Merge pull request #1089 from hashicorp/f-validate-driver-on-submit

Validate driver config on job register

api/util_test.go

@@ -21,6 +21,7 @@ func assertWriteMeta(t *testing.T, wm *WriteMeta) {
 
 func testJob() *Job {
 	task := NewTask("task1", "exec").
+		SetConfig("command", "/bin/sleep").
 		Require(&Resources{
 			CPU:      100,
 			MemoryMB: 256,

command/agent/job_endpoint_test.go

@@ -67,6 +67,7 @@ func TestHTTP_PrefixJobsList(t *testing.T) {
 			// Create the job
 			job := mock.Job()
 			job.ID = ids[i]
+			job.TaskGroups[0].Count = 1
 			args := structs.JobRegisterRequest{
 				Job:          job,
 				WriteRequest: structs.WriteRequest{Region: "global"},

command/util_test.go

@@ -40,6 +40,7 @@ func testServer(
 
 func testJob(jobID string) *api.Job {
 	task := api.NewTask("task1", "exec").
+		SetConfig("command", "/bin/sleep").
 		Require(&api.Resources{
 			MemoryMB: 256,
 			DiskMB:   20,

nomad/job_endpoint.go

@@ -6,6 +6,8 @@ import (
 
 	"github.com/armon/go-metrics"
 	"github.com/hashicorp/go-memdb"
+	"github.com/hashicorp/go-multierror"
+	"github.com/hashicorp/nomad/client/driver"
 	"github.com/hashicorp/nomad/nomad/structs"
 	"github.com/hashicorp/nomad/nomad/watch"
 )
@@ -34,6 +36,31 @@ func (j *Job) Register(args *structs.JobRegisterRequest, reply *structs.JobRegis
 		return err
 	}
 
+	// Validate the driver configurations.
+	var driverErrors multierror.Error
+	for _, tg := range args.Job.TaskGroups {
+		for _, task := range tg.Tasks {
+			d, err := driver.NewDriver(
+				task.Driver,
+				driver.NewEmptyDriverContext(),
+			)
+			if err != nil {
+				msg := "failed to create driver for task %q in group %q for validation: %v"
+				driverErrors.Errors = append(driverErrors.Errors, fmt.Errorf(msg, tg.Name, task.Name, err))
+				continue
+			}
+
+			if err := d.Validate(task.Config); err != nil {
+				formatted := fmt.Errorf("group %q -> task %q -> config: %v", tg.Name, task.Name, err)
+				driverErrors.Errors = append(driverErrors.Errors, formatted)
+			}
+		}
+	}
+
+	if len(driverErrors.Errors) != 0 {
+		return driverErrors.ErrorOrNil()
+	}
+
 	if args.Job.Type == structs.JobTypeCore {
 		return fmt.Errorf("job type cannot be core")
 	}

nomad/job_endpoint_test.go

@@ -2,6 +2,7 @@ package nomad
 
 import (
 	"reflect"
+	"strings"
 	"testing"
 	"time"
 
@@ -85,6 +86,35 @@ func TestJobEndpoint_Register(t *testing.T) {
 	}
 }
 
+func TestJobEndpoint_Register_InvalidDriverConfig(t *testing.T) {
+	s1 := testServer(t, func(c *Config) {
+		c.NumSchedulers = 0 // Prevent automatic dequeue
+	})
+	defer s1.Shutdown()
+	codec := rpcClient(t, s1)
+	testutil.WaitForLeader(t, s1.RPC)
+
+	// Create the register request with a job containing an invalid driver
+	// config
+	job := mock.Job()
+	job.TaskGroups[0].Tasks[0].Config["foo"] = 1
+	req := &structs.JobRegisterRequest{
+		Job:          job,
+		WriteRequest: structs.WriteRequest{Region: "global"},
+	}
+
+	// Fetch the response
+	var resp structs.JobRegisterResponse
+	err := msgpackrpc.CallWithCodec(codec, "Job.Register", req, &resp)
+	if err == nil {
+		t.Fatalf("expected a validation error")
+	}
+
+	if !strings.Contains(err.Error(), "-> config:") {
+		t.Fatalf("expected a driver config validation error but got: %v", err)
+	}
+}
+
 func TestJobEndpoint_Register_Existing(t *testing.T) {
 	s1 := testServer(t, func(c *Config) {
 		c.NumSchedulers = 0 // Prevent automatic dequeue

scripts/test.sh

@@ -10,4 +10,4 @@ go build -o $TEMPDIR/nomad || exit 1
 
 # Run the tests
 echo "--> Running tests"
-go list ./... | grep -v '/vendor/' | sudo -E PATH=$TEMPDIR:$PATH xargs -n1 go test -cover -timeout=300s
+go list ./... | grep -v '/vendor/' | sudo -E PATH=$TEMPDIR:$PATH xargs -n1 go test -cover -timeout=360s