Create predectable and unique job script names

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

go.mod

@@ -3,6 +3,6 @@ module github.com/gvallee/go_hpc_jobmgr
 go 1.13
 
 require (
-	github.com/gvallee/go_exec v1.0.0
-	github.com/gvallee/go_util v1.1.0
+	github.com/gvallee/go_exec v1.1.0
+	github.com/gvallee/go_util v1.5.0
 )

go.sum

@@ -1,4 +1,5 @@
-github.com/gvallee/go_exec v1.0.0 h1:43nuopP/x1+34yeZX0Z/tv7eI68SgLDwHygQMwVtAkE=
-github.com/gvallee/go_exec v1.0.0/go.mod h1:s9fSkVniLYlQUZb99zuioVH45nkvnYxw6vmGEaZCUPk=
-github.com/gvallee/go_util v1.1.0 h1:qSwjCPTejJ8zbhFmpJ3WQtzpNOHWQwc4ijFsTbCJK7s=
+github.com/gvallee/go_exec v1.1.0 h1:3xhwLIcxY4VNGGPL5YZYxnWeTgLYU0PSOEMcO0gfi+Q=
+github.com/gvallee/go_exec v1.1.0/go.mod h1:s9fSkVniLYlQUZb99zuioVH45nkvnYxw6vmGEaZCUPk=
 github.com/gvallee/go_util v1.1.0/go.mod h1:fTexpwdH/n05Ziu0TXJIQsr7E+46QpBxNdeOOsyC0/s=
+github.com/gvallee/go_util v1.5.0 h1:xxAQR2v6csFQdMX18dt9J0DATIUvkBV1zu2VW3yU3wo=
+github.com/gvallee/go_util v1.5.0/go.mod h1:rhmrHriih4is1E3KbQUyn+o8J6wrT6j2pLfAsugaJMY=

internal/pkg/openmpi/openmpi.go

@@ -72,8 +72,13 @@ func DetectFromDir(dir string) (string, string, error) {
 	versionCmd.Env = []string{"LD_LIBRARY_PATH=" + newLDPath, "PATH=" + newPath}
 	res := versionCmd.Run()
 	if res.Err != nil {
-		log.Printf("unable to run ompi_info: %s; stdout: %s; stderr: %s", res.Err, res.Stdout, res.Stderr)
-		return "", "", res.Err
+		// If it fails we try with OPAL_PREFIX set
+		versionCmd.Env = append(versionCmd.Env, "OPAL_PREFIX="+dir)
+		res = versionCmd.Run()
+		if res.Err != nil {
+			log.Printf("unable to run ompi_info: %s; stdout: %s; stderr: %s", res.Err, res.Stdout, res.Stderr)
+			return "", "", res.Err
+		}
 	}
 	version, err := parseOmpiInfoOutputForVersion(res.Stdout)
 	if err != nil {

pkg/jm/jm.go

@@ -168,10 +168,8 @@ func getBatchScriptPath(j *job.Job, sysCfg *sys.Config, batchScriptFilenamePrefi
 
 // TempFile creates a temporary file that is used to store a batch script
 func TempFile(j *job.Job, sysCfg *sys.Config) error {
-	filePrefix := "sbatch-"
-	if j.Partition != "" {
-		filePrefix += j.Partition + "-"
-	}
+	j.SetTimestamp()
+	filePrefix := "sbatch-" + j.ExecutionTimestamp + "-"
 	filePrefix += j.Name
 	var err error
 	j.BatchScript, err = getBatchScriptPath(j, sysCfg, filePrefix)

pkg/jm/jobmgr_slurm.go

@@ -14,7 +14,6 @@ import (
 	"path/filepath"
 	"strconv"
 	"strings"
-	"time"
 
 	"github.com/gvallee/go_exec/pkg/advexec"
 	"github.com/gvallee/go_hpc_jobmgr/internal/pkg/network"
@@ -216,16 +215,19 @@ func generateBatchScriptContent(j *job.Job, sysCfg *sys.Config) (string, error)
 		scriptText += slurm.ScriptCmdPrefix + " -N " + strconv.Itoa(j.NNodes) + "\n"
 	}
 
-	scriptText += slurm.ScriptCmdPrefix + " -t 0:30:0\n"
+	if j.MaxExecTime == "" {
+		scriptText += slurm.ScriptCmdPrefix + " -t 0:30:0\n"
+	} else {
+		scriptText += slurm.ScriptCmdPrefix + " -t " + j.MaxExecTime + "\n"
+	}
 
 	/*
 		if j.NP > 0 {
 			scriptText += slurm.ScriptCmdPrefix + " --ntasks=" + strconv.Itoa(j.NP) + "\n"
 		}
 	*/
 
-	now := time.Now()
-	j.ExecutionTimestamp = string(now.Format("060102150405"))
+	j.SetTimestamp()
 	scriptText += slurm.ScriptCmdPrefix + " --error=" + getJobErrorFilePath(j, sysCfg) + "\n"
 	scriptText += slurm.ScriptCmdPrefix + " --output=" + getJobOutputFilePath(j, sysCfg) + "\n"
 	scriptText += "\n"
@@ -322,8 +324,10 @@ func generateJobScript(j *job.Job, sysCfg *sys.Config) error {
 		return fmt.Errorf("undefined scratch directory")
 	}
 
-	if j.App.BinPath == "" {
-		return fmt.Errorf("application binary is undefined")
+	// If we know nothing about the app and there is no batch script to use, we do
+	// not know how to launch the application
+	if j.App.BinPath == "" && j.BatchScript == "" {
+		return fmt.Errorf("application binary and batch script are undefined")
 	}
 
 	// Create the batch script if the user did not specify a batch script to use.

pkg/job/job.go

@@ -12,6 +12,7 @@ import (
 	"github.com/gvallee/go_hpc_jobmgr/pkg/app"
 	"github.com/gvallee/go_hpc_jobmgr/pkg/mpi"
 	"github.com/gvallee/go_hpc_jobmgr/pkg/sys"
+	"github.com/gvallee/go_util/pkg/timestamp"
 )
 
 // CleanUpFn is a "function pointer" to call to clean up the system after the completion of a job
@@ -81,6 +82,8 @@ type Job struct {
 	CustomEnv map[string]string
 
 	ExecutionTimestamp string
+
+	MaxExecTime string
 }
 
 // GetOutput is the function to call to gather the output (stdout) of the application after execution of the job
@@ -102,3 +105,9 @@ func (j *Job) SetOutputFn(fn GetOutputFn) {
 func (j *Job) SetErrorFn(fn GetErrorFn) {
 	j.internalGetError = fn
 }
+
+func (j *Job) SetTimestamp() {
+	if j.ExecutionTimestamp == "" {
+		j.ExecutionTimestamp = timestamp.Now()
+	}
+}