[ruby] Make `GRPC::DeadlineExceeded` report properly (#33565)

In grpc v1.46.2 and later versions, #29155 caused the Ruby client to return `GRPC::Core::CallError`, a non-standard error, with a message: `grpc_call_start_batch failed with outstanding read or write present (code=8)`. However, the actual error should have been `GRPC::DeadlineExceeded`. This occurred because when `GRPC::DeadlineExceeded` is raised, the `@metadata_received` flag doesn't get flipped. When `receive_and_check_status` runs to determine
the error, the `RECV_INITIAL_METADATA` is erroneously sent again.

To avoid this, flip the flag in an `ensure` block whenever the `RECV_INITIAL_METADATA` op is set.

Closes #33283

Closes #33565

COPYBARA_INTEGRATE_REVIEW=https://github.com/grpc/grpc/pull/33565 from stanhu:sh-ruby-fix-issue-33283 850889558c
PiperOrigin-RevId: 614088694

src/ruby/lib/grpc/generic/active_call.rb

@@ -169,10 +169,13 @@ module GRPC
       batch_result = @call.run_batch(ops)
       unless @metadata_received
         @call.metadata = batch_result.metadata
-        @metadata_received = true
       end
       set_input_stream_done
       attach_status_results_and_complete_call(batch_result)
+    ensure
+      # Ensure we don't attempt to request the initial metadata again
+      # in case an exception occurs.
+      @metadata_received = true
     end
 
     def attach_status_results_and_complete_call(recv_status_batch_result)
@@ -258,12 +261,15 @@ module GRPC
       batch_result = @call.run_batch(ops)
       unless @metadata_received
         @call.metadata = batch_result.metadata
-        @metadata_received = true
       end
       get_message_from_batch_result(batch_result)
     rescue GRPC::Core::CallError => e
       GRPC.logger.info("remote_read: #{e}")
       nil
+    ensure
+      # Ensure we don't attempt to request the initial metadata again
+      # in case an exception occurs.
+      @metadata_received = true
     end
 
     def get_message_from_batch_result(recv_message_batch_result)

src/ruby/spec/generic/rpc_server_spec.rb

@@ -61,6 +61,7 @@ FailingStub = FailingService.rpc_stub_class
 class SlowService
   include GRPC::GenericService
   rpc :an_rpc, EchoMsg, EchoMsg
+  rpc :a_server_streaming_rpc, EchoMsg, stream(EchoMsg)
   attr_reader :received_md, :delay
 
   def initialize(_default_var = 'ignored')
@@ -74,6 +75,13 @@ class SlowService
     @received_md << call.metadata unless call.metadata.nil?
     req  # send back the req as the response
   end
+
+  def a_server_streaming_rpc(_, call)
+    GRPC.logger.info("starting a slow #{@delay} server streaming rpc")
+    sleep @delay
+    @received_md << call.metadata unless call.metadata.nil?
+    [EchoMsg.new, EchoMsg.new]
+  end
 end
 
 SlowStub = SlowService.rpc_stub_class
@@ -410,6 +418,23 @@ describe GRPC::RpcServer do
         t.join
       end
 
+      it 'should raise DeadlineExceeded', server: true do
+        service = SlowService.new
+        @srv.handle(service)
+        t = Thread.new { @srv.run }
+        @srv.wait_till_running
+        req = EchoMsg.new
+        stub = SlowStub.new(@host, :this_channel_is_insecure, **client_opts)
+        timeout = service.delay - 0.1
+        deadline = GRPC::Core::TimeConsts.from_relative_time(timeout)
+        responses = stub.a_server_streaming_rpc(req,
+                                                deadline: deadline,
+                                                metadata: { k1: 'v1', k2: 'v2' })
+        expect { responses.to_a }.to raise_error(GRPC::DeadlineExceeded)
+        @srv.stop
+        t.join
+      end
+
       it 'should handle cancellation correctly', server: true do
         request_received = false
         request_received_mu = Mutex.new