Merge branch '9437-keep-index-ns'
[arvados.git] / services / crunch-dispatch-slurm / crunch-dispatch-slurm_test.go
index 348d5e48b87fdf19d26c75423f7f07656f42b1bd..d8b94f98d2d2d298f83423d621169acf937251b1 100644 (file)
@@ -1,14 +1,14 @@
 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"
-
-       "bytes"
-       "fmt"
+       "io"
        "log"
-       "math"
        "net/http"
        "net/http/httptest"
        "os"
@@ -35,115 +35,138 @@ 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) {
+       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) {
+       container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo") }, []string{"sbatch", "--share", "--parsable",
+               fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
+               fmt.Sprintf("--mem-per-cpu=%d", 2862),
+               fmt.Sprintf("--cpus-per-task=%d", 4),
+               fmt.Sprintf("--priority=%d", 1)},
+               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
        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
 
        doneProcessing := make(chan struct{})
        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)
-                       }()
+                       container arvados.Container,
+                       status chan arvados.Container) {
+                       go runContainer(dispatcher, container)
                        run(dispatcher, container, status)
                        doneProcessing <- struct{}{}
                },
                DoneProcessing: doneProcessing}
 
+       squeueUpdater.StartMonitor(time.Duration(500) * time.Millisecond)
+
        err = dispatcher.RunDispatcher()
        c.Assert(err, IsNil)
 
-       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)
+       squeueUpdater.Done()
 
-       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"})
+       c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
 
        // There should be no queued containers now
        err = arv.List("containers", params, &containers)
@@ -151,10 +174,10 @@ 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) {
@@ -180,22 +203,20 @@ 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
 
        doneProcessing := make(chan struct{})
        dispatcher := dispatch.Dispatcher{
                Arv:          arv,
                PollInterval: 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.Sleep(1 * time.Second)
                                dispatcher.UpdateState(container.UUID, dispatch.Running)
                                dispatcher.UpdateState(container.UUID, dispatch.Complete)
                        }()