Don't start a coordinated transaction on workers.

Otherwise transaction hooks of Citus kick in and might cause unwanted errors.

src/backend/distributed/executor/insert_select_executor.c

@@ -267,17 +267,11 @@ CoordinatorInsertSelectExecScanInternal(CustomScanState *node)
 			scanState->tuplestorestate =
 				tuplestore_begin_heap(randomAccess, interTransactions, work_mem);
 
-			TransactionProperties xactProperties = {
+			uint64 rowsInserted = ExecuteTaskListIntoTupleStore(ROW_MODIFY_COMMUTATIVE,
-				.errorOnAnyFailure = true,
+																taskList,
-				.useRemoteTransactionBlocks = TRANSACTION_BLOCKS_REQUIRED,
+																tupleDescriptor,
-				.requires2PC = false
+																scanState->tuplestorestate,
-			};
+																hasReturning);
-			int64 rowsInserted = ExecuteTaskListExtended(ROW_MODIFY_COMMUTATIVE, taskList,
-														 tupleDescriptor,
-														 scanState->tuplestorestate,
-														 hasReturning,
-														 MaxAdaptiveExecutorPoolSize,
-														 &xactProperties);
 
 			executorState->es_processed = rowsInserted;
 		}
@@ -981,6 +975,7 @@ RedistributedInsertSelectTaskList(Query *insertSelectQuery,
 
 		Task *modifyTask = CreateBasicTask(jobId, taskIdIndex, MODIFY_TASK,
 										   queryString->data);
+		modifyTask->dependentTaskList = NIL;
 		modifyTask->anchorShardId = shardId;
 		modifyTask->taskPlacementList = insertShardPlacementList;
 		modifyTask->relationShardList = list_make1(relationShard);
@@ -1043,7 +1038,7 @@ IsRedistributablePlan(Plan *selectPlan, bool hasReturning)
 	/*
 	 * Don't use redistribution if only one task. This is to keep the existing
 	 * behaviour for CTEs that the last step is a read_intermediate_result()
-	 * call. It doesn't hurt much in other case too.
+	 * call. It doesn't hurt much in other cases too.
 	 */
 	if (list_length(distSelectTaskList) <= 1)
 	{

src/backend/distributed/executor/intermediate_results.c

@@ -804,7 +804,7 @@ fetch_intermediate_results(PG_FUNCTION_ARGS)
 	char *remoteHost = text_to_cstring(remoteHostText);
 	int remotePort = PG_GETARG_INT32(2);
 
-	int connectionFlags = 0;
+	int connectionFlags = FORCE_NEW_CONNECTION;
 	int resultIndex = 0;
 	int64 totalBytesWritten = 0L;
 
@@ -827,7 +827,7 @@ fetch_intermediate_results(PG_FUNCTION_ARGS)
 	 * Intermediate results will be stored in a directory that is derived
 	 * from the distributed transaction ID.
 	 */
-	UseCoordinatedTransaction();
+	EnsureDistributedTransactionId();
 
 	MultiConnection *connection = GetNodeConnection(connectionFlags, remoteHost,
 													remotePort);
@@ -838,7 +838,8 @@ fetch_intermediate_results(PG_FUNCTION_ARGS)
 							   remoteHost, remotePort)));
 	}
 
-	RemoteTransactionBeginIfNecessary(connection);
+	StringInfo beginAndSetXactId = BeginAndSetDistributedTransactionIdCommand();
+	ExecuteCriticalRemoteCommand(connection, beginAndSetXactId->data);
 
 	for (resultIndex = 0; resultIndex < resultCount; resultIndex++)
 	{
@@ -847,7 +848,9 @@ fetch_intermediate_results(PG_FUNCTION_ARGS)
 		totalBytesWritten += FetchRemoteIntermediateResult(connection, resultId);
 	}
 
-	UnclaimConnection(connection);
+	ExecuteCriticalRemoteCommand(connection, "END");
+
+	CloseConnection(connection);
 
 	PG_RETURN_INT64(totalBytesWritten);
 }

src/backend/distributed/executor/partitioned_intermediate_results.c

@@ -149,7 +149,7 @@ worker_partition_query_result(PG_FUNCTION_ARGS)
 	 * Intermediate results will be stored in a directory that is derived
 	 * from the distributed transaction ID.
 	 */
-	UseCoordinatedTransaction();
+	EnsureDistributedTransactionId();
 
 	CreateIntermediateResultsDirectory();
 

src/backend/distributed/transaction/remote_transaction.c

@@ -58,7 +58,6 @@ void
 StartRemoteTransactionBegin(struct MultiConnection *connection)
 {
 	RemoteTransaction *transaction = &connection->remoteTransaction;
-	StringInfo beginAndSetDistributedTransactionId = makeStringInfo();
 	ListCell *subIdCell = NULL;
 
 	Assert(transaction->transactionState == REMOTE_TRANS_NOT_STARTED);
@@ -68,28 +67,8 @@ StartRemoteTransactionBegin(struct MultiConnection *connection)
 
 	transaction->transactionState = REMOTE_TRANS_STARTING;
 
-	/*
+	StringInfo beginAndSetDistributedTransactionId =
-	 * Explicitly specify READ COMMITTED, the default on the remote
+		BeginAndSetDistributedTransactionIdCommand();
-	 * side might have been changed, and that would cause problematic
-	 * behaviour.
-	 */
-	appendStringInfoString(beginAndSetDistributedTransactionId,
-						   "BEGIN TRANSACTION ISOLATION LEVEL READ COMMITTED;");
-
-	/*
-	 * Append BEGIN and assign_distributed_transaction_id() statements into a single command
-	 * and send both in one step. The reason is purely performance, we don't want
-	 * seperate roundtrips for these two statements.
-	 */
-	DistributedTransactionId *distributedTransactionId =
-		GetCurrentDistributedTransactionId();
-	const char *timestamp = timestamptz_to_str(distributedTransactionId->timestamp);
-	appendStringInfo(beginAndSetDistributedTransactionId,
-					 "SELECT assign_distributed_transaction_id(%d, " UINT64_FORMAT
-					 ", '%s');",
-					 distributedTransactionId->initiatorNodeIdentifier,
-					 distributedTransactionId->transactionNumber,
-					 timestamp);
 
 	/* append context for in-progress SAVEPOINTs for this transaction */
 	List *activeSubXacts = ActiveSubXactContexts();
@@ -129,6 +108,42 @@ StartRemoteTransactionBegin(struct MultiConnection *connection)
 }
 
 
+/*
+ * BeginAndSetDistributedTransactionIdCommand returns a command which starts
+ * a transaction and assigns the current distributed transaction id.
+ */
+StringInfo
+BeginAndSetDistributedTransactionIdCommand(void)
+{
+	StringInfo beginAndSetDistributedTransactionId = makeStringInfo();
+
+	/*
+	 * Explicitly specify READ COMMITTED, the default on the remote
+	 * side might have been changed, and that would cause problematic
+	 * behaviour.
+	 */
+	appendStringInfoString(beginAndSetDistributedTransactionId,
+						   "BEGIN TRANSACTION ISOLATION LEVEL READ COMMITTED;");
+
+	/*
+	 * Append BEGIN and assign_distributed_transaction_id() statements into a single command
+	 * and send both in one step. The reason is purely performance, we don't want
+	 * seperate roundtrips for these two statements.
+	 */
+	DistributedTransactionId *distributedTransactionId =
+		GetCurrentDistributedTransactionId();
+	const char *timestamp = timestamptz_to_str(distributedTransactionId->timestamp);
+	appendStringInfo(beginAndSetDistributedTransactionId,
+					 "SELECT assign_distributed_transaction_id(%d, " UINT64_FORMAT
+					 ", '%s');",
+					 distributedTransactionId->initiatorNodeIdentifier,
+					 distributedTransactionId->transactionNumber,
+					 timestamp);
+
+	return beginAndSetDistributedTransactionId;
+}
+
+
 /*
  * FinishRemoteTransactionBegin finishes the work StartRemoteTransactionBegin
  * initiated. It blocks if necessary (i.e. if PQisBusy() would return true).

src/backend/distributed/transaction/transaction_management.c

@@ -117,8 +117,7 @@ static bool MaybeExecutingUDF(void);
 void
 UseCoordinatedTransaction(void)
 {
-	if (CurrentCoordinatedTransactionState == COORD_TRANS_STARTED ||
+	if (CurrentCoordinatedTransactionState == COORD_TRANS_STARTED)
-		CurrentCoordinatedTransactionState == COORD_TRANS_STARTED_ON_WORKER)
 	{
 		return;
 	}
@@ -130,21 +129,23 @@ UseCoordinatedTransaction(void)
 	}
 
 	CurrentCoordinatedTransactionState = COORD_TRANS_STARTED;
+	AssignDistributedTransactionId();
+}
 
-	/*
+
-	 * This might be part of bigger distributed transaction originating from
+/*
-	 * another node, in which case transaction id has already been assigned
+ * EnsureDistributedTransactionId makes sure that the current transaction
-	 * by a assign_distributed_transaction_id() call.
+ * has a distributed transaction id. It is either assigned by a previous
-	 */
+ * call of assign_distributed_transaction_id(), or by starting a coordinated
+ * transaction.
+ */
+void
+EnsureDistributedTransactionId(void)
+{
 	DistributedTransactionId *transactionId = GetCurrentDistributedTransactionId();
 	if (transactionId->transactionNumber == 0)
 	{
-		CurrentCoordinatedTransactionState = COORD_TRANS_STARTED_ON_WORKER;
+		UseCoordinatedTransaction();
-		AssignDistributedTransactionId();
-	}
-	else
-	{
-		CurrentCoordinatedTransactionState = COORD_TRANS_STARTED;
 	}
 }
 

src/include/distributed/transaction_management.h

@@ -35,13 +35,6 @@ typedef enum CoordinatedTransactionState
 	/* no coordinated transaction in progress, but connections established */
 	COORD_TRANS_IDLE,
 
-	/*
-	 * Coordinated transaction was initiated by coordinator, but the worker also
-	 * needs to start a coordinated transaction to be able to send commands to
-	 * other workers.
-	 */
-	COORD_TRANS_STARTED_ON_WORKER,
-
 	/* coordinated transaction in progress */
 	COORD_TRANS_STARTED,
 
@@ -117,6 +110,7 @@ extern void UseCoordinatedTransaction(void);
 extern bool InCoordinatedTransaction(void);
 extern void CoordinatedTransactionUse2PC(void);
 extern bool IsMultiStatementTransaction(void);
+extern void EnsureDistributedTransactionId(void);
 
 /* initialization function(s) */
 extern void InitializeTransactionManagement(void);
@@ -124,6 +118,7 @@ extern void InitializeTransactionManagement(void);
 /* other functions */
 extern List * ActiveSubXacts(void);
 extern List * ActiveSubXactContexts(void);
+extern StringInfo BeginAndSetDistributedTransactionIdCommand(void);
 
 
 #endif /*  TRANSACTION_MANAGMENT_H */

src/test/regress/expected/insert_select_repartition.out

@@ -269,8 +269,8 @@ SELECT * FROM target_table ORDER BY a;
 (4 rows)
 
 --
--- worker queries have more columns than necessary. ExpandWorkerTargetEntry() might
+-- worker queries can have more columns than necessary. ExpandWorkerTargetEntry()
--- add additional columns to the target list.
+-- might add additional columns to the target list.
 --
 TRUNCATE target_table;
 \set VERBOSITY TERSE
@@ -304,6 +304,107 @@ SELECT * FROM target_table ORDER BY a;
  -1 | {-1}
 (4 rows)
 
-DROP TABLE source_table, target_table;
+--
+-- repartitioned INSERT/SELECT followed by other DML in stame transaction
+--
+-- case 1. followed by DELETE
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -4 | {3}
+ -3 | {}
+ -2 | {4,6}
+ -1 | {1,2,3}
+(4 rows)
+
+DELETE FROM target_table;
+END;
+SELECT * FROM target_table ORDER BY a;
+ a | b
+---------------------------------------------------------------------
+(0 rows)
+
+-- case 2. followed by UPDATE
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -4 | {3}
+ -3 | {}
+ -2 | {4,6}
+ -1 | {1,2,3}
+(4 rows)
+
+UPDATE target_table SET b=array_append(b, a);
+END;
+SELECT * FROM target_table ORDER BY a;
+ a  |     b
+---------------------------------------------------------------------
+ -4 | {3,-4}
+ -3 | {-3}
+ -2 | {4,6,-2}
+ -1 | {1,2,3,-1}
+(4 rows)
+
+-- case 3. followed by multi-row INSERT
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -4 | {3}
+ -3 | {}
+ -2 | {4,6}
+ -1 | {1,2,3}
+(4 rows)
+
+INSERT INTO target_table VALUES (-5, ARRAY[10,11]), (-6, ARRAY[11,12]), (-7, ARRAY[999]);
+END;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -7 | {999}
+ -6 | {11,12}
+ -5 | {10,11}
+ -4 | {3}
+ -3 | {}
+ -2 | {4,6}
+ -1 | {1,2,3}
+(7 rows)
+
+-- case 4. followed by distributed INSERT/SELECT
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -4 | {3}
+ -3 | {}
+ -2 | {4,6}
+ -1 | {1,2,3}
+(4 rows)
+
+INSERT INTO target_table SELECT * FROM target_table;
+END;
+SELECT * FROM target_table ORDER BY a;
+ a  |    b
+---------------------------------------------------------------------
+ -4 | {3}
+ -4 | {3}
+ -3 | {}
+ -3 | {}
+ -2 | {4,6}
+ -2 | {4,6}
+ -1 | {1,2,3}
+ -1 | {1,2,3}
+(8 rows)
+
 SET client_min_messages TO WARNING;
 DROP SCHEMA insert_select_repartition CASCADE;

src/test/regress/sql/insert_select_repartition.sql

@@ -124,8 +124,8 @@ RESET client_min_messages;
 SELECT * FROM target_table ORDER BY a;
 
 --
--- worker queries have more columns than necessary. ExpandWorkerTargetEntry() might
+-- worker queries can have more columns than necessary. ExpandWorkerTargetEntry()
--- add additional columns to the target list.
+-- might add additional columns to the target list.
 --
 TRUNCATE target_table;
 \set VERBOSITY TERSE
@@ -144,7 +144,45 @@ RESET client_min_messages;
 
 SELECT * FROM target_table ORDER BY a;
 
-DROP TABLE source_table, target_table;
+--
+-- repartitioned INSERT/SELECT followed by other DML in stame transaction
+--
+
+-- case 1. followed by DELETE
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+DELETE FROM target_table;
+END;
+SELECT * FROM target_table ORDER BY a;
+
+-- case 2. followed by UPDATE
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+UPDATE target_table SET b=array_append(b, a);
+END;
+SELECT * FROM target_table ORDER BY a;
+
+-- case 3. followed by multi-row INSERT
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+INSERT INTO target_table VALUES (-5, ARRAY[10,11]), (-6, ARRAY[11,12]), (-7, ARRAY[999]);
+END;
+SELECT * FROM target_table ORDER BY a;
+
+-- case 4. followed by distributed INSERT/SELECT
+TRUNCATE target_table;
+BEGIN;
+INSERT INTO target_table SELECT mapped_key, c FROM source_table;
+SELECT * FROM target_table ORDER BY a;
+INSERT INTO target_table SELECT * FROM target_table;
+END;
+SELECT * FROM target_table ORDER BY a;
 
 SET client_min_messages TO WARNING;
 DROP SCHEMA insert_select_repartition CASCADE;