X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/876b9e64d1364770486552060222f5f6b1b5e2ea..83864f0f77a37ef8212fd4c3eca268ae9bad4bbb:/lib/crunchrun/singularity.go

diff --git a/lib/crunchrun/singularity.go b/lib/crunchrun/singularity.go
index 5af023a83d..942de4300e 100644
--- a/lib/crunchrun/singularity.go
+++ b/lib/crunchrun/singularity.go
@@ -10,6 +10,7 @@ import (
 	"os"
 	"os/exec"
 	"sort"
+	"strings"
 	"syscall"
 	"time"
 
@@ -241,11 +242,16 @@ func (e *singularityExecutor) Create(spec containerSpec) error {
 	return nil
 }
 
-func (e *singularityExecutor) Start() error {
-	args := []string{"singularity", "exec", "--containall", "--cleanenv", "--pwd", e.spec.WorkingDir}
+func (e *singularityExecutor) execCmd(path string) *exec.Cmd {
+	args := []string{path, "exec", "--containall", "--cleanenv", "--pwd", e.spec.WorkingDir}
 	if !e.spec.EnableNetwork {
 		args = append(args, "--net", "--network=none")
 	}
+
+	if e.spec.CUDADeviceCount != 0 {
+		args = append(args, "--nv")
+	}
+
 	readonlyflag := map[bool]string{
 		false: "rw",
 		true:  "ro",
@@ -279,14 +285,19 @@ func (e *singularityExecutor) Start() error {
 		env = append(env, "SINGULARITYENV_"+k+"="+v)
 	}
 
+	// Singularity always makes all nvidia devices visible to the
+	// container.  If a resource manager such as slurm or LSF told
+	// us to select specific devices we need to propagate that.
+	for _, s := range os.Environ() {
+		if strings.HasPrefix(s, "CUDA_VISIBLE_DEVICES=") {
+			env = append(env, "SINGULARITYENV_"+s)
+		}
+	}
+
 	args = append(args, e.imageFilename)
 	args = append(args, e.spec.Command...)
 
-	path, err := exec.LookPath(args[0])
-	if err != nil {
-		return err
-	}
-	child := &exec.Cmd{
+	return &exec.Cmd{
 		Path:   path,
 		Args:   args,
 		Env:    env,
@@ -294,6 +305,14 @@ func (e *singularityExecutor) Start() error {
 		Stdout: e.spec.Stdout,
 		Stderr: e.spec.Stderr,
 	}
+}
+
+func (e *singularityExecutor) Start() error {
+	path, err := exec.LookPath("singularity")
+	if err != nil {
+		return err
+	}
+	child := e.execCmd(path)
 	err = child.Start()
 	if err != nil {
 		return err