Merge branch 'master' into 10645-cr-mounts-display
[arvados.git] / services / crunch-dispatch-slurm / crunch-dispatch-slurm_test.go
index 348d5e48b87fdf19d26c75423f7f07656f42b1bd..1c366a0e5a59ecaa17233354e14cd882d1470deb 100644 (file)
@@ -1,14 +1,12 @@
 package main
 
 import (
-       "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
-       "git.curoverse.com/arvados.git/sdk/go/arvadostest"
-       "git.curoverse.com/arvados.git/sdk/go/dispatch"
-
        "bytes"
+       "context"
        "fmt"
+       "io"
+       "io/ioutil"
        "log"
-       "math"
        "net/http"
        "net/http/httptest"
        "os"
@@ -17,6 +15,10 @@ import (
        "testing"
        "time"
 
+       "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"
        . "gopkg.in/check.v1"
 )
 
@@ -35,115 +37,156 @@ var initialArgs []string
 
 func (s *TestSuite) SetUpSuite(c *C) {
        initialArgs = os.Args
-       arvadostest.StartAPI()
 }
 
 func (s *TestSuite) TearDownSuite(c *C) {
-       arvadostest.StopAPI()
 }
 
 func (s *TestSuite) SetUpTest(c *C) {
        args := []string{"crunch-dispatch-slurm"}
        os.Args = args
 
+       arvadostest.StartAPI()
        os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
 }
 
 func (s *TestSuite) TearDownTest(c *C) {
-       arvadostest.ResetEnv()
        os.Args = initialArgs
+       arvadostest.StopAPI()
 }
 
 func (s *MockArvadosServerSuite) TearDownTest(c *C) {
        arvadostest.ResetEnv()
 }
 
-func (s *TestSuite) TestIntegration(c *C) {
+func (s *TestSuite) TestIntegrationNormal(c *C) {
+       done := false
+       container := s.integrationTest(c,
+               func() *exec.Cmd {
+                       if done {
+                               return exec.Command("true")
+                       } else {
+                               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)
+                       done = true
+               })
+       c.Check(container.State, Equals, arvados.ContainerStateComplete)
+}
+
+func (s *TestSuite) TestIntegrationCancel(c *C) {
+       var cmd *exec.Cmd
+       var scancelCmdLine []string
+       defer func(orig func(arvados.Container) *exec.Cmd) {
+               scancelCmd = orig
+       }(scancelCmd)
+       attempt := 0
+       scancelCmd = func(container arvados.Container) *exec.Cmd {
+               if attempt++; attempt == 1 {
+                       return exec.Command("false")
+               } else {
+                       scancelCmdLine = scancelFunc(container).Args
+                       cmd = exec.Command("echo")
+                       return cmd
+               }
+       }
+
+       container := s.integrationTest(c,
+               func() *exec.Cmd {
+                       if cmd != nil && cmd.ProcessState != nil {
+                               return exec.Command("true")
+                       } else {
+                               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) {
+       container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo") }, []string{"sbatch",
+               fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
+               fmt.Sprintf("--mem=%d", 11445),
+               fmt.Sprintf("--cpus-per-task=%d", 4),
+               fmt.Sprintf("--tmp=%d", 45777)},
+               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,
+       newSqueueCmd func() *exec.Cmd,
+       sbatchCmdComps []string,
+       runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
+       arvadostest.ResetEnv()
+
        arv, err := arvadosclient.MakeArvadosClient()
        c.Assert(err, IsNil)
 
        var sbatchCmdLine []string
-       var striggerCmdLine []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")
        }
 
-       // Override striggerCmd
-       defer func(orig func(jobid, containerUUID, finishCommand,
-               apiHost, apiToken, apiInsecure string) *exec.Cmd) {
-               striggerCmd = orig
-       }(striggerCmd)
-       striggerCmd = func(jobid, containerUUID, finishCommand, apiHost, apiToken, apiInsecure string) *exec.Cmd {
-               striggerCmdLine = striggerFunc(jobid, containerUUID, finishCommand,
-                       apiHost, apiToken, apiInsecure).Args
-               go func() {
-                       time.Sleep(5 * time.Second)
-                       arv.Update("containers", containerUUID,
-                               arvadosclient.Dict{
-                                       "container": arvadosclient.Dict{"state": dispatch.Complete}},
-                               nil)
-               }()
-               return exec.Command("echo", striggerCmdLine...)
-       }
-
        // Override squeueCmd
        defer func(orig func() *exec.Cmd) {
                squeueCmd = orig
        }(squeueCmd)
-       squeueCmd = func() *exec.Cmd {
-               return exec.Command("echo")
-       }
+       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
-       finishCmd := "/usr/bin/crunch-finish-slurm.sh"
-       finishCommand = &finishCmd
+       theConfig.CrunchRunCommand = []string{"echo"}
 
-       doneProcessing := make(chan struct{})
+       ctx, cancel := context.WithCancel(context.Background())
        dispatcher := dispatch.Dispatcher{
-               Arv:          arv,
-               PollInterval: time.Duration(1) * time.Second,
-               RunContainer: func(dispatcher *dispatch.Dispatcher,
-                       container dispatch.Container,
-                       status chan dispatch.Container) {
-                       go func() {
-                               time.Sleep(1)
-                               dispatcher.UpdateState(container.UUID, dispatch.Running)
-                               dispatcher.UpdateState(container.UUID, dispatch.Complete)
-                       }()
-                       run(dispatcher, container, status)
-                       doneProcessing <- struct{}{}
+               Arv:        arv,
+               PollPeriod: time.Duration(1) * time.Second,
+               RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+                       go runContainer(disp, ctr)
+                       run(disp, ctr, status)
+                       cancel()
                },
-               DoneProcessing: doneProcessing}
+       }
 
-       err = dispatcher.RunDispatcher()
-       c.Assert(err, IsNil)
+       sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
 
-       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)}
-       c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
+       err = dispatcher.Run(ctx)
+       c.Assert(err, Equals, context.Canceled)
 
-       c.Check(striggerCmdLine, DeepEquals, []string{"strigger", "--set", "--jobid=zzzzz-dz642-queuedcontainer", "--fini",
-               "--program=/usr/bin/crunch-finish-slurm.sh " + os.Getenv("ARVADOS_API_HOST") + " " + arvadostest.Dispatch1Token + " 1 zzzzz-dz642-queuedcontainer"})
+       sqCheck.Stop()
+
+       c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
 
        // There should be no queued containers now
        err = arv.List("containers", params, &containers)
@@ -151,13 +194,13 @@ func (s *TestSuite) TestIntegration(c *C) {
        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)
-       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(`{}`)}
@@ -171,7 +214,7 @@ 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",
@@ -180,39 +223,126 @@ func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubRespon
        }
 
        buf := bytes.NewBuffer(nil)
-       log.SetOutput(buf)
+       log.SetOutput(io.MultiWriter(buf, os.Stderr))
        defer log.SetOutput(os.Stderr)
 
-       crunchRunCommand = &crunchCmd
-       finishCmd := "/usr/bin/crunch-finish-slurm.sh"
-       finishCommand = &finishCmd
+       theConfig.CrunchRunCommand = []string{crunchCmd}
 
-       doneProcessing := make(chan struct{})
+       ctx, cancel := context.WithCancel(context.Background())
        dispatcher := dispatch.Dispatcher{
-               Arv:          arv,
-               PollInterval: time.Duration(1) * time.Second,
-               RunContainer: func(dispatcher *dispatch.Dispatcher,
-                       container dispatch.Container,
-                       status chan dispatch.Container) {
+               Arv:        arv,
+               PollPeriod: time.Duration(1) * time.Second,
+               RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
                        go func() {
-                               time.Sleep(1)
-                               dispatcher.UpdateState(container.UUID, dispatch.Running)
-                               dispatcher.UpdateState(container.UUID, dispatch.Complete)
+                               time.Sleep(1 * time.Second)
+                               disp.UpdateState(ctr.UUID, dispatch.Running)
+                               disp.UpdateState(ctr.UUID, dispatch.Complete)
                        }()
-                       run(dispatcher, container, status)
-                       doneProcessing <- struct{}{}
+                       run(disp, ctr, status)
+                       cancel()
                },
-               DoneProcessing: doneProcessing}
+       }
 
        go func() {
                for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ {
                        time.Sleep(100 * time.Millisecond)
                }
-               dispatcher.DoneProcessing <- struct{}{}
+               cancel()
        }()
 
-       err := dispatcher.RunDispatcher()
-       c.Assert(err, IsNil)
+       err := dispatcher.Run(ctx)
+       c.Assert(err, Equals, context.Canceled)
 
        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")
+       expected = append(expected, theConfig.SbatchArguments...)
+       expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0")
+
+       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")
+       expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--partition=blurb,b2")
+
+       c.Check(sbatchCmd.Args, DeepEquals, expected)
+}