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         "context"
   9         "encoding/json"
  10         "io/ioutil"
  11         "math/rand"
  12         "net/http"
  13         "net/http/httptest"
  14         "os"
  15         "sync"
  16         "time"
  17
  18         "git.curoverse.com/arvados.git/lib/dispatchcloud/test"
  19         "git.curoverse.com/arvados.git/sdk/go/arvados"
  20         "git.curoverse.com/arvados.git/sdk/go/arvadostest"
  21         "git.curoverse.com/arvados.git/sdk/go/ctxlog"
  22         "golang.org/x/crypto/ssh"
  23         check "gopkg.in/check.v1"
  24 )
  25
  26 var _ = check.Suite(&DispatcherSuite{})
  27
  28 type DispatcherSuite struct {
  29         ctx        context.Context
  30         cancel     context.CancelFunc
  31         cluster    *arvados.Cluster
  32         stubDriver *test.StubDriver
  33         disp       *dispatcher
  34 }
  35
  36 func (s *DispatcherSuite) SetUpTest(c *check.C) {
  37         s.ctx, s.cancel = context.WithCancel(context.Background())
  38         s.ctx = ctxlog.Context(s.ctx, ctxlog.TestLogger(c))
  39         dispatchpub, _ := test.LoadTestKey(c, "test/sshkey_dispatch")
  40         dispatchprivraw, err := ioutil.ReadFile("test/sshkey_dispatch")
  41         c.Assert(err, check.IsNil)
  42
  43         _, hostpriv := test.LoadTestKey(c, "test/sshkey_vm")
  44         s.stubDriver = &test.StubDriver{
  45                 HostKey:                   hostpriv,
  46                 AuthorizedKeys:            []ssh.PublicKey{dispatchpub},
  47                 ErrorRateDestroy:          0.1,
  48                 MinTimeBetweenCreateCalls: time.Millisecond,
  49         }
  50
  51         s.cluster = &arvados.Cluster{
  52                 Containers: arvados.ContainersConfig{
  53                         DispatchPrivateKey: string(dispatchprivraw),
  54                         StaleLockTimeout:   arvados.Duration(5 * time.Millisecond),
  55                         CloudVMs: arvados.CloudVMsConfig{
  56                                 Driver:               "test",
  57                                 SyncInterval:         arvados.Duration(10 * time.Millisecond),
  58                                 TimeoutIdle:          arvados.Duration(150 * time.Millisecond),
  59                                 TimeoutBooting:       arvados.Duration(150 * time.Millisecond),
  60                                 TimeoutProbe:         arvados.Duration(15 * time.Millisecond),
  61                                 TimeoutShutdown:      arvados.Duration(5 * time.Millisecond),
  62                                 MaxCloudOpsPerSecond: 500,
  63                                 PollInterval:         arvados.Duration(5 * time.Millisecond),
  64                                 ProbeInterval:        arvados.Duration(5 * time.Millisecond),
  65                                 MaxProbesPerSecond:   1000,
  66                                 TimeoutSignal:        arvados.Duration(3 * time.Millisecond),
  67                                 TimeoutTERM:          arvados.Duration(20 * time.Millisecond),
  68                                 ResourceTags:         map[string]string{"testtag": "test value"},
  69                         },
  70                 },
  71                 InstanceTypes: arvados.InstanceTypeMap{
  72                         test.InstanceType(1).Name:  test.InstanceType(1),
  73                         test.InstanceType(2).Name:  test.InstanceType(2),
  74                         test.InstanceType(3).Name:  test.InstanceType(3),
  75                         test.InstanceType(4).Name:  test.InstanceType(4),
  76                         test.InstanceType(6).Name:  test.InstanceType(6),
  77                         test.InstanceType(8).Name:  test.InstanceType(8),
  78                         test.InstanceType(16).Name: test.InstanceType(16),
  79                 },
  80         }
  81         arvadostest.SetServiceURL(&s.cluster.Services.DispatchCloud, "http://localhost:/")
  82         arvadostest.SetServiceURL(&s.cluster.Services.Controller, "https://"+os.Getenv("ARVADOS_API_HOST")+"/")
  83
  84         arvClient, err := arvados.NewClientFromConfig(s.cluster)
  85         c.Check(err, check.IsNil)
  86
  87         s.disp = &dispatcher{
  88                 Cluster:   s.cluster,
  89                 Context:   s.ctx,
  90                 ArvClient: arvClient,
  91                 AuthToken: arvadostest.AdminToken,
  92         }
  93         // Test cases can modify s.cluster before calling
  94         // initialize(), and then modify private state before calling
  95         // go run().
  96 }
  97
  98 func (s *DispatcherSuite) TearDownTest(c *check.C) {
  99         s.cancel()
 100         s.disp.Close()
 101 }
 102
 103 // DispatchToStubDriver checks that the dispatcher wires everything
 104 // together effectively. It uses a real scheduler and worker pool with
 105 // a fake queue and cloud driver. The fake cloud driver injects
 106 // artificial errors in order to exercise a variety of code paths.
 107 func (s *DispatcherSuite) TestDispatchToStubDriver(c *check.C) {
 108         drivers["test"] = s.stubDriver
 109         s.disp.setupOnce.Do(s.disp.initialize)
 110         queue := &test.Queue{
 111                 ChooseType: func(ctr *arvados.Container) (arvados.InstanceType, error) {
 112                         return ChooseInstanceType(s.cluster, ctr)
 113                 },
 114         }
 115         for i := 0; i < 200; i++ {
 116                 queue.Containers = append(queue.Containers, arvados.Container{
 117                         UUID:     test.ContainerUUID(i + 1),
 118                         State:    arvados.ContainerStateQueued,
 119                         Priority: int64(i%20 + 1),
 120                         RuntimeConstraints: arvados.RuntimeConstraints{
 121                                 RAM:   int64(i%3+1) << 30,
 122                                 VCPUs: i%8 + 1,
 123                         },
 124                 })
 125         }
 126         s.disp.queue = queue
 127
 128         var mtx sync.Mutex
 129         done := make(chan struct{})
 130         waiting := map[string]struct{}{}
 131         for _, ctr := range queue.Containers {
 132                 waiting[ctr.UUID] = struct{}{}
 133         }
 134         finishContainer := func(ctr arvados.Container) {
 135                 mtx.Lock()
 136                 defer mtx.Unlock()
 137                 if _, ok := waiting[ctr.UUID]; !ok {
 138                         c.Errorf("container completed twice: %s", ctr.UUID)
 139                         return
 140                 }
 141                 delete(waiting, ctr.UUID)
 142                 if len(waiting) == 0 {
 143                         close(done)
 144                 }
 145         }
 146         executeContainer := func(ctr arvados.Container) int {
 147                 finishContainer(ctr)
 148                 return int(rand.Uint32() & 0x3)
 149         }
 150         n := 0
 151         s.stubDriver.Queue = queue
 152         s.stubDriver.SetupVM = func(stubvm *test.StubVM) {
 153                 n++
 154                 stubvm.Boot = time.Now().Add(time.Duration(rand.Int63n(int64(5 * time.Millisecond))))
 155                 stubvm.CrunchRunDetachDelay = time.Duration(rand.Int63n(int64(10 * time.Millisecond)))
 156                 stubvm.ExecuteContainer = executeContainer
 157                 stubvm.CrashRunningContainer = finishContainer
 158                 switch n % 7 {
 159                 case 0:
 160                         stubvm.Broken = time.Now().Add(time.Duration(rand.Int63n(90)) * time.Millisecond)
 161                 case 1:
 162                         stubvm.CrunchRunMissing = true
 163                 case 2:
 164                         stubvm.ReportBroken = time.Now().Add(time.Duration(rand.Int63n(200)) * time.Millisecond)
 165                 default:
 166                         stubvm.CrunchRunCrashRate = 0.1
 167                 }
 168         }
 169
 170         start := time.Now()
 171         go s.disp.run()
 172         err := s.disp.CheckHealth()
 173         c.Check(err, check.IsNil)
 174
 175         select {
 176         case <-done:
 177                 c.Logf("containers finished (%s), waiting for instances to shutdown and queue to clear", time.Since(start))
 178         case <-time.After(10 * time.Second):
 179                 c.Fatalf("timed out; still waiting for %d containers: %q", len(waiting), waiting)
 180         }
 181
 182         deadline := time.Now().Add(5 * time.Second)
 183         for range time.NewTicker(10 * time.Millisecond).C {
 184                 insts, err := s.stubDriver.InstanceSets()[0].Instances(nil)
 185                 c.Check(err, check.IsNil)
 186                 queue.Update()
 187                 ents, _ := queue.Entries()
 188                 if len(ents) == 0 && len(insts) == 0 {
 189                         break
 190                 }
 191                 if time.Now().After(deadline) {
 192                         c.Fatalf("timed out with %d containers (%v), %d instances (%+v)", len(ents), ents, len(insts), insts)
 193                 }
 194         }
 195 }
 196
 197 func (s *DispatcherSuite) TestAPIPermissions(c *check.C) {
 198         s.cluster.ManagementToken = "abcdefgh"
 199         drivers["test"] = s.stubDriver
 200         s.disp.setupOnce.Do(s.disp.initialize)
 201         s.disp.queue = &test.Queue{}
 202         go s.disp.run()
 203
 204         for _, token := range []string{"abc", ""} {
 205                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 206                 if token != "" {
 207                         req.Header.Set("Authorization", "Bearer "+token)
 208                 }
 209                 resp := httptest.NewRecorder()
 210                 s.disp.ServeHTTP(resp, req)
 211                 if token == "" {
 212                         c.Check(resp.Code, check.Equals, http.StatusUnauthorized)
 213                 } else {
 214                         c.Check(resp.Code, check.Equals, http.StatusForbidden)
 215                 }
 216         }
 217 }
 218
 219 func (s *DispatcherSuite) TestAPIDisabled(c *check.C) {
 220         s.cluster.ManagementToken = ""
 221         drivers["test"] = s.stubDriver
 222         s.disp.setupOnce.Do(s.disp.initialize)
 223         s.disp.queue = &test.Queue{}
 224         go s.disp.run()
 225
 226         for _, token := range []string{"abc", ""} {
 227                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 228                 if token != "" {
 229                         req.Header.Set("Authorization", "Bearer "+token)
 230                 }
 231                 resp := httptest.NewRecorder()
 232                 s.disp.ServeHTTP(resp, req)
 233                 c.Check(resp.Code, check.Equals, http.StatusForbidden)
 234         }
 235 }
 236
 237 func (s *DispatcherSuite) TestInstancesAPI(c *check.C) {
 238         s.cluster.ManagementToken = "abcdefgh"
 239         s.cluster.Containers.CloudVMs.TimeoutBooting = arvados.Duration(time.Second)
 240         drivers["test"] = s.stubDriver
 241         s.disp.setupOnce.Do(s.disp.initialize)
 242         s.disp.queue = &test.Queue{}
 243         go s.disp.run()
 244
 245         type instance struct {
 246                 Instance             string
 247                 WorkerState          string `json:"worker_state"`
 248                 Price                float64
 249                 LastContainerUUID    string `json:"last_container_uuid"`
 250                 ArvadosInstanceType  string `json:"arvados_instance_type"`
 251                 ProviderInstanceType string `json:"provider_instance_type"`
 252         }
 253         type instancesResponse struct {
 254                 Items []instance
 255         }
 256         getInstances := func() instancesResponse {
 257                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 258                 req.Header.Set("Authorization", "Bearer abcdefgh")
 259                 resp := httptest.NewRecorder()
 260                 s.disp.ServeHTTP(resp, req)
 261                 var sr instancesResponse
 262                 c.Check(resp.Code, check.Equals, http.StatusOK)
 263                 err := json.Unmarshal(resp.Body.Bytes(), &sr)
 264                 c.Check(err, check.IsNil)
 265                 return sr
 266         }
 267
 268         sr := getInstances()
 269         c.Check(len(sr.Items), check.Equals, 0)
 270
 271         ch := s.disp.pool.Subscribe()
 272         defer s.disp.pool.Unsubscribe(ch)
 273         ok := s.disp.pool.Create(test.InstanceType(1))
 274         c.Check(ok, check.Equals, true)
 275         <-ch
 276
 277         for deadline := time.Now().Add(time.Second); time.Now().Before(deadline); {
 278                 sr = getInstances()
 279                 if len(sr.Items) > 0 {
 280                         break
 281                 }
 282                 time.Sleep(time.Millisecond)
 283         }
 284         c.Assert(len(sr.Items), check.Equals, 1)
 285         c.Check(sr.Items[0].Instance, check.Matches, "stub.*")
 286         c.Check(sr.Items[0].WorkerState, check.Equals, "booting")
 287         c.Check(sr.Items[0].Price, check.Equals, 0.123)
 288         c.Check(sr.Items[0].LastContainerUUID, check.Equals, "")
 289         c.Check(sr.Items[0].ProviderInstanceType, check.Equals, test.InstanceType(1).ProviderType)
 290         c.Check(sr.Items[0].ArvadosInstanceType, check.Equals, test.InstanceType(1).Name)
 291 }