Customize error and enable DefaultHTTPErrorHandler

1. Enable `runtime.DefaultHTTPErrorHandler` to convert gRPC status code to HTTP code.
2. Since we use custom error struct, we need to implement `GRPCStatus` to make sure grpc-go can recognize the error code in our struct,
Otherwise, it will always return codes.Unknown(2), that's the reason we always see code 2 and HTTP 500 in the past.

apiserver/cmd/main.go

@@ -4,14 +4,16 @@ import (
 	"context"
 	"flag"
 
+	"google.golang.org/protobuf/encoding/protojson"
+
 	"math"
 	"net"
 	"net/http"
 	"path"
 	"strings"
 
 	assetfs "github.com/elazarl/go-bindata-assetfs"
-	"github.com/grpc-ecosystem/grpc-gateway/runtime"
+	"github.com/grpc-ecosystem/grpc-gateway/v2/runtime"
 
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/reflection"
@@ -82,7 +84,15 @@ func startHttpProxy() {
 	defer cancel()
 
 	// Create gRPC HTTP MUX and register services.
-	runtimeMux := runtime.NewServeMux(runtime.WithMarshalerOption(runtime.MIMEWildcard, &runtime.JSONPb{OrigName: true, EmitDefaults: true}))
+	runtimeMux := runtime.NewServeMux(
+		runtime.WithMarshalerOption(runtime.MIMEWildcard, &runtime.JSONPb{
+			MarshalOptions: protojson.MarshalOptions{
+				UseProtoNames:  true,
+				UseEnumNumbers: true,
+			},
+		}),
+		runtime.WithErrorHandler(runtime.DefaultHTTPErrorHandler),
+	)
 	registerHttpHandlerFromEndpoint(api.RegisterClusterServiceHandlerFromEndpoint, "ClusterService", ctx, runtimeMux)
 	registerHttpHandlerFromEndpoint(api.RegisterComputeTemplateServiceHandlerFromEndpoint, "ComputeTemplateService", ctx, runtimeMux)
 

apiserver/pkg/manager/resource_manager.go

@@ -4,6 +4,8 @@ import (
 	"context"
 	"fmt"
 
+	"k8s.io/apimachinery/pkg/api/errors"
+
 	"github.com/ray-project/kuberay/apiserver/pkg/model"
 
 	"github.com/ray-project/kuberay/apiserver/pkg/util"
@@ -264,6 +266,10 @@ func (r *ResourceManager) DeleteComputeTemplate(ctx context.Context, name string
 func getClusterByName(ctx context.Context, client rayiov1alpha1.RayClusterInterface, name string) (*v1alpha1.RayCluster, error) {
 	cluster, err := client.Get(ctx, name, metav1.GetOptions{})
 	if err != nil {
+		if errors.IsNotFound(err) {
+			return nil, util.NewNotFoundError(err, "Cluster %s not found", name)
+		}
+
 		return nil, util.Wrap(err, "Get Cluster failed")
 	}
 	if managedBy, ok := cluster.Labels[util.KubernetesManagedByLabelKey]; !ok || managedBy != util.ComponentName {
@@ -277,6 +283,10 @@ func getClusterByName(ctx context.Context, client rayiov1alpha1.RayClusterInterf
 func getComputeTemplateByName(ctx context.Context, client clientv1.ConfigMapInterface, name string) (*v1.ConfigMap, error) {
 	runtime, err := client.Get(ctx, name, metav1.GetOptions{})
 	if err != nil {
+		if errors.IsNotFound(err) {
+			return nil, util.NewNotFoundError(err, "Cluster %s not found", name)
+		}
+
 		return nil, util.Wrap(err, "Get compute template failed")
 	}
 

apiserver/pkg/util/error.go

@@ -7,6 +7,7 @@ import (
 	"github.com/golang/glog"
 	"github.com/pkg/errors"
 	"google.golang.org/grpc/codes"
+	"google.golang.org/grpc/status"
 	k8errors "k8s.io/apimachinery/pkg/api/errors"
 	k8metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 )
@@ -215,6 +216,13 @@ func (e *UserError) String() string {
 		e.internalError)
 }
 
+// GRPCStatus implements `GRPCStatus` to make sure `FromError` in grpc-go can honor the code.
+// Otherwise, it will always return codes.Unknown(2).
+// https://github.com/grpc/grpc-go/blob/2c0949c22d46095edc579d9e66edcd025192b98c/status/status.go#L91-L92
+func (e *UserError) GRPCStatus() *status.Status {
+	return status.New(e.externalStatusCode, e.externalMessage)
+}
+
 func (e *UserError) wrapf(format string, args ...interface{}) *UserError {
 	return newUserError(errors.Wrapf(e.internalError, format, args...),
 		e.externalMessage, e.externalStatusCode)