status/Details: unexpected code breaks after upgrading to v1.62.0 and higher

[povsister]

What version of gRPC are you using?

v1.62.0 and higher

What version of Go are you using (go version)?

go version go1.23.1 darwin/arm64

What operating system (Linux, Windows, …) and version?

Not related

What did you do?

After upgrading to v1.62.0, we have encountered serious problem on gRPC status detail extension: Our gRPC clients can not properly retrieve the customized err detail message from gRPC status.

After some investigation, we found it is caused by PR #6919
internal/status/status.go L157-L160
https://github.com/grpc/grpc-go/pull/6919/files#diff-d78e63ce979e5c67b7fa98a36052a2529e4232343046533343d9772c0cd52549R157-R160

The problem is: status.Details always return proto.MessageV1 before, but it now implicitly returns proto.MessageV2 because of changing protoV1.UnmarshalAny to any.UnmarshalNew.
This implicit behavior actually introduces compatibility issue, though gRPC itself doesn't make any guarantees on that.

Here is an example of our code extracting customized err details.
It unexpctly breaks after upgrading to v1.62.0 and higher.

// InternalErrMessage is a protobuf-defined and generated message.
// It is generated using protobuf v1  (aka: "github.com/golang/protobuf/proto")

func GetInternalErrMessage(gst *status.Status) *InternalErrMessage {
    for _, dt := range gst.Details() {
        // details may contain proto.Message or err.
        // For now, we only deal with proto.Message.
        // PROBLEM: this does not work with grpc-go >= v1.62.0.
        // The actual type of "dt" for now is "google.golang.org/protobuf/internal/impl/*messagelfaceWrapper" which
        // only implements proto.MessageV2 interface.
        if pbMsg, ok := dt.(proto.MessageV1); ok {
            // identify msg type and do type assertion
            if protoadapt.MessageV2Of(pbMsg).ProtoReflect().Descriptor().FullName() ==
                protoadapt.MessageV2Of(&InternalErrMessage{}).ProtoReflect().Descriptor().FullName() {
                    return pbMsg.(*InternalErrMessage)
            }
        }
    }
    // some fallback logic
}

What did you expect to see?

Our Code continues to work as expected after upgrading.

What did you see instead?

Code breaks because of implicit changes of return values from gRPC.

[arjan-bal]

I tried casting the output of status.Details() to a protoiface.MessageV1 in a unit test and it worked: https://github.com/grpc/grpc-go/compare/v1.62.x...arjan-bal:test-status-details?expand=1

@povsister can you share the generated .go file OR the protoc, protoc-gen-go versions used to generate the Go code?

[povsister]

@arjan-bal
Thank you for the investigation. The pb.go file in the test you given in that link actually implements both v1 and v2 message interface. That's the reason you can't reproduce this issue.

@povsister can you share the generated .go file OR the protoc, protoc-gen-go versions used to generate the Go code?

I am not sure which version of protoc and protoc-gen-go are used for our code generation, cause there is no comments on it.
But I am sure the generated message only implements proto.MessageV1, not both v1 and v2.

Here is the snippet of the generated code.

package pb

import (
	fmt "fmt"

	proto "github.com/golang/protobuf/proto"

	math "math"

	anypb "google.golang.org/protobuf/types/known/anypb"
)

// Reference imports to suppress errors if they are not otherwise used.
var _ = proto.Marshal
var _ = fmt.Errorf
var _ = math.Inf

// This is a compile-time assertion to ensure that this generated file
// is compatible with the proto package it is being compiled against.
// A compilation error at this line likely means your copy of the
// proto package needs to be updated.
const _ = proto.ProtoPackageIsVersion2 // please upgrade the proto package

type Error struct {
	ErrCode              int32      `protobuf:"varint,1,opt,name=err_code,json=errCode,proto3" json:"err_code,omitempty"`
	ErrMessage           string     `protobuf:"bytes,2,opt,name=err_message,json=errMessage,proto3" json:"err_message,omitempty"`
	ErrDetail            *anypb.Any `protobuf:"bytes,3,opt,name=err_detail,json=errDetail,proto3" json:"err_detail,omitempty"`
	XXX_NoUnkeyedLiteral struct{}   `json:"-"`
	XXX_unrecognized     []byte     `json:"-"`
	XXX_sizecache        int32      `json:"-"`
}

func (m *Error) Reset()         { *m = Error{} }
func (m *Error) String() string { return proto.CompactTextString(m) }
func (*Error) ProtoMessage()    {}
func (*Error) Descriptor() ([]byte, []int) {
	return fileDescriptor_ecode_9897508b96e13b12, []int{0}
}
func (m *Error) XXX_Unmarshal(b []byte) error {
	return xxx_messageInfo_Error.Unmarshal(m, b)
}
func (m *Error) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) {
	return xxx_messageInfo_Error.Marshal(b, m, deterministic)
}
func (dst *Error) XXX_Merge(src proto.Message) {
	xxx_messageInfo_Error.Merge(dst, src)
}
func (m *Error) XXX_Size() int {
	return xxx_messageInfo_Error.Size(m)
}
func (m *Error) XXX_DiscardUnknown() {
	xxx_messageInfo_Error.DiscardUnknown(m)
}

var xxx_messageInfo_Error proto.InternalMessageInfo

func (m *Error) GetErrCode() int32 {
	if m != nil {
		return m.ErrCode
	}
	return 0
}

func (m *Error) GetErrMessage() string {
	if m != nil {
		return m.ErrMessage
	}
	return ""
}

func (m *Error) GetErrDetail() *anypb.Any {
	if m != nil {
		return m.ErrDetail
	}
	return nil
}

func init() {
	proto.RegisterType((*Error)(nil), "err.Error")
}

func init() {
	proto.RegisterFile("status/internal/pb/ecode.proto", fileDescriptor_ecode_9897508b96e13b12)
}

var fileDescriptor_ecode_9897508b96e13b12 = []byte{
	// 206 bytes of a gzipped FileDescriptorProto
	0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0x84, 0x8f, 0x3f, 0x4f, 0xc3, 0x30,
	0x10, 0xc5, 0x15, 0xaa, 0x02, 0x75, 0x37, 0x8b, 0x21, 0x65, 0x21, 0x62, 0xca, 0x00, 0x3e, 0x89,
	0x7e, 0x02, 0xfe, 0x8d, 0x2c, 0x19, 0x59, 0x2a, 0x3b, 0x3e, 0xac, 0x48, 0x8e, 0x2f, 0x7a, 0x31,
	0x82, 0x7c, 0x7b, 0x94, 0x44, 0xcc, 0x1d, 0xef, 0xfd, 0xee, 0xee, 0xa7, 0xa7, 0x4c, 0xec, 0x1c,
	0x2c, 0x26, 0x4a, 0x9c, 0x09, 0x43, 0x4b, 0x3f, 0x16, 0x9e, 0x13, 0x75, 0x29, 0x33, 0x92, 0x8d,
	0x34, 0x38, 0xe2, 0x56, 0x3c, 0x9b, 0x01, 0x92, 0x45, 0x6f, 0x18, 0xb8, 0x3d, 0x04, 0x91, 0x10,
	0x99, 0x96, 0xc8, 0x7d, 0x7f, 0x91, 0x4d, 0xd3, 0xca, 0xef, 0x7f, 0xd5, 0xf6, 0x1d, 0x10, 0xe8,
	0x83, 0xba, 0x66, 0xe0, 0x34, 0x9f, 0x96, 0x45, 0x55, 0xd4, 0xdb, 0xe6, 0x8a, 0x81, 0x57, 0xf1,
	0xac, 0xef, 0xd4, 0x7e, 0x46, 0x3d, 0x8f, 0xa3, 0x0d, 0x5c, 0x5e, 0x54, 0x45, 0xbd, 0x6b, 0x14,
	0x03, 0x1f, 0x6b, 0xa2, 0x8f, 0x6a, 0x9e, 0x4e, 0x9e, 0xb3, 0xed, 0x62, 0xb9, 0xa9, 0x8a, 0x7a,
	0xff, 0x74, 0x63, 0x56, 0xa9, 0xf9, 0x97, 0x9a, 0xe7, 0x34, 0x35, 0x3b, 0x06, 0xde, 0x96, 0xb5,
	0x17, 0xf3, 0xf9, 0x10, 0xe4, 0xb1, 0x95, 0xbe, 0x97, 0x44, 0xe7, 0x5b, 0xb9, 0xcb, 0xe5, 0xd1,
	0xf1, 0x2f, 0x00, 0x00, 0xff, 0xff, 0x4e, 0xef, 0xdc, 0x8d, 0x02, 0x01, 0x00, 0x00,
}

[arjan-bal]

Thanks for the details. I was able to repro the issue after generating a pb.go file with protoc-gen-go v1.3.5.
https://github.com/grpc/grpc-go/compare/v1.62.x...arjan-bal:test-status-details?expand=1

[povsister]

Great! Please keep me updated if you are going to fix that.
I need adapt our code to future designs of gRPC.

[arjan-bal]

@povsister I discussed the issue with other maintainers. The change (#6919) seems to be backward incompatible when using old generated protobuf code. Since we've not gotten other reports about this issue since 1.62.0 was released 6 months ago, we don't want to make a change now.

You should be able to unblock yourself by either regenerating the pb.go files or casting the result of status.Details() into the V2 message format. I'm closing this issue, please feel free to re-open it in case you have more questions.