X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/3ae9a789410e93eeb31ca5670c17a6d03d77f608..c3d4f8a585202ec58df5506934b698039c200b68:/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go index d30c5dfcea..8809e7bcc6 100644 --- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go +++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go @@ -3,12 +3,13 @@ package main import ( "bytes" "fmt" + "git.curoverse.com/arvados.git/sdk/go/arvados" "git.curoverse.com/arvados.git/sdk/go/arvadosclient" "git.curoverse.com/arvados.git/sdk/go/arvadostest" "git.curoverse.com/arvados.git/sdk/go/dispatch" "io" + "io/ioutil" "log" - "math" "net/http" "net/http/httptest" "os" @@ -58,14 +59,60 @@ func (s *MockArvadosServerSuite) TearDownTest(c *C) { } func (s *TestSuite) TestIntegrationNormal(c *C) { - s.integrationTest(c, false) + container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") }, + []string(nil), + func(dispatcher *dispatch.Dispatcher, container arvados.Container) { + dispatcher.UpdateState(container.UUID, dispatch.Running) + time.Sleep(3 * time.Second) + dispatcher.UpdateState(container.UUID, dispatch.Complete) + }) + c.Check(container.State, Equals, arvados.ContainerStateComplete) +} + +func (s *TestSuite) TestIntegrationCancel(c *C) { + + // Override sbatchCmd + var scancelCmdLine []string + defer func(orig func(arvados.Container) *exec.Cmd) { + scancelCmd = orig + }(scancelCmd) + scancelCmd = func(container arvados.Container) *exec.Cmd { + scancelCmdLine = scancelFunc(container).Args + return exec.Command("echo") + } + + container := s.integrationTest(c, + func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") }, + []string(nil), + func(dispatcher *dispatch.Dispatcher, container arvados.Container) { + dispatcher.UpdateState(container.UUID, dispatch.Running) + time.Sleep(1 * time.Second) + dispatcher.Arv.Update("containers", container.UUID, + arvadosclient.Dict{ + "container": arvadosclient.Dict{"priority": 0}}, + nil) + }) + c.Check(container.State, Equals, arvados.ContainerStateCancelled) + c.Check(scancelCmdLine, DeepEquals, []string{"scancel", "--name=zzzzz-dz642-queuedcontainer"}) } func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) { - s.integrationTest(c, true) + container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo") }, []string{"sbatch", "--share", + fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"), + fmt.Sprintf("--mem-per-cpu=%d", 2862), + fmt.Sprintf("--cpus-per-task=%d", 4)}, + func(dispatcher *dispatch.Dispatcher, container arvados.Container) { + dispatcher.UpdateState(container.UUID, dispatch.Running) + time.Sleep(3 * time.Second) + dispatcher.UpdateState(container.UUID, dispatch.Complete) + }) + c.Check(container.State, Equals, arvados.ContainerStateCancelled) } -func (s *TestSuite) integrationTest(c *C, missingFromSqueue bool) { +func (s *TestSuite) integrationTest(c *C, + newSqueueCmd func() *exec.Cmd, + sbatchCmdComps []string, + runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container { arvadostest.ResetEnv() arv, err := arvadosclient.MakeArvadosClient() @@ -74,10 +121,10 @@ func (s *TestSuite) integrationTest(c *C, missingFromSqueue bool) { var sbatchCmdLine []string // Override sbatchCmd - defer func(orig func(dispatch.Container) *exec.Cmd) { + defer func(orig func(arvados.Container) *exec.Cmd) { sbatchCmd = orig }(sbatchCmd) - sbatchCmd = func(container dispatch.Container) *exec.Cmd { + sbatchCmd = func(container arvados.Container) *exec.Cmd { sbatchCmdLine = sbatchFunc(container).Args return exec.Command("sh") } @@ -86,66 +133,39 @@ func (s *TestSuite) integrationTest(c *C, missingFromSqueue bool) { defer func(orig func() *exec.Cmd) { squeueCmd = orig }(squeueCmd) - squeueCmd = func() *exec.Cmd { - if missingFromSqueue { - return exec.Command("echo") - } else { - return exec.Command("echo", "zzzzz-dz642-queuedcontainer") - } - } + squeueCmd = newSqueueCmd - // There should be no queued containers now + // There should be one queued container params := arvadosclient.Dict{ - "filters": [][]string{[]string{"state", "=", "Queued"}}, + "filters": [][]string{{"state", "=", "Queued"}}, } - var containers dispatch.ContainerList + var containers arvados.ContainerList err = arv.List("containers", params, &containers) c.Check(err, IsNil) c.Check(len(containers.Items), Equals, 1) - echo := "echo" - crunchRunCommand = &echo + theConfig.CrunchRunCommand = []string{"echo"} - doneProcessing := make(chan struct{}) dispatcher := dispatch.Dispatcher{ - Arv: arv, - PollInterval: time.Duration(1) * time.Second, + Arv: arv, + PollPeriod: time.Duration(1) * time.Second, RunContainer: func(dispatcher *dispatch.Dispatcher, - container dispatch.Container, - status chan dispatch.Container) { - go func() { - dispatcher.UpdateState(container.UUID, dispatch.Running) - time.Sleep(3 * time.Second) - dispatcher.UpdateState(container.UUID, dispatch.Complete) - }() + container arvados.Container, + status chan arvados.Container) { + go runContainer(dispatcher, container) run(dispatcher, container, status) - doneProcessing <- struct{}{} + dispatcher.Stop() }, - DoneProcessing: doneProcessing} + } - squeueUpdater.SqueueDone = make(chan struct{}) - go squeueUpdater.SyncSqueue(time.Duration(500) * time.Millisecond) + sqCheck = SqueueChecker{Period: 500 * time.Millisecond} - err = dispatcher.RunDispatcher() + err = dispatcher.Run() c.Assert(err, IsNil) - squeueUpdater.SqueueDone <- struct{}{} - close(squeueUpdater.SqueueDone) - - item := containers.Items[0] - sbatchCmdComps := []string{"sbatch", "--share", "--parsable", - fmt.Sprintf("--job-name=%s", item.UUID), - fmt.Sprintf("--mem-per-cpu=%d", int(math.Ceil(float64(item.RuntimeConstraints["ram"])/float64(item.RuntimeConstraints["vcpus"]*1048576)))), - fmt.Sprintf("--cpus-per-task=%d", int(item.RuntimeConstraints["vcpus"])), - fmt.Sprintf("--priority=%d", item.Priority)} - - if missingFromSqueue { - // not in squeue when run() started, so it will have called sbatch - c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps) - } else { - // already in squeue when run() started, will have just monitored it instead - c.Check(sbatchCmdLine, DeepEquals, []string(nil)) - } + sqCheck.Stop() + + c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps) // There should be no queued containers now err = arv.List("containers", params, &containers) @@ -153,17 +173,13 @@ func (s *TestSuite) integrationTest(c *C, missingFromSqueue bool) { c.Check(len(containers.Items), Equals, 0) // Previously "Queued" container should now be in "Complete" state - var container dispatch.Container + var container arvados.Container err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container) c.Check(err, IsNil) - if missingFromSqueue { - c.Check(container.State, Equals, "Cancelled") - } else { - c.Check(container.State, Equals, "Complete") - } + return container } -func (s *MockArvadosServerSuite) Test_APIErrorGettingContainers(c *C) { +func (s *MockArvadosServerSuite) TestAPIErrorGettingContainers(c *C) { apiStubResponses := make(map[string]arvadostest.StubResponse) apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`} apiStubResponses["/arvados/v1/containers"] = arvadostest.StubResponse{500, string(`{}`)} @@ -177,7 +193,7 @@ func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubRespon api := httptest.NewServer(&apiStub) defer api.Close() - arv := arvadosclient.ArvadosClient{ + arv := &arvadosclient.ArvadosClient{ Scheme: "http", ApiServer: api.URL[7:], ApiToken: "abc123", @@ -189,34 +205,124 @@ func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubRespon log.SetOutput(io.MultiWriter(buf, os.Stderr)) defer log.SetOutput(os.Stderr) - crunchRunCommand = &crunchCmd + theConfig.CrunchRunCommand = []string{crunchCmd} - doneProcessing := make(chan struct{}) dispatcher := dispatch.Dispatcher{ - Arv: arv, - PollInterval: time.Duration(1) * time.Second, + Arv: arv, + PollPeriod: time.Duration(1) * time.Second, RunContainer: func(dispatcher *dispatch.Dispatcher, - container dispatch.Container, - status chan dispatch.Container) { + container arvados.Container, + status chan arvados.Container) { go func() { time.Sleep(1 * time.Second) dispatcher.UpdateState(container.UUID, dispatch.Running) dispatcher.UpdateState(container.UUID, dispatch.Complete) }() run(dispatcher, container, status) - doneProcessing <- struct{}{} + dispatcher.Stop() }, - DoneProcessing: doneProcessing} + } go func() { for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ { time.Sleep(100 * time.Millisecond) } - dispatcher.DoneProcessing <- struct{}{} + dispatcher.Stop() }() - err := dispatcher.RunDispatcher() + err := dispatcher.Run() c.Assert(err, IsNil) c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`) } + +func (s *MockArvadosServerSuite) TestNoSuchConfigFile(c *C) { + var config Config + err := readConfig(&config, "/nosuchdir89j7879/8hjwr7ojgyy7") + c.Assert(err, NotNil) +} + +func (s *MockArvadosServerSuite) TestBadSbatchArgsConfig(c *C) { + var config Config + + tmpfile, err := ioutil.TempFile(os.TempDir(), "config") + c.Check(err, IsNil) + defer os.Remove(tmpfile.Name()) + + _, err = tmpfile.Write([]byte(`{"SbatchArguments": "oops this is not a string array"}`)) + c.Check(err, IsNil) + + err = readConfig(&config, tmpfile.Name()) + c.Assert(err, NotNil) +} + +func (s *MockArvadosServerSuite) TestNoSuchArgInConfigIgnored(c *C) { + var config Config + + tmpfile, err := ioutil.TempFile(os.TempDir(), "config") + c.Check(err, IsNil) + defer os.Remove(tmpfile.Name()) + + _, err = tmpfile.Write([]byte(`{"NoSuchArg": "Nobody loves me, not one tiny hunk."}`)) + c.Check(err, IsNil) + + err = readConfig(&config, tmpfile.Name()) + c.Assert(err, IsNil) + c.Check(0, Equals, len(config.SbatchArguments)) +} + +func (s *MockArvadosServerSuite) TestReadConfig(c *C) { + var config Config + + tmpfile, err := ioutil.TempFile(os.TempDir(), "config") + c.Check(err, IsNil) + defer os.Remove(tmpfile.Name()) + + args := []string{"--arg1=v1", "--arg2", "--arg3=v3"} + argsS := `{"SbatchArguments": ["--arg1=v1", "--arg2", "--arg3=v3"]}` + _, err = tmpfile.Write([]byte(argsS)) + c.Check(err, IsNil) + + err = readConfig(&config, tmpfile.Name()) + c.Assert(err, IsNil) + c.Check(3, Equals, len(config.SbatchArguments)) + c.Check(args, DeepEquals, config.SbatchArguments) +} + +func (s *MockArvadosServerSuite) TestSbatchFuncWithNoConfigArgs(c *C) { + testSbatchFuncWithArgs(c, nil) +} + +func (s *MockArvadosServerSuite) TestSbatchFuncWithEmptyConfigArgs(c *C) { + testSbatchFuncWithArgs(c, []string{}) +} + +func (s *MockArvadosServerSuite) TestSbatchFuncWithConfigArgs(c *C) { + testSbatchFuncWithArgs(c, []string{"--arg1=v1", "--arg2"}) +} + +func testSbatchFuncWithArgs(c *C, args []string) { + theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...) + + container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2}} + sbatchCmd := sbatchFunc(container) + + var expected []string + expected = append(expected, "sbatch", "--share") + expected = append(expected, theConfig.SbatchArguments...) + expected = append(expected, "--job-name=123", "--mem-per-cpu=120", "--cpus-per-task=2") + + c.Check(sbatchCmd.Args, DeepEquals, expected) +} + +func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) { + theConfig.SbatchArguments = nil + container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1}, SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}}} + sbatchCmd := sbatchFunc(container) + + var expected []string + expected = append(expected, "sbatch", "--share") + expected = append(expected, "--job-name=123", "--mem-per-cpu=239", "--cpus-per-task=1", "--partition=blurb,b2") + + c.Check(sbatchCmd.Args, DeepEquals, expected) +}