Fix for bug 3540784: Incorrect COPY for partially distributed tables

When a table was distributed on a portion of nodes. COPY process
was not correctly choosing the output node during a COPY TO. This
problem has been put into light by a case where table was replicated
without using the first datanode of the cluster. It happened that
there was the same problem for hash/modulo tables as well.
A test case in xc_copy has been added to cover this problem, and
a small portion managing COPY TO in execRemote.c has been cleaned.

4 files changed, 58 insertions, 25 deletions

diff --git a/src/backend/pgxc/locator/locator.c b/src/backend/pgxc/locator/locator.c
index 1bba86b379..8c44f70308 100644
--- a/src/backend/pgxc/locator/locator.c
+++ b/src/backend/pgxc/locator/locator.c
@@ -149,8 +149,8 @@ GetAnyDataNode(List *relNodes)
 		}
 	}
 
-	/* Nothing found? Return the 1st one */
-	return lappend_int(NULL, 0);
+	/* Nothing found? Return the first one in relation node list */
+	return lappend_int(NULL, linitial_int(relNodes));
 }
 
 /*
diff --git a/src/backend/pgxc/pool/execRemote.c b/src/backend/pgxc/pool/execRemote.c
index ed2e841ad8..73fa752f3b 100644
--- a/src/backend/pgxc/pool/execRemote.c
+++ b/src/backend/pgxc/pool/execRemote.c
@@ -2322,7 +2322,6 @@ DataNodeCopyOut(ExecNodes *exec_nodes, PGXCNodeHandle** copy_connections, FILE*
 {
 	RemoteQueryState *combiner;
 	int 		conn_count = list_length(exec_nodes->nodeList) == 0 ? NumDataNodes : list_length(exec_nodes->nodeList);
-	int 		count = 0;
 	ListCell	*nodeitem;
 	uint64		processed;
 
@@ -2334,32 +2333,33 @@ DataNodeCopyOut(ExecNodes *exec_nodes, PGXCNodeHandle** copy_connections, FILE*
 
 	foreach(nodeitem, exec_nodes->nodeList)
 	{
-		PGXCNodeHandle *handle = copy_connections[count];
-		count++;
+		PGXCNodeHandle *handle = copy_connections[lfirst_int(nodeitem)];
+		int read_status = 0;
+
+		Assert(handle && handle->state == DN_CONNECTION_STATE_COPY_OUT);
 
-		if (handle && handle->state == DN_CONNECTION_STATE_COPY_OUT)
+		/*
+		 * H message has been consumed, continue to manage data row messages.
+		 * Continue to read as long as there is data.
+		 */
+		while (read_status >= 0 && handle->state == DN_CONNECTION_STATE_COPY_OUT)
 		{
-			int read_status = 0;
-			/* H message has been consumed, continue to manage data row messages */
-			while (read_status >= 0 && handle->state == DN_CONNECTION_STATE_COPY_OUT) /* continue to read as long as there is data */
+			if (handle_response(handle,combiner) == RESPONSE_EOF)
 			{
-				if (handle_response(handle,combiner) == RESPONSE_EOF)
-				{
-					/* read some extra-data */
-					read_status = pgxc_node_read_data(handle, true);
-					if (read_status < 0)
-						ereport(ERROR,
-								(errcode(ERRCODE_CONNECTION_FAILURE),
-								 errmsg("unexpected EOF on datanode connection")));
-					else
-						/*
-						 * Set proper connection status - handle_response
-						 * has changed it to DN_CONNECTION_STATE_QUERY
-						 */
-						handle->state = DN_CONNECTION_STATE_COPY_OUT;
-				}
-				/* There is no more data that can be read from connection */
+				/* read some extra-data */
+				read_status = pgxc_node_read_data(handle, true);
+				if (read_status < 0)
+					ereport(ERROR,
+							(errcode(ERRCODE_CONNECTION_FAILURE),
+							 errmsg("unexpected EOF on datanode connection")));
+				else
+					/*
+					 * Set proper connection status - handle_response
+					 * has changed it to DN_CONNECTION_STATE_QUERY
+					 */
+					handle->state = DN_CONNECTION_STATE_COPY_OUT;
 			}
+			/* There is no more data that can be read from connection */
 		}
 	}
 
diff --git a/src/test/regress/input/xc_copy.source b/src/test/regress/input/xc_copy.source
index 2376dd9e36..efce40287b 100644
--- a/src/test/regress/input/xc_copy.source
+++ b/src/test/regress/input/xc_copy.source
@@ -1,3 +1,9 @@
+--
+-- XC_COPY
+--
+
+-- COPY tests for a Postgres-XC cluster
+
 create or replace function pgxc_nodetype() returns varchar as 
 $$
 declare
@@ -85,3 +91,11 @@ drop function deffunc_bytea();
 drop function deffunc_str();
 drop function deffunc_str_i();
 drop function deffunc_nullstring();
+
+-- Tests related to COPY for a Postgres-XC cluster
+-- Create a table not using the first node of cluster
+SELECT create_table_nodes('xc_copy_1(a int, b int)', '{2}'::int[], 'replication', NULL);
+INSERT INTO xc_copy_1 VALUES (1,23),(34,5),(9,11);
+-- Extract its data
+COPY xc_copy_1 TO STDOUT;
+DROP TABLE xc_copy_1;
diff --git a/src/test/regress/output/xc_copy.source b/src/test/regress/output/xc_copy.source
index 828325c06d..ee95542890 100644
--- a/src/test/regress/output/xc_copy.source
+++ b/src/test/regress/output/xc_copy.source
@@ -1,3 +1,7 @@
+--
+-- XC_COPY
+--
+-- COPY tests for a Postgres-XC cluster
 create or replace function pgxc_nodetype() returns varchar as 
 $$
 declare
@@ -81,3 +85,18 @@ drop function deffunc_bytea();
 drop function deffunc_str();
 drop function deffunc_str_i();
 drop function deffunc_nullstring();
+-- Tests related to COPY for a Postgres-XC cluster
+-- Create a table not using the first node of cluster
+SELECT create_table_nodes('xc_copy_1(a int, b int)', '{2}'::int[], 'replication', NULL);
+ create_table_nodes 
+--------------------
+ 
+(1 row)
+
+INSERT INTO xc_copy_1 VALUES (1,23),(34,5),(9,11);
+-- Extract its data
+COPY xc_copy_1 TO STDOUT;
+1	23
+34	5
+9	11
+DROP TABLE xc_copy_1;