Addition of integer node identifier as part of tuple information

This mechanism permits to identify uniquely a tuple fetched from
Datanode on Coordinator by adding to the tuple information a field
called xc_node_id which is used as a unique node integer identifier.
This node ID is calculated at node creation and added in tuple information
as a kind of extension of ctid. It can be used in internal XC mechanisms.

This commit also includes a fix for round robin tables using remote update
and delete plans. This was an old issue that remained in the test xc_remote but
there was no way until now to identify the node origin of a tuple with a method
which is tuple-based.

Original feature is from Abbas.
The fix regarding round robin tables is from me. I also did some cleanup and
documentation addition on the new column of pgxc_node.

1 files changed, 9 insertions, 0 deletions

diff --git a/src/backend/executor/nodeModifyTable.c b/src/backend/executor/nodeModifyTable.c
index 0cf693aec4..002e1d2991 100644
--- a/src/backend/executor/nodeModifyTable.c
+++ b/src/backend/executor/nodeModifyTable.c
@@ -351,6 +351,9 @@ ExecDelete(ItemPointer tupleid,
 		tuple.t_len = HeapTupleHeaderGetDatumLength(oldtuple);
 		ItemPointerSetInvalid(&(tuple.t_self));
 		tuple.t_tableOid = InvalidOid;
+#ifdef PGXC
+		tuple.t_xc_node_id = 0;
+#endif
 
 		dodelete = ExecIRDeleteTriggers(estate, resultRelInfo, &tuple);
 
@@ -464,6 +467,9 @@ ldelete:;
 			deltuple.t_len = HeapTupleHeaderGetDatumLength(oldtuple);
 			ItemPointerSetInvalid(&(deltuple.t_self));
 			deltuple.t_tableOid = InvalidOid;
+#ifdef PGXC
+			deltuple.t_xc_node_id = 0;
+#endif
 			delbuffer = InvalidBuffer;
 		}
 		else
@@ -575,6 +581,9 @@ ExecUpdate(ItemPointer tupleid,
 		oldtup.t_len = HeapTupleHeaderGetDatumLength(oldtuple);
 		ItemPointerSetInvalid(&(oldtup.t_self));
 		oldtup.t_tableOid = InvalidOid;
+#ifdef PGXC
+		oldtup.t_xc_node_id = 0;
+#endif
 
 		slot = ExecIRUpdateTriggers(estate, resultRelInfo,
 									&oldtup, slot);