"fmt"
"io"
"io/ioutil"
- "log"
"net/http"
"net/http/httptest"
"os"
"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"
)
arvadostest.StartAPI()
os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
s.disp = Dispatcher{}
+ s.disp.cluster = &arvados.Cluster{}
s.disp.setup()
s.slurm = slurmFake{}
}
}
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
}
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)
- s.disp.CrunchRunCommand = []string{"echo"}
+ s.disp.cluster.Containers.CrunchRunCommand = "echo"
ctx, cancel := context.WithCancel(context.Background())
doneRun := make(chan struct{})
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.disp.slurm = &s.slurm
- s.disp.sqCheck = &SqueueChecker{Period: 500 * time.Millisecond, Slurm: s.disp.slurm}
+ s.disp.sqCheck = &SqueueChecker{
+ Logger: logrus.StandardLogger(),
+ Period: 500 * time.Millisecond,
+ Slurm: s.disp.slurm,
+ }
err = s.disp.Dispatcher.Run(ctx)
<-doneRun
}
func (s *IntegrationSuite) TestNormal(c *C) {
- s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"}
+ s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
container := s.integrationTest(c,
nil,
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
}
func (s *IntegrationSuite) TestCancel(c *C) {
- s.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)
s.slurm.onCancel = func() { <-readyToCancel }
container := s.integrationTest(c,
container := s.integrationTest(c,
[][]string{{
fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
+ fmt.Sprintf("--nice=%d", 10000),
+ "--no-requeue",
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,
- [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--mem=11445", "--cpus-per-task=4", "--tmp=45777", "--nice=9990"}},
+ [][]string{{"--job-name=zzzzz-dz642-queuedcontainer", "--nice=10000", "--no-requeue", "--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) {
- s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 9990 100\n"}
- container := s.integrationTest(c, 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(s.slurm.didRenice), Not(Equals), 0)
- c.Check(s.slurm.didRenice[len(s.slurm.didRenice)-1], DeepEquals, []string{"zzzzz-dz642-queuedcontainer", "4000"})
-}
-
type StubbedSuite struct {
disp Dispatcher
}
func (s *StubbedSuite) SetUpTest(c *C) {
s.disp = Dispatcher{}
+ s.disp.cluster = &arvados.Cluster{}
s.disp.setup()
}
apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`}
apiStubResponses["/arvados/v1/containers"] = arvadostest.StubResponse{500, string(`{}`)}
- s.testWithServerStub(c, apiStubResponses, "echo", "Error getting list of containers")
+ s.testWithServerStub(c, apiStubResponses, "echo", "error getting count of containers")
}
func (s *StubbedSuite) testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
}
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)
- s.disp.CrunchRunCommand = []string{crunchCmd}
+ s.disp.cluster.Containers.CrunchRunCommand = "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)
}()
c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
}
-func (s *StubbedSuite) TestNoSuchConfigFile(c *C) {
- err := s.disp.readConfig("/nosuchdir89j7879/8hjwr7ojgyy7")
- c.Assert(err, NotNil)
-}
-
-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 = s.disp.readConfig(tmpfile.Name())
- c.Assert(err, NotNil)
-}
-
-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 = s.disp.readConfig(tmpfile.Name())
- c.Assert(err, IsNil)
- c.Check(0, Equals, len(s.disp.SbatchArguments))
-}
-
-func (s *StubbedSuite) TestReadConfig(c *C) {
- tmpfile, err := ioutil.TempFile(os.TempDir(), "config")
- c.Check(err, IsNil)
- defer os.Remove(tmpfile.Name())
-
- args := []string{"--arg1=v1", "--arg2", "--arg3=v3"}
- argsS := `{"SbatchArguments": ["--arg1=v1", "--arg2", "--arg3=v3"]}`
- _, err = tmpfile.Write([]byte(argsS))
- c.Check(err, IsNil)
-
- err = s.disp.readConfig(tmpfile.Name())
- c.Assert(err, IsNil)
- c.Check(args, DeepEquals, s.disp.SbatchArguments)
-}
-
func (s *StubbedSuite) TestSbatchArgs(c *C) {
container := arvados.Container{
UUID: "123",
{"--arg1=v1", "--arg2"},
} {
c.Logf("%#v", defaults)
- s.disp.SbatchArguments = defaults
+ s.disp.cluster.Containers.SLURM.SbatchArgumentsList = defaults
args, err := s.disp.sbatchArgs(container)
- c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990"))
+ c.Check(args, DeepEquals, append(defaults, "--job-name=123", "--nice=10000", "--no-requeue", "--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},
- {Name: "a1.large", Price: 0.16, RAM: 1024000000, VCPUs: 8},
+ 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: 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.disp.cluster = &arvados.Cluster{InstanceTypes: trial.types}
args, err := s.disp.sbatchArgs(container)
- c.Check(err, Equals, trial.err)
+ 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", "--no-requeue"}, trial.sbatchArgs...))
+ } else {
+ c.Check(len(err.(dispatchcloud.ConstraintsNotSatisfiableError).AvailableTypes), Equals, len(trial.types))
}
}
}
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", "--no-requeue",
+ "--mem=239", "--cpus-per-task=1", "--tmp=0",
"--partition=blurb,b2",
})
c.Check(err, IsNil)
}
+
+func (s *StubbedSuite) TestLoadLegacyConfig(c *C) {
+ content := []byte(`
+Client:
+ APIHost: example.com
+ AuthToken: abcdefg
+ KeepServiceURIs:
+ - https://example.com/keep1
+ - https://example.com/keep2
+SbatchArguments: ["--foo", "bar"]
+PollPeriod: 12s
+PrioritySpread: 42
+CrunchRunCommand: ["x-crunch-run", "--cgroup-parent-subsystem=memory"]
+ReserveExtraRAM: 12345
+MinRetryPeriod: 13s
+BatchSize: 99
+`)
+ tmpfile, err := ioutil.TempFile("", "example")
+ if err != nil {
+ c.Error(err)
+ }
+
+ defer os.Remove(tmpfile.Name()) // clean up
+
+ if _, err := tmpfile.Write(content); err != nil {
+ c.Error(err)
+ }
+ if err := tmpfile.Close(); err != nil {
+ c.Error(err)
+
+ }
+ err = s.disp.configure("crunch-dispatch-slurm", []string{"-config", tmpfile.Name()})
+ c.Check(err, IsNil)
+
+ c.Check(s.disp.cluster.Services.Controller.ExternalURL, Equals, arvados.URL{Scheme: "https", Host: "example.com"})
+ c.Check(s.disp.cluster.SystemRootToken, Equals, "abcdefg")
+ c.Check(s.disp.cluster.Containers.SLURM.SbatchArgumentsList, DeepEquals, []string{"--foo", "bar"})
+ c.Check(s.disp.cluster.Containers.CloudVMs.PollInterval, Equals, arvados.Duration(12*time.Second))
+ c.Check(s.disp.cluster.Containers.SLURM.PrioritySpread, Equals, int64(42))
+ c.Check(s.disp.cluster.Containers.CrunchRunCommand, Equals, "x-crunch-run")
+ c.Check(s.disp.cluster.Containers.CrunchRunArgumentsList, DeepEquals, []string{"--cgroup-parent-subsystem=memory"})
+ c.Check(s.disp.cluster.Containers.ReserveExtraRAM, Equals, arvados.ByteSize(12345))
+ c.Check(s.disp.cluster.Containers.MinRetryPeriod, Equals, arvados.Duration(13*time.Second))
+ c.Check(s.disp.cluster.API.MaxItemsPerResponse, Equals, 99)
+ c.Check(s.disp.cluster.Containers.SLURM.KeepServices, DeepEquals, map[string]arvados.Service{
+ "00000-bi6l4-000000000000000": arvados.Service{
+ InternalURLs: map[arvados.URL]arvados.ServiceInstance{
+ arvados.URL{Scheme: "https", Path: "/keep1", Host: "example.com"}: struct{}{},
+ arvados.URL{Scheme: "https", Path: "/keep2", Host: "example.com"}: struct{}{},
+ },
+ },
+ })
+ ks := os.Getenv("ARVADOS_KEEP_SERVICES")
+ if ks != "https://example.com/keep1 https://example.com/keep2" && ks != "https://example.com/keep2 https://example.com/keep1" {
+ c.Assert(ks, Equals, "https://example.com/keep1 https://example.com/keep2")
+ }
+}
projects / arvados.git / blobdiff