"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"
- "git.curoverse.com/arvados.git/services/dispatchcloud"
. "gopkg.in/check.v1"
)
var _ = Suite(&StubbedSuite{})
type IntegrationSuite struct {
- cmd command
+ disp Dispatcher
+ slurm slurmFake
}
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 *IntegrationSuite) TearDownTest(c *C) {
}
type slurmFake struct {
- didBatch [][]string
- didCancel []string
- didRenice [][]string
- queue string
+ 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()
return exec.Command("echo", sf.queue)
}
-func (sf *slurmFake) Renice(name string, nice int) error {
+func (sf *slurmFake) Release(name string) error {
+ sf.didRelease = append(sf.didRelease, name)
+ return nil
+}
+
+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
}
return nil
}
-func (s *IntegrationSuite) integrationTest(c *C, slurm *slurmFake,
+func (s *IntegrationSuite) integrationTest(c *C,
expectBatch [][]string,
runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
arvadostest.ResetEnv()
arv, err := arvadosclient.MakeArvadosClient()
c.Assert(err, IsNil)
- s.cmd.slurm = slurm
-
// There should be one queued container
params := arvadosclient.Dict{
"filters": [][]string{{"state", "=", "Queued"}},
c.Check(err, IsNil)
c.Check(len(containers.Items), Equals, 1)
- s.cmd.CrunchRunCommand = []string{"echo"}
+ s.disp.CrunchRunCommand = []string{"echo"}
ctx, cancel := context.WithCancel(context.Background())
doneRun := make(chan struct{})
- s.cmd.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)
- slurm.queue = ""
+ s.slurm.queue = ""
doneRun <- struct{}{}
}()
- s.cmd.run(disp, ctr, status)
+ s.disp.runContainer(disp, ctr, status)
cancel()
},
}
- s.cmd.sqCheck = &SqueueChecker{Period: 500 * time.Millisecond, Slurm: slurm}
+ s.disp.slurm = &s.slurm
+ s.disp.sqCheck = &SqueueChecker{Period: 500 * time.Millisecond, Slurm: s.disp.slurm}
- err = s.cmd.dispatcher.Run(ctx)
+ err = s.disp.Dispatcher.Run(ctx)
<-doneRun
c.Assert(err, Equals, context.Canceled)
- s.cmd.sqCheck.Stop()
+ s.disp.sqCheck.Stop()
- c.Check(slurm.didBatch, DeepEquals, expectBatch)
+ c.Check(s.slurm.didBatch, DeepEquals, expectBatch)
// There should be no queued containers now
err = arv.List("containers", params, &containers)
}
func (s *IntegrationSuite) TestNormal(c *C) {
+ s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
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)
}
func (s *IntegrationSuite) TestCancel(c *C) {
- slurm := &slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"}
+ s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
readyToCancel := make(chan bool)
- slurm.onCancel = func() { <-readyToCancel }
+ s.slurm.onCancel = func() { <-readyToCancel }
container := s.integrationTest(c,
- slurm,
nil,
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
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"})
+ 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 *IntegrationSuite) TestMissingFromSqueue(c *C) {
- container := s.integrationTest(c, &slurmFake{},
+ container := s.integrationTest(c,
[][]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)
}
func (s *IntegrationSuite) TestSbatchFail(c *C) {
+ s.slurm = slurmFake{errBatch: errors.New("something terrible happened")}
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"}},
+ [][]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 *IntegrationSuite) TestChangePriority(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"})
-}
-
type StubbedSuite struct {
- cmd command
+ disp Dispatcher
}
func (s *StubbedSuite) SetUpTest(c *C) {
- s.cmd = command{}
+ s.disp = Dispatcher{}
+ s.disp.setup()
}
func (s *StubbedSuite) TestAPIErrorGettingContainers(c *C) {
log.SetOutput(io.MultiWriter(buf, os.Stderr))
defer log.SetOutput(os.Stderr)
- s.cmd.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)
}()
- s.cmd.run(disp, ctr, status)
+ s.disp.runContainer(disp, ctr, status)
cancel()
},
}
}
func (s *StubbedSuite) TestNoSuchConfigFile(c *C) {
- err := s.cmd.readConfig("/nosuchdir89j7879/8hjwr7ojgyy7")
+ err := s.disp.readConfig("/nosuchdir89j7879/8hjwr7ojgyy7")
c.Assert(err, NotNil)
}
_, err = tmpfile.Write([]byte(`{"SbatchArguments": "oops this is not a string array"}`))
c.Check(err, IsNil)
- err = s.cmd.readConfig(tmpfile.Name())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, NotNil)
}
_, err = tmpfile.Write([]byte(`{"NoSuchArg": "Nobody loves me, not one tiny hunk."}`))
c.Check(err, IsNil)
- err = s.cmd.readConfig(tmpfile.Name())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, IsNil)
- c.Check(0, Equals, len(s.cmd.SbatchArguments))
+ c.Check(0, Equals, len(s.disp.SbatchArguments))
}
func (s *StubbedSuite) TestReadConfig(c *C) {
_, err = tmpfile.Write([]byte(argsS))
c.Check(err, IsNil)
- err = s.cmd.readConfig(tmpfile.Name())
+ err = s.disp.readConfig(tmpfile.Name())
c.Assert(err, IsNil)
- c.Check(args, DeepEquals, s.cmd.SbatchArguments)
+ c.Check(args, DeepEquals, s.disp.SbatchArguments)
}
func (s *StubbedSuite) TestSbatchArgs(c *C) {
{"--arg1=v1", "--arg2"},
} {
c.Logf("%#v", defaults)
- s.cmd.SbatchArguments = defaults
+ s.disp.SbatchArguments = defaults
- args, err := s.cmd.sbatchArgs(container)
- c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990"))
+ 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)
}
}
}
for _, trial := range []struct {
- types []arvados.InstanceType
+ types map[string]arvados.InstanceType
sbatchArgs []string
err error
}{
// Choose node type => use --constraint arg
{
- types: []arvados.InstanceType{
- {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
- {Name: "a1.small", Price: 0.04, RAM: 256000000, VCPUs: 2},
- {Name: "a1.medium", Price: 0.08, RAM: 512000000, VCPUs: 4},
+ 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=a1.small"},
+ sbatchArgs: []string{"--constraint=instancetype=a1.medium"},
},
// No node types configured => no slurm constraint
{
types: nil,
- sbatchArgs: nil,
+ sbatchArgs: []string{"--mem=239", "--cpus-per-task=2", "--tmp=0"},
},
// No node type is big enough => error
{
- types: []arvados.InstanceType{
- {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
+ types: map[string]arvados.InstanceType{
+ "a1.tiny": {Name: "a1.tiny", Price: 0.02, RAM: 128000000, VCPUs: 1},
},
- err: dispatchcloud.ErrConstraintsNotSatisfiable,
+ err: dispatchcloud.ConstraintsNotSatisfiableError{},
},
} {
c.Logf("%#v", trial)
- s.cmd.cluster = &arvados.Cluster{InstanceTypes: trial.types}
+ s.disp.cluster = &arvados.Cluster{InstanceTypes: trial.types}
- args, err := s.cmd.sbatchArgs(container)
- c.Check(err, Equals, trial.err)
+ 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", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990"}, trial.sbatchArgs...))
+ 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))
}
}
}
Priority: 1,
}
- args, err := s.cmd.sbatchArgs(container)
+ args, err := s.disp.sbatchArgs(container)
c.Check(args, DeepEquals, []string{
- "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--nice=9990",
+ "--job-name=123", "--nice=10000",
+ "--mem=239", "--cpus-per-task=1", "--tmp=0",
"--partition=blurb,b2",
})
c.Check(err, IsNil)