diff --git a/src/objective-c/GRPCClient/GRPCCall.m b/src/objective-c/GRPCClient/GRPCCall.m
index 0eb1065..c34df99 100644
--- a/src/objective-c/GRPCClient/GRPCCall.m
+++ b/src/objective-c/GRPCClient/GRPCCall.m
@@ -378,6 +378,7 @@
       [strongSelf finishWithError:[NSError errorWithDomain:kGRPCErrorDomain
                                                       code:GRPCErrorCodeUnavailable
                                                   userInfo:@{NSLocalizedDescriptionKey: @"Connectivity lost."}]];
+      [[GRPCHost hostWithAddress:strongSelf->_host] discardChannel:channel];
     }
   }];
 }
diff --git a/src/objective-c/GRPCClient/private/GRPCWrappedCall.m b/src/objective-c/GRPCClient/private/GRPCWrappedCall.m
index 16e5bff..160344a 100644
--- a/src/objective-c/GRPCClient/private/GRPCWrappedCall.m
+++ b/src/objective-c/GRPCClient/private/GRPCWrappedCall.m
@@ -250,7 +250,7 @@
 
     // Each completion queue consumes one thread. There's a trade to be made between creating and
     // consuming too many threads and having contention of multiple calls in a single completion
-    // queue. Currently we favor latency and use one per call.
+    // queue. Currently we use a singleton queue.
     _queue = [GRPCCompletionQueue completionQueue];
 
     _call = [[GRPCHost hostWithAddress:host] unmanagedCallWithPath:path completionQueue:_queue];