fix: increase MaxMessageSize to 16 MiB (#24599)

enterprise/tailnet/pgcoord.go

@@ -739,9 +739,12 @@ func (m *mapper) run() {
 			m.logger.Debug(m.ctx, "skipping nil node update")
 			continue
 		}
-		if err := m.c.Enqueue(update); err != nil {
-			// lots of reasons this could happen, most usually, the peer has disconnected.
-			m.logger.Debug(m.ctx, "failed to enqueue node update", slog.Error(err))
+		for _, chunk := range update.Chunked() {
+			if err := m.c.Enqueue(chunk); err != nil {
+				// lots of reasons this could happen, most usually, the peer has disconnected.
+				m.logger.Debug(m.ctx, "failed to enqueue chunk", slog.Error(err))
+				break
+			}
 		}
 	}
 }

tailnet/peer.go

@@ -70,14 +70,16 @@ func (p *peer) batchUpdateMappingLocked(others []*peer, k proto.CoordinateRespon
 	if len(req.PeerUpdates) == 0 {
 		return nil
 	}
-	select {
-	case p.resps <- req:
-		p.lastWrite = time.Now()
-		p.logger.Debug(context.Background(), "wrote batched update", slog.F("num_peer_updates", len(req.PeerUpdates)))
-		return nil
-	default:
-		return ErrWouldBlock
+	for _, chunk := range req.Chunked() {
+		select {
+		case p.resps <- chunk:
+			p.lastWrite = time.Now()
+		default:
+			return ErrWouldBlock
+		}
 	}
+	p.logger.Debug(context.Background(), "wrote batched update", slog.F("num_peer_updates", len(req.PeerUpdates)))
+	return nil
 }
 
 var errNoResp = xerrors.New("no response needed")

tailnet/proto/response.go

@@ -0,0 +1,22 @@
+package proto
+
+// maxPeerUpdatesPerMessage is the maximum number of peer updates that
+// can be sent in a single CoordinateResponse to stay under DRPC
+// message size limits.
+const maxPeerUpdatesPerMessage = 1024
+
+// Chunked splits the response into multiple responses, each containing
+// at most maxPeerUpdatesPerMessage peer updates to stay under the DRPC
+// 4 MiB transport limit.
+func (r *CoordinateResponse) Chunked() []*CoordinateResponse {
+	updates := r.GetPeerUpdates()
+	if len(updates) <= maxPeerUpdatesPerMessage {
+		return []*CoordinateResponse{r}
+	}
+	var chunks []*CoordinateResponse
+	for i := 0; i < len(updates); i += maxPeerUpdatesPerMessage {
+		end := min(i+maxPeerUpdatesPerMessage, len(updates))
+		chunks = append(chunks, &CoordinateResponse{PeerUpdates: updates[i:end]})
+	}
+	return chunks
+}

tailnet/proto/response_test.go

@@ -0,0 +1,55 @@
+package proto_test
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/coder/coder/v2/tailnet/proto"
+)
+
+func TestCoordinateResponse_Chunked(t *testing.T) {
+	t.Parallel()
+
+	t.Run("NoChunkingNeeded", func(t *testing.T) {
+		t.Parallel()
+		resp := &proto.CoordinateResponse{
+			PeerUpdates: make([]*proto.CoordinateResponse_PeerUpdate, 100),
+		}
+		chunks := resp.Chunked()
+		require.Len(t, chunks, 1)
+		require.Equal(t, resp, chunks[0])
+	})
+
+	t.Run("ExactLimit", func(t *testing.T) {
+		t.Parallel()
+		resp := &proto.CoordinateResponse{
+			PeerUpdates: make([]*proto.CoordinateResponse_PeerUpdate, 1024),
+		}
+		chunks := resp.Chunked()
+		require.Len(t, chunks, 1)
+		require.Equal(t, resp, chunks[0])
+	})
+
+	t.Run("MultipleChunks", func(t *testing.T) {
+		t.Parallel()
+		n := 1024*3 + 500
+		resp := &proto.CoordinateResponse{
+			PeerUpdates: make([]*proto.CoordinateResponse_PeerUpdate, n),
+		}
+		chunks := resp.Chunked()
+		require.Len(t, chunks, 4)
+		total := 0
+		for _, c := range chunks {
+			total += len(c.GetPeerUpdates())
+		}
+		require.Equal(t, n, total)
+	})
+
+	t.Run("EmptyResponse", func(t *testing.T) {
+		t.Parallel()
+		resp := &proto.CoordinateResponse{}
+		chunks := resp.Chunked()
+		require.Len(t, chunks, 1)
+	})
+}