// Dispatcher service for Crunch that submits containers to the slurm queue.
import (
+ "bytes"
"context"
"flag"
"fmt"
"strings"
"time"
+ "git.curoverse.com/arvados.git/lib/config"
"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/config"
"git.curoverse.com/arvados.git/sdk/go/dispatch"
"github.com/coreos/go-systemd/daemon"
+ "github.com/ghodss/yaml"
+ "github.com/sirupsen/logrus"
)
+type logger interface {
+ dispatch.Logger
+ Fatalf(string, ...interface{})
+}
+
const initialNiceValue int64 = 10000
var (
- version = "dev"
- defaultConfigPath = "/etc/arvados/crunch-dispatch-slurm/crunch-dispatch-slurm.yml"
+ version = "dev"
)
type Dispatcher struct {
*dispatch.Dispatcher
+ logger logrus.FieldLogger
cluster *arvados.Cluster
sqCheck *SqueueChecker
slurm Slurm
Client arvados.Client
-
- SbatchArguments []string
- PollPeriod arvados.Duration
- PrioritySpread int64
-
- // crunch-run command to invoke. The container UUID will be
- // appended. If nil, []string{"crunch-run"} will be used.
- //
- // Example: []string{"crunch-run", "--cgroup-parent-subsystem=memory"}
- CrunchRunCommand []string
-
- // Extra RAM to reserve (in Bytes) for SLURM job, in addition
- // to the amount specified in the container's RuntimeConstraints
- ReserveExtraRAM int64
-
- // Minimum time between two attempts to run the same container
- MinRetryPeriod arvados.Duration
}
func main() {
- disp := &Dispatcher{}
+ logger := logrus.StandardLogger()
+ if os.Getenv("DEBUG") != "" {
+ logger.SetLevel(logrus.DebugLevel)
+ }
+ logger.Formatter = &logrus.JSONFormatter{
+ TimestampFormat: "2006-01-02T15:04:05.000000000Z07:00",
+ }
+ disp := &Dispatcher{logger: logger}
err := disp.Run(os.Args[0], os.Args[1:])
if err != nil {
- log.Fatal(err)
+ logrus.Fatalf("%s", err)
}
}
// configure() loads config files. Tests skip this.
func (disp *Dispatcher) configure(prog string, args []string) error {
+ if disp.logger == nil {
+ disp.logger = logrus.StandardLogger()
+ }
flags := flag.NewFlagSet(prog, flag.ExitOnError)
flags.Usage = func() { usage(flags) }
- configPath := flags.String(
- "config",
- defaultConfigPath,
- "`path` to JSON or YAML configuration file")
+ loader := config.NewLoader(nil, disp.logger)
+ loader.SetupFlags(flags)
+
dumpConfig := flag.Bool(
"dump-config",
false,
"version",
false,
"Print version information and exit.")
+
+ args = loader.MungeLegacyConfigArgs(logrus.StandardLogger(), args, "-legacy-crunch-dispatch-slurm-config")
+
// Parse args; omit the first arg which is the command name
- flags.Parse(args)
+ err := flags.Parse(args)
+
+ if err == flag.ErrHelp {
+ return nil
+ }
// Print version information if requested
if *getVersion {
return nil
}
- log.Printf("crunch-dispatch-slurm %s started", version)
+ disp.logger.Printf("crunch-dispatch-slurm %s started", version)
- err := disp.readConfig(*configPath)
+ cfg, err := loader.Load()
if err != nil {
return err
}
- if disp.CrunchRunCommand == nil {
- disp.CrunchRunCommand = []string{"crunch-run"}
+ if disp.cluster, err = cfg.GetCluster(""); err != nil {
+ return fmt.Errorf("config error: %s", err)
}
- if disp.PollPeriod == 0 {
- disp.PollPeriod = arvados.Duration(10 * time.Second)
- }
+ disp.Client.APIHost = disp.cluster.Services.Controller.ExternalURL.Host
+ disp.Client.AuthToken = disp.cluster.SystemRootToken
+ disp.Client.Insecure = disp.cluster.TLS.Insecure
if disp.Client.APIHost != "" || disp.Client.AuthToken != "" {
// Copy real configs into env vars so [a]
if disp.Client.Insecure {
os.Setenv("ARVADOS_API_HOST_INSECURE", "1")
}
- os.Setenv("ARVADOS_KEEP_SERVICES", strings.Join(disp.Client.KeepServiceURIs, " "))
+ ks := ""
+ if length(disp.cluster.Containers.SLURM.KeepServices) > 0 {
+ for _, svc := range disp.cluster.Containers.SLURM.KeepServices {
+ for k, _ := range svc.InternalURLs {
+ ks += k
+ ks += " "
+ }
+ }
+ }
+ os.Setenv("ARVADOS_KEEP_SERVICES", ks)
os.Setenv("ARVADOS_EXTERNAL_CLIENT", "")
} else {
- log.Printf("warning: Client credentials missing from config, so falling back on environment variables (deprecated).")
+ disp.logger.Warnf("Client credentials missing from config, so falling back on environment variables (deprecated).")
}
if *dumpConfig {
- return config.DumpAndExit(disp)
- }
-
- siteConfig, err := arvados.GetConfig(arvados.DefaultConfigFile)
- if os.IsNotExist(err) {
- log.Printf("warning: no cluster config (%s), proceeding with no node types defined", err)
- } else if err != nil {
- return fmt.Errorf("error loading config: %s", err)
- } else if disp.cluster, err = siteConfig.GetCluster(""); err != nil {
- return fmt.Errorf("config error: %s", err)
+ out, err := yaml.Marshal(cfg)
+ if err != nil {
+ return err
+ }
+ _, err = os.Stdout.Write(out)
+ if err != nil {
+ return err
+ }
}
return nil
func (disp *Dispatcher) setup() {
arv, err := arvadosclient.MakeArvadosClient()
if err != nil {
- log.Fatalf("Error making Arvados client: %v", err)
+ disp.logger.Fatalf("Error making Arvados client: %v", err)
}
arv.Retries = 25
- disp.slurm = &slurmCLI{}
+ disp.slurm = NewSlurmCLI()
disp.sqCheck = &SqueueChecker{
- Period: time.Duration(disp.PollPeriod),
- PrioritySpread: disp.PrioritySpread,
+ Logger: disp.logger,
+ Period: time.Duration(disp.cluster.Containers.CloudVMs.PollInterval),
+ PrioritySpread: disp.cluster.Containers.SLURM.PrioritySpread,
Slurm: disp.slurm,
}
disp.Dispatcher = &dispatch.Dispatcher{
Arv: arv,
+ Logger: disp.logger,
+ BatchSize: disp.cluster.API.MaxItemsPerResponse,
RunContainer: disp.runContainer,
- PollPeriod: time.Duration(disp.PollPeriod),
- MinRetryPeriod: time.Duration(disp.MinRetryPeriod),
+ PollPeriod: time.Duration(disp.cluster.Containers.CloudVMs.PollInterval),
+ MinRetryPeriod: time.Duration(disp.cluster.Containers.MinRetryPeriod),
}
}
defer disp.sqCheck.Stop()
if disp.cluster != nil && len(disp.cluster.InstanceTypes) > 0 {
- go dispatchcloud.SlurmNodeTypeFeatureKludge(disp.cluster)
+ go SlurmNodeTypeFeatureKludge(disp.cluster)
}
if _, err := daemon.SdNotify(false, "READY=1"); err != nil {
}
func (disp *Dispatcher) slurmConstraintArgs(container arvados.Container) []string {
- mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+container.RuntimeConstraints.KeepCacheRAM+disp.ReserveExtraRAM) / float64(1048576)))
+ mem := int64(math.Ceil(float64(container.RuntimeConstraints.RAM+
+ container.RuntimeConstraints.KeepCacheRAM+
+ int64(disp.cluster.Containers.ReserveExtraRAM)) / float64(1048576)))
- var disk int64
- for _, m := range container.Mounts {
- if m.Kind == "tmp" {
- disk += m.Capacity
- }
- }
+ disk := dispatchcloud.EstimateScratchSpace(&container)
disk = int64(math.Ceil(float64(disk) / float64(1048576)))
return []string{
fmt.Sprintf("--mem=%d", mem),
func (disp *Dispatcher) sbatchArgs(container arvados.Container) ([]string, error) {
var args []string
- args = append(args, disp.SbatchArguments...)
- args = append(args, "--job-name="+container.UUID, fmt.Sprintf("--nice=%d", initialNiceValue))
+ args = append(args, disp.cluster.Containers.SLURM.SbatchArgumentsList...)
+ args = append(args, "--job-name="+container.UUID, fmt.Sprintf("--nice=%d", initialNiceValue), "--no-requeue")
if disp.cluster == nil {
// no instance types configured
crArgs = append(crArgs, container.UUID)
crScript := strings.NewReader(execScript(crArgs))
- disp.sqCheck.L.Lock()
- defer disp.sqCheck.L.Unlock()
-
sbArgs, err := disp.sbatchArgs(container)
if err != nil {
return err
if ctr.State == dispatch.Locked && !disp.sqCheck.HasUUID(ctr.UUID) {
log.Printf("Submitting container %s to slurm", ctr.UUID)
- if err := disp.submit(ctr, disp.CrunchRunCommand); err != nil {
+ cmd := []string{disp.cluster.Containers.CrunchRunCommand}
+ cmd = append(cmd, disp.cluster.Containers.CrunchRunArgumentsList...)
+ if err := disp.submit(ctr, cmd); err != nil {
var text string
- if err == dispatchcloud.ErrConstraintsNotSatisfiable {
- text = fmt.Sprintf("cannot run container %s: %s", ctr.UUID, err)
+ if err, ok := err.(dispatchcloud.ConstraintsNotSatisfiableError); ok {
+ var logBuf bytes.Buffer
+ fmt.Fprintf(&logBuf, "cannot run container %s: %s\n", ctr.UUID, err)
+ if len(err.AvailableTypes) == 0 {
+ fmt.Fprint(&logBuf, "No instance types are configured.\n")
+ } else {
+ fmt.Fprint(&logBuf, "Available instance types:\n")
+ for _, t := range err.AvailableTypes {
+ fmt.Fprintf(&logBuf,
+ "Type %q: %d VCPUs, %d RAM, %d Scratch, %f Price\n",
+ t.Name, t.VCPUs, t.RAM, t.Scratch, t.Price,
+ )
+ }
+ }
+ text = logBuf.String()
disp.UpdateState(ctr.UUID, dispatch.Cancelled)
} else {
text = fmt.Sprintf("Error submitting container %s to slurm: %s", ctr.UUID, err)
case <-ctx.Done():
// Disappeared from squeue
if err := disp.Arv.Get("containers", ctr.UUID, nil, &ctr); err != nil {
- log.Printf("Error getting final container state for %s: %s", ctr.UUID, err)
+ log.Printf("error getting final container state for %s: %s", ctr.UUID, err)
}
switch ctr.State {
case dispatch.Running:
}
}
func (disp *Dispatcher) scancel(ctr arvados.Container) {
- disp.sqCheck.L.Lock()
err := disp.slurm.Cancel(ctr.UUID)
- disp.sqCheck.L.Unlock()
-
if err != nil {
log.Printf("scancel: %s", err)
time.Sleep(time.Second)
time.Sleep(time.Second)
}
}
-
-func (disp *Dispatcher) readConfig(path string) error {
- err := config.LoadFile(disp, path)
- if err != nil && os.IsNotExist(err) && path == defaultConfigPath {
- log.Printf("Config not specified. Continue with default configuration.")
- err = nil
- }
- return err
-}
projects / arvados.git / blobdiff