         TimeoutProbe:    arvados.Duration(15 * time.Millisecond),
  59                         TimeoutShutdown: arvados.Duration(5 * time.Millisecond),
  60                 },
  61                 Dispatch: arvados.Dispatch{
  62                         PrivateKey:         dispatchprivraw,
  63                         PollInterval:       arvados.Duration(5 * time.Millisecond),
  64                         ProbeInterval:      arvados.Duration(5 * time.Millisecond),
  65                         StaleLockTimeout:   arvados.Duration(5 * time.Millisecond),
  66                         MaxProbesPerSecond: 1000,
  67                 },
  68                 InstanceTypes: arvados.InstanceTypeMap{
  69                         test.InstanceType(1).Name:  test.InstanceType(1),
  70                         test.InstanceType(2).Name:  test.InstanceType(2),
  71                         test.InstanceType(3).Name:  test.InstanceType(3),
  72                         test.InstanceType(4).Name:  test.InstanceType(4),
  73                         test.InstanceType(6).Name:  test.InstanceType(6),
  74                         test.InstanceType(8).Name:  test.InstanceType(8),
  75                         test.InstanceType(16).Name: test.InstanceType(16),
  76                 },
  77                 NodeProfiles: map[string]arvados.NodeProfile{
  78                         "*": {
  79                                 Controller:    arvados.SystemServiceInstance{Listen: os.Getenv("ARVADOS_API_HOST")},
  80                                 DispatchCloud: arvados.SystemServiceInstance{Listen: ":"},
  81                         },
  82                 },
  83         }
  84         s.disp = &dispatcher{Cluster: s.cluster}
  85         // Test cases can modify s.cluster before calling
  86         // initialize(), and then modify private state before calling
  87         // go run().
  88 }
  89
  90 func (s *DispatcherSuite) TearDownTest(c *check.C) {
  91         s.disp.Close()
  92 }
  93
  94 // DispatchToStubDriver checks that the dispatcher wires everything
  95 // together effectively. It uses a real scheduler and worker pool with
  96 // a fake queue and cloud driver. The fake cloud driver injects
  97 // artificial errors in order to exercise a variety of code paths.
  98 func (s *DispatcherSuite) TestDispatchToStubDriver(c *check.C) {
  99         drivers["test"] = s.stubDriver
 100         s.disp.setupOnce.Do(s.disp.initialize)
 101         queue := &test.Queue{
 102                 ChooseType: func(ctr *arvados.Container) (arvados.InstanceType, error) {
 103                         return ChooseInstanceType(s.cluster, ctr)
 104                 },
 105         }
 106         for i := 0; i < 200; i++ {
 107                 queue.Containers = append(queue.Containers, arvados.Container{
 108                         UUID:     test.ContainerUUID(i + 1),
 109                         State:    arvados.ContainerStateQueued,
 110                         Priority: int64(i%20 + 1),
 111                         RuntimeConstraints: arvados.RuntimeConstraints{
 112                                 RAM:   int64(i%3+1) << 30,
 113                                 VCPUs: i%8 + 1,
 114                         },
 115                 })
 116         }
 117         s.disp.queue = queue
 118
 119         var mtx sync.Mutex
 120         done := make(chan struct{})
 121         waiting := map[string]struct{}{}
 122         for _, ctr := range queue.Containers {
 123                 waiting[ctr.UUID] = struct{}{}
 124         }
 125         executeContainer := func(ctr arvados.Container) int {
 126                 mtx.Lock()
 127                 defer mtx.Unlock()
 128                 if _, ok := waiting[ctr.UUID]; !ok {
 129                         c.Logf("container completed twice: %s -- perhaps completed after stub instance was killed?", ctr.UUID)
 130                         return 1
 131                 }
 132                 delete(waiting, ctr.UUID)
 133                 if len(waiting) == 0 {
 134                         close(done)
 135                 }
 136                 return int(rand.Uint32() & 0x3)
 137         }
 138         n := 0
 139         s.stubDriver.Queue = queue
 140         s.stubDriver.SetupVM = func(stubvm *test.StubVM) {
 141                 n++
 142                 stubvm.Boot = time.Now().Add(time.Duration(rand.Int63n(int64(5 * time.Millisecond))))
 143                 stubvm.CrunchRunDetachDelay = time.Duration(rand.Int63n(int64(10 * time.Millisecond)))
 144                 stubvm.ExecuteContainer = executeContainer
 145                 switch n % 7 {
 146                 case 0:
 147                         stubvm.Broken = time.Now().Add(time.Duration(rand.Int63n(90)) * time.Millisecond)
 148                 case 1:
 149                         stubvm.CrunchRunMissing = true
 150                 default:
 151                         stubvm.CrunchRunCrashRate = 0.1
 152                 }
 153         }
 154
 155         start := time.Now()
 156         go s.disp.run()
 157         err := s.disp.CheckHealth()
 158         c.Check(err, check.IsNil)
 159
 160         select {
 161         case <-done:
 162                 c.Logf("containers finished (%s), waiting for instances to shutdown and queue to clear", time.Since(start))
 163         case <-time.After(10 * time.Second):
 164                 c.Fatalf("timed out; still waiting for %d containers: %q", len(waiting), waiting)
 165         }
 166
 167         deadline := time.Now().Add(time.Second)
 168         for range time.NewTicker(10 * time.Millisecond).C {
 169                 insts, err := s.stubDriver.InstanceSets()[0].Instances(nil)
 170                 c.Check(err, check.IsNil)
 171                 queue.Update()
 172                 ents, _ := queue.Entries()
 173                 if len(ents) == 0 && len(insts) == 0 {
 174                         break
 175                 }
 176                 if time.Now().After(deadline) {
 177                         c.Fatalf("timed out with %d containers (%v), %d instances (%+v)", len(ents), ents, len(insts), insts)
 178                 }
 179         }
 180 }
 181
 182 func (s *DispatcherSuite) TestAPIPermissions(c *check.C) {
 183         s.cluster.ManagementToken = "abcdefgh"
 184         drivers["test"] = s.stubDriver
 185         s.disp.setupOnce.Do(s.disp.initialize)
 186         s.disp.queue = &test.Queue{}
 187         go s.disp.run()
 188
 189         for _, token := range []string{"abc", ""} {
 190                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 191                 if token != "" {
 192                         req.Header.Set("Authorization", "Bearer "+token)
 193                 }
 194                 resp := httptest.NewRecorder()
 195                 s.disp.ServeHTTP(resp, req)
 196                 if token == "" {
 197                         c.Check(resp.Code, check.Equals, http.StatusUnauthorized)
 198                 } else {
 199                         c.Check(resp.Code, check.Equals, http.StatusForbidden)
 200                 }
 201         }
 202 }
 203
 204 func (s *DispatcherSuite) TestAPIDisabled(c *check.C) {
 205         s.cluster.ManagementToken = ""
 206         drivers["test"] = s.stubDriver
 207         s.disp.setupOnce.Do(s.disp.initialize)
 208         s.disp.queue = &test.Queue{}
 209         go s.disp.run()
 210
 211         for _, token := range []string{"abc", ""} {
 212                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 213                 if token != "" {
 214                         req.Header.Set("Authorization", "Bearer "+token)
 215                 }
 216                 resp := httptest.NewRecorder()
 217                 s.disp.ServeHTTP(resp, req)
 218                 c.Check(resp.Code, check.Equals, http.StatusForbidden)
 219         }
 220 }
 221
 222 func (s *DispatcherSuite) TestInstancesAPI(c *check.C) {
 223         s.cluster.ManagementToken = "abcdefgh"
 224         s.cluster.CloudVMs.TimeoutBooting = arvados.Duration(time.Second)
 225         drivers["test"] = s.stubDriver
 226         s.disp.setupOnce.Do(s.disp.initialize)
 227         s.disp.queue = &test.Queue{}
 228         go s.disp.run()
 229
 230         type instance struct {
 231                 Instance             string
 232                 WorkerState          string `json:"worker_state"`
 233                 Price                float64
 234                 LastContainerUUID    string `json:"last_container_uuid"`
 235                 ArvadosInstanceType  string `json:"arvados_instance_type"`
 236                 ProviderInstanceType string `json:"provider_instance_type"`
 237         }
 238         type instancesResponse struct {
 239                 Items []instance
 240         }
 241         getInstances := func() instancesResponse {
 242                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 243                 req.Header.Set("Authorization", "Bearer abcdefgh")
 244                 resp := httptest.NewRecorder()
 245                 s.disp.ServeHTTP(resp, req)
 246                 var sr instancesResponse
 247                 c.Check(resp.Code, check.Equals, http.StatusOK)
 248                 err := json.Unmarshal(resp.Body.Bytes(), &sr)
 249                 c.Check(err, check.IsNil)
 250                 return sr
 251         }
 252
 253         sr := getInstances()
 254         c.Check(len(sr.Items), check.Equals, 0)
 255
 256         ch := s.disp.pool.Subscribe()
 257         defer s.disp.pool.Unsubscribe(ch)
 258         ok := s.disp.pool.Create(test.InstanceType(1))
 259         c.Check(ok, check.Equals, true)
 260         <-ch
 261
 262         sr = getInstances()
 263         c.Assert(len(sr.Items), check.Equals, 1)
 264         c.Check(sr.Items[0].Instance, check.Matches, "stub.*")
 265         c.Check(sr.Items[0].WorkerState, check.Equals, "booting")
 266         c.Check(sr.Items[0].Price, check.Equals, 0.123)
 267         c.Check(sr.Items[0].LastContainerUUID, check.Equals, "")
 268         c.Check(sr.Items[0].ProviderInstanceType, check.Equals, test.InstanceType(1).ProviderType)
 269         c.Check(sr.Items[0].ArvadosInstanceType, check.Equals, test.InstanceType(1).Name)
 270 }