fix(GODT-2526): Unlimited memory usage during fetch/search

This bug was introduced when we switch for a queued channel to handle
the client's responses. This had the unfortunate side effect that if the
responses would arrive faster than they can be sent to client they can
pile up. When this is combined with fetch queries that read the message
literal, an unhealthy amount of RAM can be consumed by this.

internal/session/handle.go

@@ -8,25 +8,14 @@ import (
 	"github.com/ProtonMail/gluon/internal/response"
 	"github.com/ProtonMail/gluon/internal/state"
 	"github.com/ProtonMail/gluon/logging"
-	"github.com/ProtonMail/gluon/queue"
 )
 
 func (s *Session) handleOther(
 	ctx context.Context,
 	tag string,
 	cmd command.Payload,
 ) <-chan response.Response {
-	resCh := make(chan response.Response)
-
-	outCh := queue.NewQueuedChannel[response.Response](0, 0)
-
-	go func() {
-		defer outCh.Close()
-
-		for res := range resCh {
-			outCh.Enqueue(res)
-		}
-	}()
+	resCh := make(chan response.Response, 8)
 
 	s.handleWG.Go(func() {
 		logging.DoAnnotated(state.NewStateContext(ctx, s.state), func(ctx context.Context) {
@@ -45,7 +34,7 @@ func (s *Session) handleOther(
 		})
 	})
 
-	return outCh.GetChannel()
+	return resCh
 }
 
 // handleCommand returns a response instance if a command needs to force an exit of the client.

internal/session/session.go

@@ -213,8 +213,16 @@ func (s *Session) serve(ctx context.Context) error {
 				}
 
 			default:
-				for res := range s.handleOther(withStartTime(ctx, time.Now()), res.command.Tag, cmd) {
+				respCh := s.handleOther(withStartTime(ctx, time.Now()), res.command.Tag, cmd)
+				for res := range respCh {
 					if err := res.Send(s); err != nil {
+						go func() {
+							for range respCh {
+								// Consume all invalid input on error that is still being produced by the ongoing
+								// command.
+							}
+						}()
+
 						return fmt.Errorf("failed to send response to client: %w", err)
 					}
 				}