Fix leaving connection in a broken state after preliminary cancellation distributed queries

azat · azat · commit 9ac4c8be3f54 · 2025-12-24T21:48:36.000+01:00
The problem is that some places in the code (i.e. RemoteSource::onUpdatePorts()), can call finish() even before sendQueryAsync() finishes sending the query, and so if after it will try to call finish() again (after sendQueryAsync() finishes) it will be no-op. The problem pops up after #92807, since RemoteSource has these (anti-)pattern.
diff --git a/src/QueryPipeline/RemoteQueryExecutor.cpp b/src/QueryPipeline/RemoteQueryExecutor.cpp
@@ -778,9 +778,6 @@ void RemoteQueryExecutor::finish()
 {
     LockAndBlocker guard(was_cancelled_mutex);
 
-    /// To make sure finish is only called once
-    SCOPE_EXIT({ finished = true; });
-
     /** If one of:
       * - nothing started to do;
       * - received all packets before EndOfStream;
@@ -791,6 +788,9 @@ void RemoteQueryExecutor::finish()
     if (!isQueryPending() || hasThrownException())
         return;
 
+    /// To make sure finish is only called once
+    SCOPE_EXIT({ finished = true; });
+
     /** If you have not read all the data yet, but they are no longer needed.
       * This may be due to the fact that the data is sufficient (for example, when using LIMIT).
       */
