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"
31 // Gocheck boilerplate
32 func Test(t *testing.T) {
36 var _ = Suite(&IntegrationSuite{})
37 var _ = Suite(&StubbedSuite{})
39 type IntegrationSuite struct {
44 func (s *IntegrationSuite) SetUpTest(c *C) {
45 arvadostest.StartAPI()
46 os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
52 func (s *IntegrationSuite) TearDownTest(c *C) {
53 arvadostest.ResetEnv()
57 type slurmFake struct {
64 // If non-nil, run this func during the 2nd+ call to Cancel()
66 // Error returned by Batch()
70 func (sf *slurmFake) Batch(script io.Reader, args []string) error {
71 sf.didBatch = append(sf.didBatch, args)
75 func (sf *slurmFake) QueueCommand(args []string) *exec.Cmd {
76 return exec.Command("echo", sf.queue)
79 func (sf *slurmFake) Release(name string) error {
80 sf.didRelease = append(sf.didRelease, name)
84 func (sf *slurmFake) Renice(name string, nice int64) error {
85 sf.didRenice = append(sf.didRenice, []string{name, fmt.Sprintf("%d", nice)})
86 if sf.rejectNice10K && nice > 10000 {
87 return errors.New("scontrol: error: Invalid nice value, must be between -10000 and 10000")
92 func (sf *slurmFake) Cancel(name string) error {
93 sf.didCancel = append(sf.didCancel, name)
94 if len(sf.didCancel) == 1 {
95 // simulate error on first attempt
96 return errors.New("something terrible happened")
98 if sf.onCancel != nil {
104 func (s *IntegrationSuite) integrationTest(c *C,
105 expectBatch [][]string,
106 runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
107 arvadostest.ResetEnv()
109 arv, err := arvadosclient.MakeArvadosClient()
112 // There should be one queued container
113 params := arvadosclient.Dict{
114 "filters": [][]string{{"state", "=", "Queued"}},
116 var containers arvados.ContainerList
117 err = arv.List("containers", params, &containers)
119 c.Assert(len(containers.Items), Equals, 1)
121 s.disp.CrunchRunCommand = []string{"echo"}
123 ctx, cancel := context.WithCancel(context.Background())
124 doneRun := make(chan struct{})
126 s.disp.Dispatcher = &dispatch.Dispatcher{
128 PollPeriod: time.Second,
129 RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
131 runContainer(disp, ctr)
133 doneRun <- struct{}{}
135 s.disp.runContainer(disp, ctr, status)
140 s.disp.slurm = &s.slurm
141 s.disp.sqCheck = &SqueueChecker{Period: 500 * time.Millisecond, Slurm: s.disp.slurm}
143 err = s.disp.Dispatcher.Run(ctx)
145 c.Assert(err, Equals, context.Canceled)
147 s.disp.sqCheck.Stop()
149 c.Check(s.slurm.didBatch, DeepEquals, expectBatch)
151 // There should be no queued containers now
152 err = arv.List("containers", params, &containers)
154 c.Check(len(containers.Items), Equals, 0)
156 // Previously "Queued" container should now be in "Complete" state
157 var container arvados.Container
158 err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
163 func (s *IntegrationSuite) TestNormal(c *C) {
164 s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
165 container := s.integrationTest(c,
167 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
168 dispatcher.UpdateState(container.UUID, dispatch.Running)
169 time.Sleep(3 * time.Second)
170 dispatcher.UpdateState(container.UUID, dispatch.Complete)
172 c.Check(container.State, Equals, arvados.ContainerStateComplete)
175 func (s *IntegrationSuite) TestCancel(c *C) {
176 s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
177 readyToCancel := make(chan bool)
178 s.slurm.onCancel = func() { <-readyToCancel }
179 container := s.integrationTest(c,
181 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
182 dispatcher.UpdateState(container.UUID, dispatch.Running)
183 time.Sleep(time.Second)
184 dispatcher.Arv.Update("containers", container.UUID,
186 "container": arvadosclient.Dict{"priority": 0}},
188 readyToCancel <- true
191 c.Check(container.State, Equals, arvados.ContainerStateCancelled)
192 c.Check(len(s.slurm.didCancel) > 1, Equals, true)
193 c.Check(s.slurm.didCancel[:2], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "zzzzz-dz642-queuedcontainer"})
196 func (s *IntegrationSuite) TestMissingFromSqueue(c *C) {
197 container := s.integrationTest(c,
199 fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
200 fmt.Sprintf("--nice=%d", 10000),
201 fmt.Sprintf("--mem=%d", 11445),
202 fmt.Sprintf("--cpus-per-task=%d", 4),
203 fmt.Sprintf("--tmp=%d", 45777),
205 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
206 dispatcher.UpdateState(container.UUID, dispatch.Running)
207 time.Sleep(3 * time.Second)
208 dispatcher.UpdateState(container.UUID, dispatch.Complete)
210 c.Check(container.State, Equals, arvados.ContainerStateCancelled)
213 func (s *IntegrationSuite) TestSbatchFail(c *C) {
214 s.slurm = slurmFake{errBatch: errors.New("something terrible happened")}
215 container := s.integrationTest(c,
216 [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--nice=10000", "--mem=11445", "--cpus-per-task=4", "--tmp=45777"}},
217 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
218 dispatcher.UpdateState(container.UUID, dispatch.Running)
219 dispatcher.UpdateState(container.UUID, dispatch.Complete)
221 c.Check(container.State, Equals, arvados.ContainerStateComplete)
223 arv, err := arvadosclient.MakeArvadosClient()
226 var ll arvados.LogList
227 err = arv.List("logs", arvadosclient.Dict{"filters": [][]string{
228 {"object_uuid", "=", container.UUID},
229 {"event_type", "=", "dispatch"},
232 c.Assert(len(ll.Items), Equals, 1)
235 type StubbedSuite struct {
239 func (s *StubbedSuite) SetUpTest(c *C) {
240 s.disp = Dispatcher{}
244 func (s *StubbedSuite) TestAPIErrorGettingContainers(c *C) {
245 apiStubResponses := make(map[string]arvadostest.StubResponse)
246 apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`}
247 apiStubResponses["/arvados/v1/containers"] = arvadostest.StubResponse{500, string(`{}`)}
249 s.testWithServerStub(c, apiStubResponses, "echo", "error getting count of containers")
252 func (s *StubbedSuite) testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
253 apiStub := arvadostest.ServerStub{apiStubResponses}
255 api := httptest.NewServer(&apiStub)
258 arv := &arvadosclient.ArvadosClient{
260 ApiServer: api.URL[7:],
262 Client: &http.Client{Transport: &http.Transport{}},
266 buf := bytes.NewBuffer(nil)
267 log.SetOutput(io.MultiWriter(buf, os.Stderr))
268 defer log.SetOutput(os.Stderr)
270 s.disp.CrunchRunCommand = []string{crunchCmd}
272 ctx, cancel := context.WithCancel(context.Background())
273 dispatcher := dispatch.Dispatcher{
275 PollPeriod: time.Second,
276 RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
278 time.Sleep(time.Second)
279 disp.UpdateState(ctr.UUID, dispatch.Running)
280 disp.UpdateState(ctr.UUID, dispatch.Complete)
282 s.disp.runContainer(disp, ctr, status)
288 for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ {
289 time.Sleep(100 * time.Millisecond)
294 err := dispatcher.Run(ctx)
295 c.Assert(err, Equals, context.Canceled)
297 c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
300 func (s *StubbedSuite) TestNoSuchConfigFile(c *C) {
301 err := s.disp.readConfig("/nosuchdir89j7879/8hjwr7ojgyy7")
302 c.Assert(err, NotNil)
305 func (s *StubbedSuite) TestBadSbatchArgsConfig(c *C) {
306 tmpfile, err := ioutil.TempFile(os.TempDir(), "config")
308 defer os.Remove(tmpfile.Name())
310 _, err = tmpfile.Write([]byte(`{"SbatchArguments": "oops this is not a string array"}`))
313 err = s.disp.readConfig(tmpfile.Name())
314 c.Assert(err, NotNil)
317 func (s *StubbedSuite) TestNoSuchArgInConfigIgnored(c *C) {
318 tmpfile, err := ioutil.TempFile(os.TempDir(), "config")
320 defer os.Remove(tmpfile.Name())
322 _, err = tmpfile.Write([]byte(`{"NoSuchArg": "Nobody loves me, not one tiny hunk."}`))
325 err = s.disp.readConfig(tmpfile.Name())
327 c.Check(0, Equals, len(s.disp.SbatchArguments))
330 func (s *StubbedSuite) TestReadConfig(c *C) {
331 tmpfile, err := ioutil.TempFile(os.TempDir(), "config")
333 defer os.Remove(tmpfile.Name())
335 args := []string{"--arg1=v1", "--arg2", "--arg3=v3"}
336 argsS := `{"SbatchArguments": ["--arg1=v1", "--arg2", "--arg3=v3"]}`
337 _, err = tmpfile.Write([]byte(argsS))
340 err = s.disp.readConfig(tmpfile.Name())
342 c.Check(args, DeepEquals, s.disp.SbatchArguments)
345 func (s *StubbedSuite) TestSbatchArgs(c *C) {
346 container := arvados.Container{
348 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
352 for _, defaults := range [][]string{
355 {"--arg1=v1", "--arg2"},
357 c.Logf("%#v", defaults)
358 s.disp.SbatchArguments = defaults
360 args, err := s.disp.sbatchArgs(container)
361 c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--nice=10000", "--mem=239", "--cpus-per-task=2", "--tmp=0"))
366 func (s *StubbedSuite) TestSbatchInstanceTypeConstraint(c *C) {
367 container := arvados.Container{
369 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
373 for _, trial := range []struct {
374 types map[string]arvados.InstanceType
378 // Choose node type => use --constraint arg
380 types: map[string]arvados.InstanceType{
381 "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
382 "a1.small": {Name: "a1.small", Price: 0.04, RAM: 256000000, VCPUs: 2},
383 "a1.medium": {Name: "a1.medium", Price: 0.08, RAM: 512000000, VCPUs: 4},
384 "a1.large": {Name: "a1.large", Price: 0.16, RAM: 1024000000, VCPUs: 8},
386 sbatchArgs: []string{"--constraint=instancetype=a1.medium"},
388 // No node types configured => no slurm constraint
391 sbatchArgs: []string{"--mem=239", "--cpus-per-task=2", "--tmp=0"},
393 // No node type is big enough => error
395 types: map[string]arvados.InstanceType{
396 "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
398 err: dispatchcloud.ConstraintsNotSatisfiableError{},
402 s.disp.cluster = &arvados.Cluster{InstanceTypes: trial.types}
404 args, err := s.disp.sbatchArgs(container)
405 c.Check(err == nil, Equals, trial.err == nil)
406 if trial.err == nil {
407 c.Check(args, DeepEquals, append([]string{"--job-name=123", "--nice=10000"}, trial.sbatchArgs...))
409 c.Check(len(err.(dispatchcloud.ConstraintsNotSatisfiableError).AvailableTypes), Equals, len(trial.types))
414 func (s *StubbedSuite) TestSbatchPartition(c *C) {
415 container := arvados.Container{
417 RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
418 SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
422 args, err := s.disp.sbatchArgs(container)
423 c.Check(args, DeepEquals, []string{
424 "--job-name=123", "--nice=10000",
425 "--mem=239", "--cpus-per-task=1", "--tmp=0",
426 "--partition=blurb,b2",