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.arvados.org/arvados.git/lib/dispatchcloud/test"
  19         "git.arvados.org/arvados.git/sdk/go/arvados"
  20         "git.arvados.org/arvados.git/sdk/go/arvadostest"
  21         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  22         "github.com/prometheus/client_golang/prometheus"
  23         "golang.org/x/crypto/ssh"
  24         check "gopkg.in/check.v1"
  25 )
  26
  27 var _ = check.Suite(&DispatcherSuite{})
  28
  29 type DispatcherSuite struct {
  30         ctx        context.Context
  31         cancel     context.CancelFunc
  32         cluster    *arvados.Cluster
  33         stubDriver *test.StubDriver
  34         disp       *dispatcher
  35 }
  36
  37 func (s *DispatcherSuite) SetUpTest(c *check.C) {
  38         s.ctx, s.cancel = context.WithCancel(context.Background())
  39         s.ctx = ctxlog.Context(s.ctx, ctxlog.TestLogger(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                 ManagementToken: "test-management-token",
  54                 Containers: arvados.ContainersConfig{
  55                         DispatchPrivateKey: string(dispatchprivraw),
  56                         StaleLockTimeout:   arvados.Duration(5 * time.Millisecond),
  57                         CloudVMs: arvados.CloudVMsConfig{
  58                                 Driver:               "test",
  59                                 SyncInterval:         arvados.Duration(10 * time.Millisecond),
  60                                 TimeoutIdle:          arvados.Duration(150 * time.Millisecond),
  61                                 TimeoutBooting:       arvados.Duration(150 * time.Millisecond),
  62                                 TimeoutProbe:         arvados.Duration(15 * time.Millisecond),
  63                                 TimeoutShutdown:      arvados.Duration(5 * time.Millisecond),
  64                                 MaxCloudOpsPerSecond: 500,
  65                                 PollInterval:         arvados.Duration(5 * time.Millisecond),
  66                                 ProbeInterval:        arvados.Duration(5 * time.Millisecond),
  67                                 MaxProbesPerSecond:   1000,
  68                                 TimeoutSignal:        arvados.Duration(3 * time.Millisecond),
  69                                 TimeoutTERM:          arvados.Duration(20 * time.Millisecond),
  70                                 ResourceTags:         map[string]string{"testtag": "test value"},
  71                                 TagKeyPrefix:         "test:",
  72                         },
  73                 },
  74                 InstanceTypes: arvados.InstanceTypeMap{
  75                         test.InstanceType(1).Name:  test.InstanceType(1),
  76                         test.InstanceType(2).Name:  test.InstanceType(2),
  77                         test.InstanceType(3).Name:  test.InstanceType(3),
  78                         test.InstanceType(4).Name:  test.InstanceType(4),
  79                         test.InstanceType(6).Name:  test.InstanceType(6),
  80                         test.InstanceType(8).Name:  test.InstanceType(8),
  81                         test.InstanceType(16).Name: test.InstanceType(16),
  82                 },
  83         }
  84         arvadostest.SetServiceURL(&s.cluster.Services.DispatchCloud, "http://localhost:/")
  85         arvadostest.SetServiceURL(&s.cluster.Services.Controller, "https://"+os.Getenv("ARVADOS_API_HOST")+"/")
  86
  87         arvClient, err := arvados.NewClientFromConfig(s.cluster)
  88         c.Check(err, check.IsNil)
  89
  90         s.disp = &dispatcher{
  91                 Cluster:   s.cluster,
  92                 Context:   s.ctx,
  93                 ArvClient: arvClient,
  94                 AuthToken: arvadostest.AdminToken,
  95                 Registry:  prometheus.NewRegistry(),
  96         }
  97         // Test cases can modify s.cluster before calling
  98         // initialize(), and then modify private state before calling
  99         // go run().
 100 }
 101
 102 func (s *DispatcherSuite) TearDownTest(c *check.C) {
 103         s.cancel()
 104         s.disp.Close()
 105 }
 106
 107 // DispatchToStubDriver checks that the dispatcher wires everything
 108 // together effectively. It uses a real scheduler and worker pool with
 109 // a fake queue and cloud driver. The fake cloud driver injects
 110 // artificial errors in order to exercise a variety of code paths.
 111 func (s *DispatcherSuite) TestDispatchToStubDriver(c *check.C) {
 112         Drivers["test"] = s.stubDriver
 113         s.disp.setupOnce.Do(s.disp.initialize)
 114         queue := &test.Queue{
 115                 ChooseType: func(ctr *arvados.Container) (arvados.InstanceType, error) {
 116                         return ChooseInstanceType(s.cluster, ctr)
 117                 },
 118                 Logger: ctxlog.TestLogger(c),
 119         }
 120         for i := 0; i < 200; i++ {
 121                 queue.Containers = append(queue.Containers, arvados.Container{
 122                         UUID:     test.ContainerUUID(i + 1),
 123                         State:    arvados.ContainerStateQueued,
 124                         Priority: int64(i%20 + 1),
 125                         RuntimeConstraints: arvados.RuntimeConstraints{
 126                                 RAM:   int64(i%3+1) << 30,
 127                                 VCPUs: i%8 + 1,
 128                         },
 129                 })
 130         }
 131         s.disp.queue = queue
 132
 133         var mtx sync.Mutex
 134         done := make(chan struct{})
 135         waiting := map[string]struct{}{}
 136         for _, ctr := range queue.Containers {
 137                 waiting[ctr.UUID] = struct{}{}
 138         }
 139         finishContainer := func(ctr arvados.Container) {
 140                 mtx.Lock()
 141                 defer mtx.Unlock()
 142                 if _, ok := waiting[ctr.UUID]; !ok {
 143                         c.Errorf("container completed twice: %s", ctr.UUID)
 144                         return
 145                 }
 146                 delete(waiting, ctr.UUID)
 147                 if len(waiting) == 0 {
 148                         close(done)
 149                 }
 150         }
 151         executeContainer := func(ctr arvados.Container) int {
 152                 finishContainer(ctr)
 153                 return int(rand.Uint32() & 0x3)
 154         }
 155         n := 0
 156         s.stubDriver.Queue = queue
 157         s.stubDriver.SetupVM = func(stubvm *test.StubVM) {
 158                 n++
 159                 stubvm.Boot = time.Now().Add(time.Duration(rand.Int63n(int64(5 * time.Millisecond))))
 160                 stubvm.CrunchRunDetachDelay = time.Duration(rand.Int63n(int64(10 * time.Millisecond)))
 161                 stubvm.ExecuteContainer = executeContainer
 162                 stubvm.CrashRunningContainer = finishContainer
 163                 switch n % 7 {
 164                 case 0:
 165                         stubvm.Broken = time.Now().Add(time.Duration(rand.Int63n(90)) * time.Millisecond)
 166                 case 1:
 167                         stubvm.CrunchRunMissing = true
 168                 case 2:
 169                         stubvm.ReportBroken = time.Now().Add(time.Duration(rand.Int63n(200)) * time.Millisecond)
 170                 default:
 171                         stubvm.CrunchRunCrashRate = 0.1
 172                 }
 173         }
 174
 175         start := time.Now()
 176         go s.disp.run()
 177         err := s.disp.CheckHealth()
 178         c.Check(err, check.IsNil)
 179
 180         select {
 181         case <-done:
 182                 c.Logf("containers finished (%s), waiting for instances to shutdown and queue to clear", time.Since(start))
 183         case <-time.After(10 * time.Second):
 184                 c.Fatalf("timed out; still waiting for %d containers: %q", len(waiting), waiting)
 185         }
 186
 187         deadline := time.Now().Add(5 * time.Second)
 188         for range time.NewTicker(10 * time.Millisecond).C {
 189                 insts, err := s.stubDriver.InstanceSets()[0].Instances(nil)
 190                 c.Check(err, check.IsNil)
 191                 queue.Update()
 192                 ents, _ := queue.Entries()
 193                 if len(ents) == 0 && len(insts) == 0 {
 194                         break
 195                 }
 196                 if time.Now().After(deadline) {
 197                         c.Fatalf("timed out with %d containers (%v), %d instances (%+v)", len(ents), ents, len(insts), insts)
 198                 }
 199         }
 200
 201         req := httptest.NewRequest("GET", "/metrics", nil)
 202         req.Header.Set("Authorization", "Bearer "+s.cluster.ManagementToken)
 203         resp := httptest.NewRecorder()
 204         s.disp.ServeHTTP(resp, req)
 205         c.Check(resp.Code, check.Equals, http.StatusOK)
 206         c.Check(resp.Body.String(), check.Matches, `(?ms).*driver_operations{error="0",operation="Create"} [^0].*`)
 207         c.Check(resp.Body.String(), check.Matches, `(?ms).*driver_operations{error="0",operation="List"} [^0].*`)
 208         c.Check(resp.Body.String(), check.Matches, `(?ms).*driver_operations{error="0",operation="Destroy"} [^0].*`)
 209         c.Check(resp.Body.String(), check.Matches, `(?ms).*driver_operations{error="1",operation="Create"} [^0].*`)
 210         c.Check(resp.Body.String(), check.Matches, `(?ms).*driver_operations{error="1",operation="List"} 0\n.*`)
 211         c.Check(resp.Body.String(), check.Matches, `(?ms).*boot_outcomes{outcome="aborted"} 0.*`)
 212         c.Check(resp.Body.String(), check.Matches, `(?ms).*boot_outcomes{outcome="disappeared"} [^0].*`)
 213         c.Check(resp.Body.String(), check.Matches, `(?ms).*boot_outcomes{outcome="failure"} [^0].*`)
 214         c.Check(resp.Body.String(), check.Matches, `(?ms).*boot_outcomes{outcome="success"} [^0].*`)
 215         c.Check(resp.Body.String(), check.Matches, `(?ms).*instances_disappeared{state="shutdown"} [^0].*`)
 216         c.Check(resp.Body.String(), check.Matches, `(?ms).*instances_disappeared{state="unknown"} 0\n.*`)
 217 }
 218
 219 func (s *DispatcherSuite) TestAPIPermissions(c *check.C) {
 220         s.cluster.ManagementToken = "abcdefgh"
 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                 if token == "" {
 234                         c.Check(resp.Code, check.Equals, http.StatusUnauthorized)
 235                 } else {
 236                         c.Check(resp.Code, check.Equals, http.StatusForbidden)
 237                 }
 238         }
 239 }
 240
 241 func (s *DispatcherSuite) TestAPIDisabled(c *check.C) {
 242         s.cluster.ManagementToken = ""
 243         Drivers["test"] = s.stubDriver
 244         s.disp.setupOnce.Do(s.disp.initialize)
 245         s.disp.queue = &test.Queue{}
 246         go s.disp.run()
 247
 248         for _, token := range []string{"abc", ""} {
 249                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 250                 if token != "" {
 251                         req.Header.Set("Authorization", "Bearer "+token)
 252                 }
 253                 resp := httptest.NewRecorder()
 254                 s.disp.ServeHTTP(resp, req)
 255                 c.Check(resp.Code, check.Equals, http.StatusForbidden)
 256         }
 257 }
 258
 259 func (s *DispatcherSuite) TestInstancesAPI(c *check.C) {
 260         s.cluster.ManagementToken = "abcdefgh"
 261         s.cluster.Containers.CloudVMs.TimeoutBooting = arvados.Duration(time.Second)
 262         Drivers["test"] = s.stubDriver
 263         s.disp.setupOnce.Do(s.disp.initialize)
 264         s.disp.queue = &test.Queue{}
 265         go s.disp.run()
 266
 267         type instance struct {
 268                 Instance             string
 269                 WorkerState          string `json:"worker_state"`
 270                 Price                float64
 271                 LastContainerUUID    string `json:"last_container_uuid"`
 272                 ArvadosInstanceType  string `json:"arvados_instance_type"`
 273                 ProviderInstanceType string `json:"provider_instance_type"`
 274         }
 275         type instancesResponse struct {
 276                 Items []instance
 277         }
 278         getInstances := func() instancesResponse {
 279                 req := httptest.NewRequest("GET", "/arvados/v1/dispatch/instances", nil)
 280                 req.Header.Set("Authorization", "Bearer abcdefgh")
 281                 resp := httptest.NewRecorder()
 282                 s.disp.ServeHTTP(resp, req)
 283                 var sr instancesResponse
 284                 c.Check(resp.Code, check.Equals, http.StatusOK)
 285                 err := json.Unmarshal(resp.Body.Bytes(), &sr)
 286                 c.Check(err, check.IsNil)
 287                 return sr
 288         }
 289
 290         sr := getInstances()
 291         c.Check(len(sr.Items), check.Equals, 0)
 292
 293         ch := s.disp.pool.Subscribe()
 294         defer s.disp.pool.Unsubscribe(ch)
 295         ok := s.disp.pool.Create(test.InstanceType(1))
 296         c.Check(ok, check.Equals, true)
 297         <-ch
 298
 299         for deadline := time.Now().Add(time.Second); time.Now().Before(deadline); {
 300                 sr = getInstances()
 301                 if len(sr.Items) > 0 {
 302                         break
 303                 }
 304                 time.Sleep(time.Millisecond)
 305         }
 306         c.Assert(len(sr.Items), check.Equals, 1)
 307         c.Check(sr.Items[0].Instance, check.Matches, "stub.*")
 308         c.Check(sr.Items[0].WorkerState, check.Equals, "booting")
 309         c.Check(sr.Items[0].Price, check.Equals, 0.123)
 310         c.Check(sr.Items[0].LastContainerUUID, check.Equals, "")
 311         c.Check(sr.Items[0].ProviderInstanceType, check.Equals, test.InstanceType(1).ProviderType)
 312         c.Check(sr.Items[0].ArvadosInstanceType, check.Equals, test.InstanceType(1).Name)
 313 }