//
// SPDX-License-Identifier: AGPL-3.0
-package main
+package dispatchslurm
import (
"bytes"
"context"
"errors"
+ "flag"
"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"
+ "git.arvados.org/arvados.git/lib/cmd"
+ "git.arvados.org/arvados.git/lib/config"
+ "git.arvados.org/arvados.git/lib/dispatchcloud"
+ "git.arvados.org/arvados.git/sdk/go/arvados"
+ "git.arvados.org/arvados.git/sdk/go/arvadosclient"
+ "git.arvados.org/arvados.git/sdk/go/arvadostest"
+ "git.arvados.org/arvados.git/sdk/go/ctxlog"
+ "git.arvados.org/arvados.git/sdk/go/dispatch"
+ "github.com/sirupsen/logrus"
. "gopkg.in/check.v1"
)
}
func (s *IntegrationSuite) SetUpTest(c *C) {
- arvadostest.StartAPI()
+ arvadostest.ResetEnv()
+ arvadostest.ResetDB(c)
os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
s.disp = Dispatcher{}
+ s.disp.cluster = &arvados.Cluster{}
s.disp.setup()
s.slurm = slurmFake{}
}
func (s *IntegrationSuite) TearDownTest(c *C) {
arvadostest.ResetEnv()
- arvadostest.StopAPI()
+ arvadostest.ResetDB(c)
}
type slurmFake struct {
- didBatch [][]string
- didCancel []string
- didRelease []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()
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
}
func (s *IntegrationSuite) integrationTest(c *C,
expectBatch [][]string,
- runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
+ runContainer func(*dispatch.Dispatcher, arvados.Container)) (arvados.Container, error) {
arvadostest.ResetEnv()
arv, err := arvadosclient.MakeArvadosClient()
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{})
+ doneDispatch := make(chan error)
s.disp.Dispatcher = &dispatch.Dispatcher{
Arv: arv,
PollPeriod: time.Second,
- RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+ RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) error {
go func() {
runContainer(disp, ctr)
s.slurm.queue = ""
doneRun <- struct{}{}
}()
- s.disp.runContainer(disp, ctr, status)
+ err := s.disp.runContainer(disp, ctr, status)
cancel()
+ doneDispatch <- err
+ return nil
},
}
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
c.Assert(err, Equals, context.Canceled)
+ errDispatch := <-doneDispatch
s.disp.sqCheck.Stop()
var container arvados.Container
err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
c.Check(err, IsNil)
- return container
+ return container, errDispatch
}
func (s *IntegrationSuite) TestNormal(c *C) {
s.slurm = slurmFake{queue: "zzzzz-dz642-queuedcontainer 10000 100 PENDING Resources\n"}
- container := s.integrationTest(c,
+ container, _ := s.integrationTest(c,
nil,
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
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,
nil,
func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
dispatcher.UpdateState(container.UUID, dispatch.Running)
}
func (s *IntegrationSuite) TestMissingFromSqueue(c *C) {
- 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),
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", "--nice=10000", "--mem=11445", "--cpus-per-task=4", "--tmp=45777"}},
+ container, err := s.integrationTest(c,
+ [][]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)
})
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(err, IsNil)
- c.Assert(len(ll.Items), Equals, 1)
+ c.Check(err, ErrorMatches, `something terrible happened`)
}
type StubbedSuite struct {
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.Second,
- RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+ RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) error {
go func() {
time.Sleep(time.Second)
disp.UpdateState(ctr.UUID, dispatch.Running)
}()
s.disp.runContainer(disp, ctr, status)
cancel()
+ return nil
},
}
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", "--nice=10000", "--mem=239", "--cpus-per-task=2", "--tmp=0"))
+ 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 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", "--nice=10000"}, 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", "--nice=10000",
+ "--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) {
+ log := ctxlog.TestLogger(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 := c.MkDir() + "/config.yml"
+ err := ioutil.WriteFile(tmpfile, content, 0777)
+ c.Assert(err, IsNil)
+
+ os.Setenv("ARVADOS_KEEP_SERVICES", "")
+
+ flags := flag.NewFlagSet("", flag.ContinueOnError)
+ flags.SetOutput(os.Stderr)
+ loader := config.NewLoader(&bytes.Buffer{}, log)
+ loader.SetupFlags(flags)
+ args := loader.MungeLegacyConfigArgs(log, []string{"-config", tmpfile}, "-legacy-"+string(arvados.ServiceNameDispatchSLURM)+"-config")
+ ok, _ := cmd.ParseFlags(flags, "crunch-dispatch-slurm", args, "", os.Stderr)
+ c.Check(ok, Equals, true)
+ cfg, err := loader.Load()
+ c.Assert(err, IsNil)
+ cluster, err := cfg.GetCluster("")
+ c.Assert(err, IsNil)
+
+ c.Check(cluster.Services.Controller.ExternalURL, Equals, arvados.URL{Scheme: "https", Host: "example.com", Path: "/"})
+ c.Check(cluster.SystemRootToken, Equals, "abcdefg")
+ c.Check(cluster.Containers.SLURM.SbatchArgumentsList, DeepEquals, []string{"--foo", "bar"})
+ c.Check(cluster.Containers.CloudVMs.PollInterval, Equals, arvados.Duration(12*time.Second))
+ c.Check(cluster.Containers.SLURM.PrioritySpread, Equals, int64(42))
+ c.Check(cluster.Containers.CrunchRunCommand, Equals, "x-crunch-run")
+ c.Check(cluster.Containers.CrunchRunArgumentsList, DeepEquals, []string{"--cgroup-parent-subsystem=memory"})
+ c.Check(cluster.Containers.ReserveExtraRAM, Equals, arvados.ByteSize(12345))
+ c.Check(cluster.Containers.MinRetryPeriod, Equals, arvados.Duration(13*time.Second))
+ c.Check(cluster.API.MaxItemsPerResponse, Equals, 99)
+ c.Check(cluster.Containers.SLURM.SbatchEnvironmentVariables, DeepEquals, map[string]string{
+ "ARVADOS_KEEP_SERVICES": "https://example.com/keep1 https://example.com/keep2",
+ })
+
+ // Ensure configure() copies SbatchEnvironmentVariables into
+ // the current process's environment (that's how they end up
+ // getting passed to sbatch).
+ s.disp.cluster = cluster
+ s.disp.configure()
+ c.Check(os.Getenv("ARVADOS_KEEP_SERVICES"), Equals, "https://example.com/keep1 https://example.com/keep2")
+}
projects / arvados.git / blobdiff