+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
package main
import (
"bytes"
+ "context"
+ "errors"
"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"
"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"
)
func (s *TestSuite) TearDownTest(c *C) {
os.Args = initialArgs
+ arvadostest.ResetEnv()
arvadostest.StopAPI()
}
arvadostest.ResetEnv()
}
-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)
+type slurmFake struct {
+ didBatch [][]string
+ didCancel []string
+ didRenice [][]string
+ queue string
+ // If non-nil, run this func during the 2nd+ call to Cancel()
+ onCancel func()
+ // Error returned by Batch()
+ errBatch error
}
-func (s *TestSuite) TestIntegrationCancel(c *C) {
+func (sf *slurmFake) Batch(script io.Reader, args []string) error {
+ sf.didBatch = append(sf.didBatch, args)
+ return sf.errBatch
+}
- // 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")
- }
+func (sf *slurmFake) QueueCommand(args []string) *exec.Cmd {
+ return exec.Command("echo", sf.queue)
+}
- 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 (sf *slurmFake) Renice(name string, nice int) error {
+ sf.didRenice = append(sf.didRenice, []string{name, fmt.Sprintf("%d", nice)})
+ return nil
}
-func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
- 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 (sf *slurmFake) Cancel(name string) error {
+ sf.didCancel = append(sf.didCancel, name)
+ if len(sf.didCancel) == 1 {
+ // simulate error on first attempt
+ return errors.New("something terrible happened")
+ }
+ if sf.onCancel != nil {
+ sf.onCancel()
+ }
+ return nil
}
-func (s *TestSuite) integrationTest(c *C,
- newSqueueCmd func() *exec.Cmd,
- sbatchCmdComps []string,
+func (s *TestSuite) integrationTest(c *C, slurm *slurmFake,
+ expectBatch [][]string,
runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
arvadostest.ResetEnv()
arv, err := arvadosclient.MakeArvadosClient()
c.Assert(err, IsNil)
- var sbatchCmdLine []string
+ defer func(orig Slurm) {
+ theConfig.slurm = orig
+ }(theConfig.slurm)
+ theConfig.slurm = slurm
- // Override sbatchCmd
- defer func(orig func(arvados.Container) *exec.Cmd) {
- sbatchCmd = orig
- }(sbatchCmd)
- 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
-
- // There should be no queued containers now
+ // There should be one queued container
params := arvadosclient.Dict{
"filters": [][]string{{"state", "=", "Queued"}},
}
theConfig.CrunchRunCommand = []string{"echo"}
- doneProcessing := make(chan struct{})
+ ctx, cancel := context.WithCancel(context.Background())
+ doneRun := make(chan struct{})
+
dispatcher := dispatch.Dispatcher{
- Arv: arv,
- PollInterval: time.Duration(1) * time.Second,
- RunContainer: func(dispatcher *dispatch.Dispatcher,
- container arvados.Container,
- status chan arvados.Container) {
- go runContainer(dispatcher, container)
- 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 func() {
+ runContainer(disp, ctr)
+ slurm.queue = ""
+ doneRun <- struct{}{}
+ }()
+ run(disp, ctr, status)
+ cancel()
},
- DoneProcessing: doneProcessing}
+ }
- squeueUpdater.StartMonitor(time.Duration(500) * time.Millisecond)
+ sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
- err = dispatcher.RunDispatcher()
- c.Assert(err, IsNil)
+ err = dispatcher.Run(ctx)
+ <-doneRun
+ c.Assert(err, Equals, context.Canceled)
- squeueUpdater.Done()
+ sqCheck.Stop()
- c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
+ c.Check(slurm.didBatch, DeepEquals, expectBatch)
// There should be no queued containers now
err = arv.List("containers", params, &containers)
return container
}
+func (s *TestSuite) TestIntegrationNormal(c *C) {
+ container := s.integrationTest(c,
+ &slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"},
+ 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) {
+ slurm := &slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"}
+ readyToCancel := make(chan bool)
+ slurm.onCancel = func() { <-readyToCancel }
+ container := s.integrationTest(c,
+ slurm,
+ nil,
+ func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+ dispatcher.UpdateState(container.UUID, dispatch.Running)
+ time.Sleep(time.Second)
+ dispatcher.Arv.Update("containers", container.UUID,
+ arvadosclient.Dict{
+ "container": arvadosclient.Dict{"priority": 0}},
+ nil)
+ readyToCancel <- true
+ close(readyToCancel)
+ })
+ c.Check(container.State, Equals, arvados.ContainerStateCancelled)
+ c.Check(len(slurm.didCancel) > 1, Equals, true)
+ c.Check(slurm.didCancel[:2], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "zzzzz-dz642-queuedcontainer"})
+}
+
+func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
+ container := s.integrationTest(c, &slurmFake{},
+ [][]string{{
+ 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("--nice=%d", 9990)}},
+ 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) TestSbatchFail(c *C) {
+ container := s.integrationTest(c,
+ &slurmFake{errBatch: errors.New("something terrible happened")},
+ [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--mem=11445", "--cpus-per-task=4", "--tmp=45777", "--nice=9990"}},
+ func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+ dispatcher.UpdateState(container.UUID, dispatch.Running)
+ dispatcher.UpdateState(container.UUID, dispatch.Complete)
+ })
+ c.Check(container.State, Equals, arvados.ContainerStateComplete)
+
+ arv, err := arvadosclient.MakeArvadosClient()
+ c.Assert(err, IsNil)
+
+ var ll arvados.LogList
+ err = arv.List("logs", arvadosclient.Dict{"filters": [][]string{
+ {"object_uuid", "=", container.UUID},
+ {"event_type", "=", "dispatch"},
+ }}, &ll)
+ c.Assert(len(ll.Items), Equals, 1)
+}
+
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 + `"}`}
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 arvados.Container,
- status chan arvados.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 * time.Second)
- dispatcher.UpdateState(container.UUID, dispatch.Running)
- dispatcher.UpdateState(container.UUID, dispatch.Complete)
+ 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 testSbatchFuncWithArgs(c *C, args []string) {
+ defer func() { theConfig.SbatchArguments = nil }()
theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...)
- container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 1000000, VCPUs: 2}}
- sbatchCmd := sbatchFunc(container)
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
+ Priority: 1}
var expected []string
- expected = append(expected, "sbatch", "--share")
expected = append(expected, theConfig.SbatchArguments...)
- expected = append(expected, "--job-name=123", "--mem-per-cpu=1", "--cpus-per-task=2")
+ expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990")
+ c.Check(sbatchArgs(container), DeepEquals, expected)
+}
+
+func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) {
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
+ SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
+ Priority: 1}
+
+ c.Check(sbatchArgs(container), DeepEquals, []string{
+ "--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) {
+ slurm := &slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"}
+ container := s.integrationTest(c, slurm, nil,
+ func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+ dispatcher.UpdateState(container.UUID, dispatch.Running)
+ time.Sleep(time.Second)
+ dispatcher.Arv.Update("containers", container.UUID,
+ arvadosclient.Dict{
+ "container": arvadosclient.Dict{"priority": 600}},
+ nil)
+ time.Sleep(time.Second)
+ dispatcher.UpdateState(container.UUID, dispatch.Complete)
+ })
+ c.Check(container.State, Equals, arvados.ContainerStateComplete)
+ c.Assert(len(slurm.didRenice), Not(Equals), 0)
+ c.Check(slurm.didRenice[len(slurm.didRenice)-1], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "4000"})
}