import (
"bytes"
"context"
+ "errors"
"fmt"
"io"
"io/ioutil"
- "log"
"net/http"
"net/http/httptest"
"os"
"testing"
"time"
+ "git.curoverse.com/arvados.git/lib/dispatchcloud"
"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"
+ "github.com/Sirupsen/logrus"
. "gopkg.in/check.v1"
)
TestingT(t)
}
-var _ = Suite(&TestSuite{})
-var _ = Suite(&MockArvadosServerSuite{})
+var _ = Suite(&IntegrationSuite{})
+var _ = Suite(&StubbedSuite{})
-type TestSuite struct{}
-type MockArvadosServerSuite struct{}
-
-var initialArgs []string
-
-func (s *TestSuite) SetUpSuite(c *C) {
- initialArgs = os.Args
-}
-
-func (s *TestSuite) TearDownSuite(c *C) {
+type IntegrationSuite struct {
+ disp Dispatcher
+ slurm slurmFake
}
-func (s *TestSuite) SetUpTest(c *C) {
- args := []string{"crunch-dispatch-slurm"}
- os.Args = args
-
+func (s *IntegrationSuite) SetUpTest(c *C) {
arvadostest.StartAPI()
os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
+ s.disp = Dispatcher{}
+ s.disp.setup()
+ s.slurm = slurmFake{}
}
-func (s *TestSuite) TearDownTest(c *C) {
- os.Args = initialArgs
+func (s *IntegrationSuite) TearDownTest(c *C) {
arvadostest.ResetEnv()
arvadostest.StopAPI()
}
-func (s *MockArvadosServerSuite) TearDownTest(c *C) {
- arvadostest.ResetEnv()
+type slurmFake struct {
+ didBatch [][]string
+ didCancel []string
+ didRelease []string
+ didRenice [][]string
+ queue string
+ rejectNice10K bool
+ // If non-nil, run this func during the 2nd+ call to Cancel()
+ onCancel func()
+ // Error returned by Batch()
+ errBatch error
}
-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)
+func (sf *slurmFake) Batch(script io.Reader, args []string) error {
+ sf.didBatch = append(sf.didBatch, args)
+ return sf.errBatch
+}
- var sbatchCmdLine []string
+func (sf *slurmFake) QueueCommand(args []string) *exec.Cmd {
+ return exec.Command("echo", sf.queue)
+}
- // Override sbatchCmd
- defer func(orig func(arvados.Container) *exec.Cmd) {
- sbatchCmd = orig
- }(sbatchCmd)
+func (sf *slurmFake) Release(name string) error {
+ sf.didRelease = append(sf.didRelease, name)
+ return nil
+}
- if newSbatchCmd != nil {
- sbatchCmd = newSbatchCmd
- } else {
- sbatchCmd = func(container arvados.Container) *exec.Cmd {
- sbatchCmdLine = sbatchFunc(container).Args
- return exec.Command("sh")
- }
+func (sf *slurmFake) Renice(name string, nice int64) error {
+ sf.didRenice = append(sf.didRenice, []string{name, fmt.Sprintf("%d", nice)})
+ if sf.rejectNice10K && nice > 10000 {
+ return errors.New("scontrol: error: Invalid nice value, must be between -10000 and 10000")
}
+ return nil
+}
- // Override squeueCmd
- defer func(orig func() *exec.Cmd) {
- squeueCmd = orig
- }(squeueCmd)
- squeueCmd = newSqueueCmd
+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
+}
- // Override scancel
- defer func(orig func(arvados.Container) *exec.Cmd) {
- scancelCmd = orig
- }(scancelCmd)
- scancelCmd = newScancelCmd
+func (s *IntegrationSuite) integrationTest(c *C,
+ expectBatch [][]string,
+ runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
+ arvadostest.ResetEnv()
- // Override scontrol
- defer func(orig func(arvados.Container) *exec.Cmd) {
- scontrolCmd = orig
- }(scontrolCmd)
- scontrolCmd = newScontrolCmd
+ arv, err := arvadosclient.MakeArvadosClient()
+ c.Assert(err, IsNil)
// There should be one queued container
params := arvadosclient.Dict{
var containers arvados.ContainerList
err = arv.List("containers", params, &containers)
c.Check(err, IsNil)
- c.Check(len(containers.Items), Equals, 1)
+ c.Assert(len(containers.Items), Equals, 1)
- theConfig.CrunchRunCommand = []string{"echo"}
+ s.disp.CrunchRunCommand = []string{"echo"}
ctx, cancel := context.WithCancel(context.Background())
doneRun := make(chan struct{})
- dispatcher := dispatch.Dispatcher{
+ s.disp.Dispatcher = &dispatch.Dispatcher{
Arv: arv,
- PollPeriod: time.Duration(1) * time.Second,
+ PollPeriod: time.Second,
RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
go func() {
runContainer(disp, ctr)
+ s.slurm.queue = ""
doneRun <- struct{}{}
}()
- run(disp, ctr, status)
+ s.disp.runContainer(disp, ctr, status)
cancel()
},
}
- sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
+ s.disp.slurm = &s.slurm
+ s.disp.sqCheck = &SqueueChecker{
+ Logger: logrus.StandardLogger(),
+ Period: 500 * time.Millisecond,
+ Slurm: s.disp.slurm,
+ }
- err = dispatcher.Run(ctx)
+ err = s.disp.Dispatcher.Run(ctx)
<-doneRun
c.Assert(err, Equals, context.Canceled)
- sqCheck.Stop()
+ s.disp.sqCheck.Stop()
- c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
+ c.Check(s.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) {
- done := false
+func (s *IntegrationSuite) TestNormal(c *C) {
+ s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
container := s.integrationTest(c,
- func() *exec.Cmd {
- if done {
- return exec.Command("true")
- } else {
- 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)
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
- attempt := 0
-
+func (s *IntegrationSuite) TestCancel(c *C) {
+ s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
+ readyToCancel := make(chan bool)
+ s.slurm.onCancel = func() { <-readyToCancel }
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 9990 100")
- }
- },
- 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
- }
- },
nil,
- nil,
- []string(nil),
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
- time.Sleep(1 * time.Second)
+ 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(scancelCmdLine, DeepEquals, []string{"scancel", "--name=zzzzz-dz642-queuedcontainer"})
+ c.Check(len(s.slurm.didCancel) > 1, Equals, true)
+ c.Check(s.slurm.didCancel[:2], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "zzzzz-dz642-queuedcontainer"})
}
-func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
+func (s *IntegrationSuite) TestMissingFromSqueue(c *C) {
container := s.integrationTest(c,
- func() *exec.Cmd { return exec.Command("echo") },
- nil,
- nil,
- nil,
- []string{"sbatch",
+ [][]string{{
fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
+ fmt.Sprintf("--nice=%d", 10000),
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)
c.Check(container.State, Equals, arvados.ContainerStateCancelled)
}
-func (s *TestSuite) TestSbatchFail(c *C) {
+func (s *IntegrationSuite) TestSbatchFail(c *C) {
+ s.slurm = slurmFake{errBatch: errors.New("something terrible happened")}
container := s.integrationTest(c,
- func() *exec.Cmd { return exec.Command("echo") },
- nil,
- func(container arvados.Container) *exec.Cmd {
- return exec.Command("false")
- },
- nil,
- []string(nil),
+ [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--nice=10000", "--mem=11445", "--cpus-per-task=4", "--tmp=45777"}},
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
dispatcher.UpdateState(container.UUID, dispatch.Complete)
{"object_uuid", "=", container.UUID},
{"event_type", "=", "dispatch"},
}}, &ll)
+ c.Assert(err, IsNil)
c.Assert(len(ll.Items), Equals, 1)
}
-func (s *MockArvadosServerSuite) TestAPIErrorGettingContainers(c *C) {
+type StubbedSuite struct {
+ disp Dispatcher
+}
+
+func (s *StubbedSuite) SetUpTest(c *C) {
+ s.disp = Dispatcher{}
+ s.disp.setup()
+}
+
+func (s *StubbedSuite) 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(`{}`)}
- testWithServerStub(c, apiStubResponses, "echo", "Error getting list of containers")
+ s.testWithServerStub(c, apiStubResponses, "echo", "error getting count of containers")
}
-func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
+func (s *StubbedSuite) testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
apiStub := arvadostest.ServerStub{apiStubResponses}
api := httptest.NewServer(&apiStub)
}
buf := bytes.NewBuffer(nil)
- log.SetOutput(io.MultiWriter(buf, os.Stderr))
- defer log.SetOutput(os.Stderr)
+ logrus.SetOutput(io.MultiWriter(buf, os.Stderr))
+ defer logrus.SetOutput(os.Stderr)
- theConfig.CrunchRunCommand = []string{crunchCmd}
+ s.disp.CrunchRunCommand = []string{crunchCmd}
ctx, cancel := context.WithCancel(context.Background())
dispatcher := dispatch.Dispatcher{
Arv: arv,
- PollPeriod: time.Duration(1) * time.Second,
+ PollPeriod: time.Second,
RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
go func() {
- time.Sleep(1 * time.Second)
+ time.Sleep(time.Second)
disp.UpdateState(ctr.UUID, dispatch.Running)
disp.UpdateState(ctr.UUID, dispatch.Complete)
}()
- run(disp, ctr, status)
+ s.disp.runContainer(disp, ctr, status)
cancel()
},
}
c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
}
-func (s *MockArvadosServerSuite) TestNoSuchConfigFile(c *C) {
- var config Config
- err := readConfig(&config, "/nosuchdir89j7879/8hjwr7ojgyy7")
+func (s *StubbedSuite) TestNoSuchConfigFile(c *C) {
+ err := s.disp.readConfig("/nosuchdir89j7879/8hjwr7ojgyy7")
c.Assert(err, NotNil)
}
-func (s *MockArvadosServerSuite) TestBadSbatchArgsConfig(c *C) {
- var config Config
-
+func (s *StubbedSuite) TestBadSbatchArgsConfig(c *C) {
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())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, NotNil)
}
-func (s *MockArvadosServerSuite) TestNoSuchArgInConfigIgnored(c *C) {
- var config Config
-
+func (s *StubbedSuite) TestNoSuchArgInConfigIgnored(c *C) {
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())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, IsNil)
- c.Check(0, Equals, len(config.SbatchArguments))
+ c.Check(0, Equals, len(s.disp.SbatchArguments))
}
-func (s *MockArvadosServerSuite) TestReadConfig(c *C) {
- var config Config
-
+func (s *StubbedSuite) TestReadConfig(c *C) {
tmpfile, err := ioutil.TempFile(os.TempDir(), "config")
c.Check(err, IsNil)
defer os.Remove(tmpfile.Name())
_, err = tmpfile.Write([]byte(argsS))
c.Check(err, IsNil)
- err = readConfig(&config, tmpfile.Name())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, IsNil)
- c.Check(3, Equals, len(config.SbatchArguments))
- c.Check(args, DeepEquals, config.SbatchArguments)
+ c.Check(args, DeepEquals, s.disp.SbatchArguments)
}
-func (s *MockArvadosServerSuite) TestSbatchFuncWithNoConfigArgs(c *C) {
- testSbatchFuncWithArgs(c, nil)
-}
-
-func (s *MockArvadosServerSuite) TestSbatchFuncWithEmptyConfigArgs(c *C) {
- testSbatchFuncWithArgs(c, []string{})
-}
+func (s *StubbedSuite) TestSbatchArgs(c *C) {
+ container := arvados.Container{
+ UUID: "123",
+ RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
+ Priority: 1,
+ }
-func (s *MockArvadosServerSuite) TestSbatchFuncWithConfigArgs(c *C) {
- testSbatchFuncWithArgs(c, []string{"--arg1=v1", "--arg2"})
+ for _, defaults := range [][]string{
+ nil,
+ {},
+ {"--arg1=v1", "--arg2"},
+ } {
+ c.Logf("%#v", defaults)
+ s.disp.SbatchArguments = defaults
+
+ args, err := s.disp.sbatchArgs(container)
+ c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--nice=10000", "--mem=239", "--cpus-per-task=2", "--tmp=0"))
+ c.Check(err, IsNil)
+ }
}
-func testSbatchFuncWithArgs(c *C, args []string) {
- theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...)
-
+func (s *StubbedSuite) TestSbatchInstanceTypeConstraint(c *C) {
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", "--nice=9990")
+ Priority: 1,
+ }
- c.Check(sbatchCmd.Args, DeepEquals, expected)
+ for _, trial := range []struct {
+ types map[string]arvados.InstanceType
+ sbatchArgs []string
+ err error
+ }{
+ // Choose node type => use --constraint arg
+ {
+ types: map[string]arvados.InstanceType{
+ "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
+ "a1.small": {Name: "a1.small", Price: 0.04, RAM: 256000000, VCPUs: 2},
+ "a1.medium": {Name: "a1.medium", Price: 0.08, RAM: 512000000, VCPUs: 4},
+ "a1.large": {Name: "a1.large", Price: 0.16, RAM: 1024000000, VCPUs: 8},
+ },
+ sbatchArgs: []string{"--constraint=instancetype=a1.medium"},
+ },
+ // No node types configured => no slurm constraint
+ {
+ types: nil,
+ sbatchArgs: []string{"--mem=239", "--cpus-per-task=2", "--tmp=0"},
+ },
+ // No node type is big enough => error
+ {
+ types: map[string]arvados.InstanceType{
+ "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
+ },
+ err: dispatchcloud.ConstraintsNotSatisfiableError{},
+ },
+ } {
+ c.Logf("%#v", trial)
+ s.disp.cluster = &arvados.Cluster{InstanceTypes: trial.types}
+
+ args, err := s.disp.sbatchArgs(container)
+ c.Check(err == nil, Equals, trial.err == nil)
+ if trial.err == nil {
+ c.Check(args, DeepEquals, append([]string{"--job-name=123", "--nice=10000"}, trial.sbatchArgs...))
+ } else {
+ c.Check(len(err.(dispatchcloud.ConstraintsNotSatisfiableError).AvailableTypes), Equals, len(trial.types))
+ }
+ }
}
-func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) {
- theConfig.SbatchArguments = nil
+func (s *StubbedSuite) TestSbatchPartition(c *C) {
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", "--nice=9990", "--partition=blurb,b2")
-
- c.Check(sbatchCmd.Args, DeepEquals, expected)
-}
-
-func (s *TestSuite) TestIntegrationChangePriority(c *C) {
- var scontrolCmdLine []string
- step := 0
+ Priority: 1,
+ }
- 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"})
+ args, err := s.disp.sbatchArgs(container)
+ c.Check(args, DeepEquals, []string{
+ "--job-name=123", "--nice=10000",
+ "--mem=239", "--cpus-per-task=1", "--tmp=0",
+ "--partition=blurb,b2",
+ })
+ c.Check(err, IsNil)
}