First version as separate package

Signed-off-by: Geoffroy Vallee <[email protected]>

go.mod

@@ -0,0 +1,9 @@
+module github.com/gvallee/go_hpc_jobmgr
+
+go 1.13
+
+require (
+	github.com/gvallee/go_exec v0.0.1
+	github.com/gvallee/go_util v1.0.1
+	github.com/gvallee/kv v1.0.0
+)

go.sum

@@ -0,0 +1,4 @@
+github.com/gvallee/go_exec v0.0.1 h1:yNP5/fTWnnym4wT17JIEBiRAxOEhclOePdwxCxWsEZ8=
+github.com/gvallee/go_exec v0.0.1/go.mod h1:4AwegK9oPhkgwkd0rjlTwxRw//8cW4pPcCSFLZ6+LZg=
+github.com/gvallee/go_util v1.0.1/go.mod h1:fTexpwdH/n05Ziu0TXJIQsr7E+46QpBxNdeOOsyC0/s=
+github.com/gvallee/kv v1.0.0/go.mod h1:sfSclfFfLV+Y+9e9FayIbBUOtvbt1779S6q52bSSU5E=

internal/pkg/job/job.go

@@ -0,0 +1,70 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package job
+
+import (
+	"bytes"
+
+	"github.com/gvallee/go_hpc_jobmgr/internal/pkg/sys"
+	"github.com/gvallee/go_hpc_jobmgr/pkg/app"
+	"github.com/gvallee/go_hpc_jobmgr/pkg/implem"
+	"github.com/gvallee/go_hpc_jobmgr/pkg/mpi"
+)
+
+// CleanUpFn is a "function pointer" to call to clean up the system after the completion of a job
+type CleanUpFn func(...interface{}) error
+
+// GetOutputFn is a "function pointer" to call to gather the output of an application after completion of a job
+type GetOutputFn func(*Job, *sys.Config) string
+
+// GetErrorFn is a "function pointer" to call to gather stderr from an application after completion of a job
+type GetErrorFn func(*Job, *sys.Config) string
+
+// Job represents a job
+type Job struct {
+	// Name is the name of the job
+	Name string
+
+	// NP is the number of ranks
+	NP int
+
+	// NNodes is the number of nodes
+	NNodes int
+
+	// CleanUp is the function to call once the job is completed to clean the system
+	CleanUp CleanUpFn
+
+	// BatchScript is the path to the script required to start a job (optional)
+	BatchScript string
+
+	// HostCfg is the MPI configuration to use on the host
+	HostCfg *implem.Info
+
+	// App is the path to the application's binary, i.e., the binary to start
+	App app.Info
+
+	// OutBuffer is a buffer with the output of the job
+	OutBuffer bytes.Buffer
+
+	// ErrBuffer is a buffer with the stderr of the job
+	ErrBuffer bytes.Buffer
+
+	// GetOutput is the function to call to gather the output of the application based on the use of a given job manager
+	GetOutput GetOutputFn
+
+	// GetError is the function to call to gather stderr of the application based on the use of a given job manager
+	GetError GetErrorFn
+
+	// Args is a set of arguments to be used for launching the job
+	Args []string
+
+	// MPICfg is the MPI configuration to use for the execution of the job
+	MPICfg *mpi.Config
+
+	// Partition is the name of the partition to use with the jobmgr (optional)
+	Partition string
+}

internal/pkg/mpich/mpich.go

@@ -0,0 +1,27 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package mpich
+
+const (
+	// VersionTag is the tag used to refer to the MPI version in MPICH template(s)
+	VersionTag = "MPICHVERSION"
+	// URLTag is the tag used to refer to the MPI URL in MPICH template(s)
+	URLTag = "MPICHURL"
+	// TarballTag is the tag used to refer to the MPI tarball in MPICH template(s)
+	TarballTag = "MPICHTARBALL"
+)
+
+// MPICHGetExtraMpirunArgs returns the extra mpirun arguments required by MPICH for a specific configuration
+func MPICHGetExtraMpirunArgs() []string {
+	var extraArgs []string
+	return extraArgs
+}
+
+// MPICHGetConfigureExtraArgs returns the extra arguments required to configure MPICH
+func MPICHGetConfigureExtraArgs() []string {
+	var extraArgs []string
+	return extraArgs
+}

internal/pkg/openmpi/openmpi.go

@@ -0,0 +1,35 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package openmpi
+
+import (
+	"github.com/gvallee/go_hpc_jobmgr/internal/pkg/sys"
+)
+
+const (
+	// VersionTag is the tag used to refer to the MPI version in Open MPI template(s)
+	VersionTag = "OMPIVERSION"
+
+	// URLTag is the tag used to refer to the MPI URL in Open MPI template(s)
+	URLTag = "OMPIURL"
+
+	// TarballTag is the tag used to refer to the MPI tarball in Open MPI template(s)
+	TarballTag = "OMPITARBALL"
+)
+
+// GetExtraMpirunArgs returns the set of arguments required for the mpirun command for the target platform
+func GetExtraMpirunArgs(sys *sys.Config) []string {
+	var extraArgs []string
+	/*
+		if sys.IBEnabled {
+			extraArgs = append(extraArgs, "--mca")
+			extraArgs = append(extraArgs, "btl")
+			extraArgs = append(extraArgs, "openib,self,vader")
+		}
+	*/
+
+	return extraArgs
+}

internal/pkg/slurm/slurm.go

@@ -0,0 +1,18 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package slurm
+
+const (
+	// SlurmParitionKey is the key to use to retrieve the optinal parition id that
+	// can be specified in the tool's configuration file.
+	PartitionKey = "slurm_partition"
+
+	// EnabledKey is the key used in the singularity-mpi.conf file to specify if Slurm shall be used
+	EnabledKey = "enable_slurm"
+
+	// ScriptCmdPrefix is the prefix to add to a script
+	ScriptCmdPrefix = "#SBATCH"
+)

internal/pkg/sys/sys.go

@@ -0,0 +1,12 @@
+// Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package sys
+
+type Config struct {
+	ScratchDir string
+
+	Persistent string
+}

pkg/app/app.go

@@ -0,0 +1,21 @@
+// Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package app
+
+// Info gathers information about a given application
+type Info struct {
+	// Name is the name of the application
+	Name string
+
+	// BinName is the name of the binary to start executing the application
+	BinName string
+
+	// BinPath is the path to the binary to start executing the application
+	BinPath string
+
+	// BinArgs is the list of argument that the application's binary needs
+	BinArgs []string
+}

pkg/implem/implem.go

@@ -0,0 +1,35 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package implem
+
+const (
+	// OMPI is the identifier for Open MPI
+	OMPI = "openmpi"
+
+	// MPICH is the identifier for MPICH
+	MPICH = "mpich"
+)
+
+// Info gathers all data about a specific MPI implementation
+type Info struct {
+	// ID is the string idenfifying the MPI implementation
+	ID string
+
+	// Version is the version of the MPI implementation
+	Version string
+
+	// InstallDir is where the MPI implementation is installed
+	InstallDir string
+}
+
+// IsMPI checks if information passed in is an MPI implementation
+func IsMPI(i *Info) bool {
+	if i != nil && (i.ID == OMPI || i.ID == MPICH) {
+		return true
+	}
+
+	return false
+}

pkg/jm/jm.go

@@ -0,0 +1,118 @@
+// Copyright (c) 2019, Sylabs Inc. All rights reserved.
+// This software is licensed under a 3-clause BSD license. Please consult the
+// LICENSE.md file distributed with the sources of this project regarding your
+// rights to use or distribute this software.
+
+package jm
+
+import (
+	"fmt"
+	"io/ioutil"
+	"log"
+	"os"
+	"path/filepath"
+
+	"github.com/gvallee/go_exec/pkg/advexec"
+	"github.com/gvallee/go_hpc_jobmgr/internal/pkg/job"
+	"github.com/gvallee/go_hpc_jobmgr/internal/pkg/sys"
+	"github.com/gvallee/go_util/pkg/util"
+)
+
+const (
+	// NativeID is the value set to JM.ID when mpirun shall be used to submit a job
+	NativeID = "native"
+
+	// SlurmID is the value set to JM.ID when Slurm shall be used to submit a job
+	SlurmID = "slurm"
+
+	// PrunID is the value set to JM.ID when prun shall be used to submit a job
+	PrunID = "prun"
+)
+
+type Environment struct {
+	InstallDir string
+
+	mpiBin string
+}
+
+// Loader checks whether a giv job manager is applicable or not
+type Loader interface {
+	Load() bool
+}
+
+// LoadFn loads a specific job manager once detected
+type LoadFn func(*JM, *sys.Config) error
+
+// SubmitFn is a "function pointer" that lets us job a new job
+type SubmitFn func(*job.Job, *sys.Config) (advexec.Advcmd, error)
+
+// JM is the structure representing a specific JM
+type JM struct {
+	// ID identifies which job manager has been detected on the system
+	ID string
+
+	Load LoadFn
+
+	// Submit is the function to submit a job through the current job manager
+	Submit SubmitFn
+}
+
+// Detect figures out which job manager must be used on the system and return a
+// structure that gather all the data necessary to interact with it
+func Detect() JM {
+	// Default job manager
+	loaded, comp := NativeDetect()
+	if !loaded {
+		log.Fatalln("unable to find a default job manager")
+	}
+
+	// Now we check if we can find better
+	loaded, slurmComp := SlurmDetect()
+	if loaded {
+		return slurmComp
+	}
+
+	loaded, prunComp := PrunDetect()
+	if loaded {
+		return prunComp
+	}
+
+	return comp
+}
+
+// Load is the function to use to load the JM component
+func Load(jm *JM) error {
+	return nil
+}
+
+// TempFile creates a temporary file that is used to store a batch script
+func TempFile(j *job.Job, sysCfg *sys.Config) error {
+	filePrefix := "sbash-" + j.Name
+	path := ""
+	if sysCfg.Persistent == "" {
+		f, err := ioutil.TempFile("", filePrefix+"-")
+		if err != nil {
+			return fmt.Errorf("failed to create temporary file: %s", err)
+		}
+		path = f.Name()
+		f.Close()
+		j.BatchScript = path
+	} else {
+		fileName := filePrefix + ".sh"
+		path = filepath.Join(j.MPICfg.Implem.InstallDir, fileName)
+		j.BatchScript = path
+		if util.PathExists(path) {
+			return fmt.Errorf("file %s already exists", path)
+		}
+	}
+
+	j.CleanUp = func(...interface{}) error {
+		err := os.RemoveAll(path)
+		if err != nil {
+			return fmt.Errorf("unable to delete %s: %s", path, err)
+		}
+		return nil
+	}
+
+	return nil
+}