1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
23 "git.curoverse.com/arvados.git/lib/dispatchcloud"
24 "git.curoverse.com/arvados.git/sdk/go/arvados"
25 "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
26 "git.curoverse.com/arvados.git/sdk/go/arvadostest"
27 "git.curoverse.com/arvados.git/sdk/go/dispatch"
28 "github.com/sirupsen/logrus"
32 // Gocheck boilerplate
33 func Test(t *testing.T) {
37 var _ = Suite(&IntegrationSuite{})
38 var _ = Suite(&StubbedSuite{})
40 type IntegrationSuite struct {
45 func (s *IntegrationSuite) SetUpTest(c *C) {
46 arvadostest.StartAPI()
47 os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
49 s.disp.cluster = &arvados.Cluster{}
54 func (s *IntegrationSuite) TearDownTest(c *C) {
55 arvadostest.ResetEnv()
59 type slurmFake struct {
66 // If non-nil, run this func during the 2nd+ call to Cancel()
68 // Error returned by Batch()
72 func (sf *slurmFake) Batch(script io.Reader, args []string) error {
73 sf.didBatch = append(sf.didBatch, args)
77 func (sf *slurmFake) QueueCommand(args []string) *exec.Cmd {
78 return exec.Command("echo", sf.queue)
81 func (sf *slurmFake) Release(name string) error {
82 sf.didRelease = append(sf.didRelease, name)
86 func (sf *slurmFake) Renice(name string, nice int64) error {
87 sf.didRenice = append(sf.didRenice, []string{name, fmt.Sprintf("%d", nice)})
88 if sf.rejectNice10K && nice > 10000 {
89 return errors.New("scontrol: error: Invalid nice value, must be between -10000 and 10000")
94 func (sf *slurmFake) Cancel(name string) error {
95 sf.didCancel = append(sf.didCancel, name)
96 if len(sf.didCancel) == 1 {
97 // simulate error on first attempt
98 return errors.New("something terrible happened")
100 if sf.onCancel != nil {
106 func (s *IntegrationSuite) integrationTest(c *C,
107 expectBatch [][]string,
108 runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
109 arvadostest.ResetEnv()
111 arv, err := arvadosclient.MakeArvadosClient()
114 // There should be one queued container
115 params := arvadosclient.Dict{
116 "filters": [][]string{{"state", "=", "Queued"}},
118 var containers arvados.ContainerList
119 err = arv.List("containers", params, &containers)
121 c.Assert(len(containers.Items), Equals, 1)
123 s.disp.cluster.Containers.CrunchRunCommand = "echo"
125 ctx, cancel := context.WithCancel(context.Background())
126 doneRun := make(chan struct{})
128 s.disp.Dispatcher = &dispatch.Dispatcher{
130 PollPeriod: time.Second,
131 RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
133 runContainer(disp, ctr)
135 doneRun <- struct{}{}
137 s.disp.runContainer(disp, ctr, status)
142 s.disp.slurm = &s.slurm
143 s.disp.sqCheck = &SqueueChecker{
144 Logger: logrus.StandardLogger(),
145 Period: 500 * time.Millisecond,
149 err = s.disp.Dispatcher.Run(ctx)
151 c.Assert(err, Equals, context.Canceled)
153 s.disp.sqCheck.Stop()
155 c.Check(s.slurm.didBatch, DeepEquals, expectBatch)
157 // There should be no queued containers now
158 err = arv.List("containers", params, &containers)
160 c.Check(len(containers.Items), Equals, 0)
162 // Previously "Queued" container should now be in "Complete" state
163 var container arvados.Container
164 err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
169 func (s *IntegrationSuite) TestNormal(c *C) {
170 s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
171 container := s.integrationTest(c,
173 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
174 dispatcher.UpdateState(container.UUID, dispatch.Running)
175 time.Sleep(3 * time.Second)
176 dispatcher.UpdateState(container.UUID, dispatch.Complete)
178 c.Check(container.State, Equals, arvados.ContainerStateComplete)
181 func (s *IntegrationSuite) TestCancel(c *C) {
182 s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
183 readyToCancel := make(chan bool)
184 s.slurm.onCancel = func() { <-readyToCancel }
185 container := s.integrationTest(c,
187 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
188 dispatcher.UpdateState(container.UUID, dispatch.Running)
189 time.Sleep(time.Second)
190 dispatcher.Arv.Update("containers", container.UUID,
192 "container": arvadosclient.Dict{"priority": 0}},
194 readyToCancel <- true
197 c.Check(container.State, Equals, arvados.ContainerStateCancelled)
198 c.Check(len(s.slurm.didCancel) > 1, Equals, true)
199 c.Check(s.slurm.didCancel[:2], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "zzzzz-dz642-queuedcontainer"})
202 func (s *IntegrationSuite) TestMissingFromSqueue(c *C) {
203 container := s.integrationTest(c,
205 fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
206 fmt.Sprintf("--nice=%d", 10000),
208 fmt.Sprintf("--mem=%d", 11445),
209 fmt.Sprintf("--cpus-per-task=%d", 4),
210 fmt.Sprintf("--tmp=%d", 45777),
212 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
213 dispatcher.UpdateState(container.UUID, dispatch.Running)
214 time.Sleep(3 * time.Second)
215 dispatcher.UpdateState(container.UUID, dispatch.Complete)
217 c.Check(container.State, Equals, arvados.ContainerStateCancelled)
220 func (s *IntegrationSuite) TestSbatchFail(c *C) {
221 s.slurm = slurmFake{errBatch: errors.New("something terrible happened")}
222 container := s.integrationTest(c,
223 [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--nice=10000", "--no-requeue", "--mem=11445", "--cpus-per-task=4", "--tmp=45777"}},
224 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
225 dispatcher.UpdateState(container.UUID, dispatch.Running)
226 dispatcher.UpdateState(container.UUID, dispatch.Complete)
228 c.Check(container.State, Equals, arvados.ContainerStateComplete)
230 arv, err := arvadosclient.MakeArvadosClient()
233 var ll arvados.LogList
234 err = arv.List("logs", arvadosclient.Dict{"filters": [][]string{
235 {"object_uuid", "=", container.UUID},
236 {"event_type", "=", "dispatch"},
239 c.Assert(len(ll.Items), Equals, 1)
242 type StubbedSuite struct {
246 func (s *StubbedSuite) SetUpTest(c *C) {
247 s.disp = Dispatcher{}
248 s.disp.cluster = &arvados.Cluster{}
252 func (s *StubbedSuite) TestAPIErrorGettingContainers(c *C) {
253 apiStubResponses := make(map[string]arvadostest.StubResponse)
254 apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`}
255 apiStubResponses["/arvados/v1/containers"] = arvadostest.StubResponse{500, string(`{}`)}
257 s.testWithServerStub(c, apiStubResponses, "echo", "error getting count of containers")
260 func (s *StubbedSuite) testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
261 apiStub := arvadostest.ServerStub{apiStubResponses}
263 api := httptest.NewServer(&apiStub)
266 arv := &arvadosclient.ArvadosClient{
268 ApiServer: api.URL[7:],
270 Client: &http.Client{Transport: &http.Transport{}},
274 buf := bytes.NewBuffer(nil)
275 logrus.SetOutput(io.MultiWriter(buf, os.Stderr))
276 defer logrus.SetOutput(os.Stderr)
278 s.disp.cluster.Containers.CrunchRunCommand = "crunchCmd"
280 ctx, cancel := context.WithCancel(context.Background())
281 dispatcher := dispatch.Dispatcher{
283 PollPeriod: time.Second,
284 RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
286 time.Sleep(time.Second)
287 disp.UpdateState(ctr.UUID, dispatch.Running)
288 disp.UpdateState(ctr.UUID, dispatch.Complete)
290 s.disp.runContainer(disp, ctr, status)
296 for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ {
297 time.Sleep(100 * time.Millisecond)
302 err := dispatcher.Run(ctx)
303 c.Assert(err, Equals, context.Canceled)
305 c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
308 func (s *StubbedSuite) TestSbatchArgs(c *C) {
309 container := arvados.Container{
311 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
315 for _, defaults := range [][]string{
318 {"--arg1=v1", "--arg2"},
320 c.Logf("%#v", defaults)
321 s.disp.cluster.Containers.SLURM.SbatchArgumentsList = defaults
323 args, err := s.disp.sbatchArgs(container)
324 c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--nice=10000", "--no-requeue", "--mem=239", "--cpus-per-task=2", "--tmp=0"))
329 func (s *StubbedSuite) TestSbatchInstanceTypeConstraint(c *C) {
330 container := arvados.Container{
332 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
336 for _, trial := range []struct {
337 types map[string]arvados.InstanceType
341 // Choose node type => use --constraint arg
343 types: map[string]arvados.InstanceType{
344 "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
345 "a1.small": {Name: "a1.small", Price: 0.04, RAM: 256000000, VCPUs: 2},
346 "a1.medium": {Name: "a1.medium", Price: 0.08, RAM: 512000000, VCPUs: 4},
347 "a1.large": {Name: "a1.large", Price: 0.16, RAM: 1024000000, VCPUs: 8},
349 sbatchArgs: []string{"--constraint=instancetype=a1.medium"},
351 // No node types configured => no slurm constraint
354 sbatchArgs: []string{"--mem=239", "--cpus-per-task=2", "--tmp=0"},
356 // No node type is big enough => error
358 types: map[string]arvados.InstanceType{
359 "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
361 err: dispatchcloud.ConstraintsNotSatisfiableError{},
365 s.disp.cluster = &arvados.Cluster{InstanceTypes: trial.types}
367 args, err := s.disp.sbatchArgs(container)
368 c.Check(err == nil, Equals, trial.err == nil)
369 if trial.err == nil {
370 c.Check(args, DeepEquals, append([]string{"--job-name=123", "--nice=10000", "--no-requeue"}, trial.sbatchArgs...))
372 c.Check(len(err.(dispatchcloud.ConstraintsNotSatisfiableError).AvailableTypes), Equals, len(trial.types))
377 func (s *StubbedSuite) TestSbatchPartition(c *C) {
378 container := arvados.Container{
380 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
381 SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
385 args, err := s.disp.sbatchArgs(container)
386 c.Check(args, DeepEquals, []string{
387 "--job-name=123", "--nice=10000", "--no-requeue",
388 "--mem=239", "--cpus-per-task=1", "--tmp=0",
389 "--partition=blurb,b2",
394 func (s *StubbedSuite) TestLoadLegacyConfig(c *C) {
399 SbatchArguments: ["--foo", "bar"]
402 CrunchRunCommand: ["x-crunch-run", "--cgroup-parent-subsystem=memory"]
403 ReserveExtraRAM: 12345
407 tmpfile, err := ioutil.TempFile("", "example")
412 defer os.Remove(tmpfile.Name()) // clean up
414 if _, err := tmpfile.Write(content); err != nil {
417 if err := tmpfile.Close(); err != nil {
421 err = s.disp.configure("crunch-dispatch-slurm", []string{"-config", tmpfile.Name()})
424 c.Check(s.disp.cluster.Services.Controller.ExternalURL, Equals, arvados.URL{Scheme: "https", Host: "example.com"})
425 c.Check(s.disp.cluster.SystemRootToken, Equals, "abcdefg")
426 c.Check(s.disp.cluster.Containers.SLURM.SbatchArgumentsList, DeepEquals, []string{"--foo", "bar"})
427 c.Check(s.disp.cluster.Containers.CloudVMs.PollInterval, Equals, arvados.Duration(12*time.Second))
428 c.Check(s.disp.cluster.Containers.SLURM.PrioritySpread, Equals, int64(42))
429 c.Check(s.disp.cluster.Containers.CrunchRunCommand, Equals, "x-crunch-run")
430 c.Check(s.disp.cluster.Containers.CrunchRunArgumentsList, DeepEquals, []string{"--cgroup-parent-subsystem=memory"})
431 c.Check(s.disp.cluster.Containers.ReserveExtraRAM, Equals, arvados.ByteSize(12345))
432 c.Check(s.disp.cluster.Containers.MinRetryPeriod, Equals, arvados.Duration(13*time.Second))
433 c.Check(s.disp.cluster.API.MaxItemsPerResponse, Equals, 99)