}
}
+func niceness(priority int) int {
+ if priority > 1000 {
+ priority = 1000
+ }
+ if priority < 0 {
+ priority = 0
+ }
+ // Niceness range 1-10000
+ return (1000 - priority) * 10
+}
+
// sbatchCmd
func sbatchFunc(container arvados.Container) *exec.Cmd {
mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+container.RuntimeConstraints.KeepCacheRAM) / float64(1048576)))
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--mem=%d", mem))
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--cpus-per-task=%d", container.RuntimeConstraints.VCPUs))
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--tmp=%d", disk))
+ sbatchArgs = append(sbatchArgs, fmt.Sprintf("--nice=%d", niceness(container.Priority)))
if len(container.SchedulingParameters.Partitions) > 0 {
sbatchArgs = append(sbatchArgs, fmt.Sprintf("--partition=%s", strings.Join(container.SchedulingParameters.Partitions, ",")))
}
return exec.Command("scancel", "--name="+container.UUID)
}
+// scontrolCmd
+func scontrolFunc(container arvados.Container) *exec.Cmd {
+ return exec.Command("scontrol", "update", "JobName="+container.UUID, fmt.Sprintf("Nice=%d", niceness(container.Priority)))
+}
+
// Wrap these so that they can be overridden by tests
var sbatchCmd = sbatchFunc
var scancelCmd = scancelFunc
+var scontrolCmd = scontrolFunc
// Submit job to slurm using sbatch.
func submit(dispatcher *dispatch.Dispatcher, container arvados.Container, crunchRunCommand []string) error {
} else if updated.Priority == 0 {
log.Printf("Container %s has state %q, priority %d: cancel slurm job", ctr.UUID, updated.State, updated.Priority)
scancel(ctr)
+ } else if niceness(updated.Priority) != sqCheck.GetNiceness(ctr.UUID) && sqCheck.GetNiceness(ctr.UUID) != -1 {
+ // dynamically adjust priority
+ log.Printf("Container priority %v != %v", niceness(updated.Priority), sqCheck.GetNiceness(ctr.UUID))
+ scontrolUpdate(updated)
}
}
}
}
}
+func scontrolUpdate(ctr arvados.Container) {
+ sqCheck.L.Lock()
+ cmd := scontrolCmd(ctr)
+ msg, err := cmd.CombinedOutput()
+ sqCheck.L.Unlock()
+
+ if err != nil {
+ log.Printf("%q %q: %s %q", cmd.Path, cmd.Args, err, msg)
+ time.Sleep(time.Second)
+ } else if sqCheck.HasUUID(ctr.UUID) {
+ log.Printf("Container %s priority is now %v, niceness is now %v",
+ ctr.UUID, ctr.Priority, sqCheck.GetNiceness(ctr.UUID))
+ }
+}
+
func readConfig(dst interface{}, path string) error {
err := config.LoadFile(dst, path)
if err != nil && os.IsNotExist(err) && path == defaultConfigPath {
arvadostest.ResetEnv()
}
+func (s *TestSuite) integrationTest(c *C,
+ newSqueueCmd func() *exec.Cmd,
+ newScancelCmd func(arvados.Container) *exec.Cmd,
+ newSbatchCmd func(arvados.Container) *exec.Cmd,
+ newScontrolCmd func(arvados.Container) *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
+
+ // Override sbatchCmd
+ defer func(orig func(arvados.Container) *exec.Cmd) {
+ sbatchCmd = orig
+ }(sbatchCmd)
+
+ if newSbatchCmd != nil {
+ sbatchCmd = newSbatchCmd
+ } else {
+ sbatchCmd = func(container arvados.Container) *exec.Cmd {
+ sbatchCmdLine = sbatchFunc(container).Args
+ return exec.Command("sh")
+ }
+ }
+
+ // Override squeueCmd
+ defer func(orig func() *exec.Cmd) {
+ squeueCmd = orig
+ }(squeueCmd)
+ squeueCmd = newSqueueCmd
+
+ // Override scancel
+ defer func(orig func(arvados.Container) *exec.Cmd) {
+ scancelCmd = orig
+ }(scancelCmd)
+ scancelCmd = newScancelCmd
+
+ // Override scontrol
+ defer func(orig func(arvados.Container) *exec.Cmd) {
+ scontrolCmd = orig
+ }(scontrolCmd)
+ scontrolCmd = newScontrolCmd
+
+ // There should be one queued container
+ params := arvadosclient.Dict{
+ "filters": [][]string{{"state", "=", "Queued"}},
+ }
+ var containers arvados.ContainerList
+ err = arv.List("containers", params, &containers)
+ c.Check(err, IsNil)
+ c.Check(len(containers.Items), Equals, 1)
+
+ theConfig.CrunchRunCommand = []string{"echo"}
+
+ ctx, cancel := context.WithCancel(context.Background())
+ doneRun := make(chan struct{})
+
+ dispatcher := dispatch.Dispatcher{
+ Arv: arv,
+ PollPeriod: time.Duration(1) * time.Second,
+ RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+ go func() {
+ runContainer(disp, ctr)
+ doneRun <- struct{}{}
+ }()
+ run(disp, ctr, status)
+ cancel()
+ },
+ }
+
+ sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
+
+ err = dispatcher.Run(ctx)
+ <-doneRun
+ c.Assert(err, Equals, context.Canceled)
+
+ sqCheck.Stop()
+
+ c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
+
+ // There should be no queued containers now
+ err = arv.List("containers", params, &containers)
+ c.Check(err, IsNil)
+ c.Check(len(containers.Items), Equals, 0)
+
+ // Previously "Queued" container should now be in "Complete" state
+ var container arvados.Container
+ err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
+ c.Check(err, IsNil)
+ return container
+}
+
func (s *TestSuite) TestIntegrationNormal(c *C) {
done := false
container := s.integrationTest(c,
if done {
return exec.Command("true")
} else {
- return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 9990 100")
}
},
nil,
nil,
+ nil,
[]string(nil),
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
if cmd != nil && cmd.ProcessState != nil {
return exec.Command("true")
} else {
- return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 9990 100")
}
},
func(container arvados.Container) *exec.Cmd {
}
},
nil,
+ nil,
[]string(nil),
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
func() *exec.Cmd { return exec.Command("echo") },
nil,
nil,
+ nil,
[]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)},
+ fmt.Sprintf("--tmp=%d", 45777),
+ fmt.Sprintf("--nice=%d", 9990)},
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
time.Sleep(3 * time.Second)
func(container arvados.Container) *exec.Cmd {
return exec.Command("false")
},
+ nil,
[]string(nil),
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
c.Assert(len(ll.Items), Equals, 1)
}
-func (s *TestSuite) integrationTest(c *C,
- newSqueueCmd func() *exec.Cmd,
- newScancelCmd func(arvados.Container) *exec.Cmd,
- newSbatchCmd func(arvados.Container) *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
-
- // Override sbatchCmd
- defer func(orig func(arvados.Container) *exec.Cmd) {
- sbatchCmd = orig
- }(sbatchCmd)
-
- if newSbatchCmd != nil {
- sbatchCmd = newSbatchCmd
- } else {
- sbatchCmd = func(container arvados.Container) *exec.Cmd {
- sbatchCmdLine = sbatchFunc(container).Args
- return exec.Command("sh")
- }
- }
-
- // Override squeueCmd
- defer func(orig func() *exec.Cmd) {
- squeueCmd = orig
- }(squeueCmd)
- squeueCmd = newSqueueCmd
-
- // Override scancel
- defer func(orig func(arvados.Container) *exec.Cmd) {
- scancelCmd = orig
- }(scancelCmd)
- scancelCmd = newScancelCmd
-
- // There should be one queued container
- params := arvadosclient.Dict{
- "filters": [][]string{{"state", "=", "Queued"}},
- }
- var containers arvados.ContainerList
- err = arv.List("containers", params, &containers)
- c.Check(err, IsNil)
- c.Check(len(containers.Items), Equals, 1)
-
- theConfig.CrunchRunCommand = []string{"echo"}
-
- ctx, cancel := context.WithCancel(context.Background())
- doneRun := make(chan struct{})
-
- dispatcher := dispatch.Dispatcher{
- Arv: arv,
- PollPeriod: time.Duration(1) * time.Second,
- RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
- go func() {
- runContainer(disp, ctr)
- doneRun <- struct{}{}
- }()
- run(disp, ctr, status)
- cancel()
- },
- }
-
- sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
-
- err = dispatcher.Run(ctx)
- <-doneRun
- c.Assert(err, Equals, context.Canceled)
-
- sqCheck.Stop()
-
- c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
-
- // There should be no queued containers now
- err = arv.List("containers", params, &containers)
- c.Check(err, IsNil)
- c.Check(len(containers.Items), Equals, 0)
-
- // Previously "Queued" container should now be in "Complete" state
- var container arvados.Container
- err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
- c.Check(err, IsNil)
- return container
-}
-
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 + `"}`}
func testSbatchFuncWithArgs(c *C, args []string) {
theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...)
- container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2}}
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
+ Priority: 1}
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")
+ expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990")
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"}}}
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
+ SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
+ Priority: 1}
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")
+ expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--nice=9990", "--partition=blurb,b2")
c.Check(sbatchCmd.Args, DeepEquals, expected)
}
+
+func (s *TestSuite) TestIntegrationChangePriority(c *C) {
+ var scontrolCmdLine []string
+ step := 0
+
+ container := s.integrationTest(c,
+ func() *exec.Cmd {
+ if step == 0 {
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 9990 100")
+ } else if step == 1 {
+ return exec.Command("echo", "zzzzz-dz642-queuedcontainer 4000 100")
+ } else {
+ return exec.Command("echo")
+ }
+ },
+ func(arvados.Container) *exec.Cmd { return exec.Command("true") },
+ nil,
+ func(container arvados.Container) *exec.Cmd {
+ scontrolCmdLine = scontrolFunc(container).Args
+ step = 1
+ return exec.Command("true")
+ },
+ []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": 600}},
+ nil)
+ time.Sleep(1 * time.Second)
+ step = 2
+ dispatcher.UpdateState(container.UUID, dispatch.Complete)
+ })
+ c.Check(container.State, Equals, arvados.ContainerStateComplete)
+ c.Check(scontrolCmdLine, DeepEquals, []string{"scontrol", "update", "JobName=zzzzz-dz642-queuedcontainer", "Nice=4000"})
+}