v11-0003-Support-2PC-txn-pgoutput.patch

From e7b21894ccd62ecb6c447ffb370dc9c2184fd55f Mon Sep 17 00:00:00 2001
From: Ajin Cherian <ajinc@fast.au.fujitsu.com>
Date: Fri, 23 Oct 2020 05:37:43 -0400
Subject: [PATCH v11] Support 2PC txn - pgoutput

This patch adds support in the pgoutput plugin and subscriber for handling
of two-phase commits.

Includes pgoutput changes.

Includes subscriber changes.

Includes two-phase commit test code (streaming and not streaming).
---
 src/backend/access/transam/twophase.c             |  27 ++
 src/backend/replication/logical/proto.c           | 162 +++++++++-
 src/backend/replication/logical/worker.c          | 364 ++++++++++++++++++++-
 src/backend/replication/pgoutput/pgoutput.c       |  79 ++++-
 src/include/access/twophase.h                     |   1 +
 src/include/replication/logicalproto.h            |  39 +++
 src/test/subscription/t/020_twophase.pl           | 163 ++++++++++
 src/test/subscription/t/021_twophase_streaming.pl | 366 ++++++++++++++++++++++
 8 files changed, 1198 insertions(+), 3 deletions(-)
 create mode 100644 src/test/subscription/t/020_twophase.pl
 create mode 100644 src/test/subscription/t/021_twophase_streaming.pl

diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index 7940060..2e0a408 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -548,6 +548,33 @@ MarkAsPrepared(GlobalTransaction gxact, bool lock_held)
 }
 
 /*
+ * LookupGXact
+ *		Check if the prepared transaction with the given GID is	around
+ */
+bool
+LookupGXact(const char *gid)
+{
+	int			i;
+	bool found = false;
+
+	LWLockAcquire(TwoPhaseStateLock, LW_SHARED);
+	for (i = 0; i < TwoPhaseState->numPrepXacts; i++)
+	{
+		GlobalTransaction gxact = TwoPhaseState->prepXacts[i];
+
+		/* Ignore not-yet-valid GIDs */
+		if (gxact->valid && strcmp(gxact->gid, gid) == 0)
+		{
+			found = true;
+			break;
+		}
+
+	}
+	LWLockRelease(TwoPhaseStateLock);
+	return found;
+}
+
+/*
  * LockGXact
  *		Locate the prepared transaction and mark it busy for COMMIT or PREPARE.
  */
diff --git a/src/backend/replication/logical/proto.c b/src/backend/replication/logical/proto.c
index eb19142..691ef0c 100644
--- a/src/backend/replication/logical/proto.c
+++ b/src/backend/replication/logical/proto.c
@@ -78,7 +78,7 @@ logicalrep_write_commit(StringInfo out, ReorderBufferTXN *txn,
 
 	pq_sendbyte(out, 'C');		/* sending COMMIT */
 
-	/* send the flags field (unused for now) */
+	/* send the flags field */
 	pq_sendbyte(out, flags);
 
 	/* send fields */
@@ -99,6 +99,7 @@ logicalrep_read_commit(StringInfo in, LogicalRepCommitData *commit_data)
 	if (flags != 0)
 		elog(ERROR, "unrecognized flags %u in commit message", flags);
 
+
 	/* read fields */
 	commit_data->commit_lsn = pq_getmsgint64(in);
 	commit_data->end_lsn = pq_getmsgint64(in);
@@ -106,6 +107,165 @@ logicalrep_read_commit(StringInfo in, LogicalRepCommitData *commit_data)
 }
 
 /*
+ * Write PREPARE to the output stream.
+ */
+void
+logicalrep_write_prepare(StringInfo out, ReorderBufferTXN *txn,
+						 XLogRecPtr prepare_lsn)
+{
+	uint8		flags = 0;
+
+	pq_sendbyte(out, 'P');		/* sending PREPARE protocol */
+
+	/*
+	 * This should only ever happen for two-phase commit transactions. In which case we
+	 * expect to have a non-empty GID.
+	 */
+	Assert(rbtxn_prepared(txn));
+	Assert(txn->gid != NULL);
+
+	/*
+	 * Flags are determined from the state of the transaction. We know we
+	 * always get PREPARE first and then [COMMIT|ROLLBACK] PREPARED, so if
+	 * it's already marked as committed then it has to be COMMIT PREPARED (and
+	 * likewise for abort / ROLLBACK PREPARED).
+	 */
+	if (rbtxn_commit_prepared(txn))
+		flags = LOGICALREP_IS_COMMIT_PREPARED;
+	else if (rbtxn_rollback_prepared(txn))
+		flags = LOGICALREP_IS_ROLLBACK_PREPARED;
+	else
+		flags = LOGICALREP_IS_PREPARE;
+
+	/* Make sure exactly one of the expected flags is set. */
+	if (!PrepareFlagsAreValid(flags))
+		elog(ERROR, "unrecognized flags %u in prepare message", flags);
+
+	/* send the flags field */
+	pq_sendbyte(out, flags);
+
+	/* send fields */
+	pq_sendint64(out, prepare_lsn);
+	pq_sendint64(out, txn->end_lsn);
+	pq_sendint64(out, txn->commit_time);
+
+	/* send gid */
+	pq_sendstring(out, txn->gid);
+}
+
+/*
+ * Read transaction PREPARE from the stream.
+ */
+void
+logicalrep_read_prepare(StringInfo in, LogicalRepPrepareData * prepare_data)
+{
+	/* read flags */
+	uint8		flags = pq_getmsgbyte(in);
+
+	if (!PrepareFlagsAreValid(flags))
+		elog(ERROR, "unrecognized flags %u in prepare message", flags);
+
+	/* set the action (reuse the constants used for the flags) */
+	prepare_data->prepare_type = flags;
+
+	/* read fields */
+	prepare_data->prepare_lsn = pq_getmsgint64(in);
+	prepare_data->end_lsn = pq_getmsgint64(in);
+	prepare_data->preparetime = pq_getmsgint64(in);
+
+	/* read gid (copy it into a pre-allocated buffer) */
+	strcpy(prepare_data->gid, pq_getmsgstring(in));
+}
+
+/*
+ * Write STREAM PREPARE to the output stream.
+ *
+ * (For stream PREPARE, stream COMMIT PREPARED, stream ROLLBACK PREPARED)
+ *
+ * TODO- This is mostly cut/paste from logicalrep_write_prepare. Consider refactoring for commonality.
+ */
+void
+logicalrep_write_stream_prepare(StringInfo out,
+								ReorderBufferTXN *txn,
+								XLogRecPtr prepare_lsn)
+{
+	uint8	flags = 0;
+
+#ifdef DEBUG_STREAM_2PC
+	elog(LOG, "proto: logicalrep_write_stream_prepare");
+#endif
+
+	pq_sendbyte(out, 'p');		/* sending STREAM PREPARE protocol */
+
+	/*
+	 * This should only ever happen for two-phase transactions. In which case we
+	 * expect to have a non-empty GID.
+	 */
+	Assert(rbtxn_prepared(txn));
+	Assert(txn->gid != NULL);
+
+	/*
+	 * For streaming APIs only PREPARE is supported. [COMMIT|ROLLBACK] PREPARED
+	 * uses non-streaming APIs
+	 */
+	flags = LOGICALREP_IS_PREPARE;
+
+	/* transaction ID */
+	Assert(TransactionIdIsValid(txn->xid));
+	pq_sendint32(out, txn->xid);
+
+	/* send the flags field */
+	pq_sendbyte(out, flags);
+
+	/* send fields */
+	pq_sendint64(out, prepare_lsn);
+	pq_sendint64(out, txn->end_lsn);
+	pq_sendint64(out, txn->commit_time);
+
+	/* send gid */
+	pq_sendstring(out, txn->gid);
+}
+
+/*
+ * Read STREAM PREPARE from the output stream.
+ *
+ * (For stream PREPARE, stream COMMIT PREPARED, stream ROLLBACK PREPARED)
+ *
+ * TODO - This is mostly cut/paste from logicalrep_read_prepare. Consider refactoring for commonality.
+ */
+TransactionId
+logicalrep_read_stream_prepare(StringInfo in, LogicalRepPrepareData *prepare_data)
+{
+	TransactionId	xid;
+	uint8			flags;
+
+#ifdef DEBUG_STREAM_2PC
+	elog(LOG, "proto: logicalrep_read_stream_prepare");
+#endif
+
+	xid = pq_getmsgint(in, 4);
+
+	/* read flags */
+	flags = pq_getmsgbyte(in);
+
+	if (!PrepareFlagsAreValid(flags))
+		elog(ERROR, "unrecognized flags %u in prepare message", flags);
+
+	/* set the action (reuse the constants used for the flags) */
+	prepare_data->prepare_type = flags;
+
+	/* read fields */
+	prepare_data->prepare_lsn = pq_getmsgint64(in);
+	prepare_data->end_lsn = pq_getmsgint64(in);
+	prepare_data->preparetime = pq_getmsgint64(in);
+
+	/* read gid (copy it into a pre-allocated buffer) */
+	strcpy(prepare_data->gid, pq_getmsgstring(in));
+
+	return xid;
+}
+
+/*
  * Write ORIGIN to the output stream.
  */
 void
diff --git a/src/backend/replication/logical/worker.c b/src/backend/replication/logical/worker.c
index 3a5b733..26ad790 100644
--- a/src/backend/replication/logical/worker.c
+++ b/src/backend/replication/logical/worker.c
@@ -722,6 +722,7 @@ apply_handle_commit(StringInfo s)
 		replorigin_session_origin_timestamp = commit_data.committime;
 
 		CommitTransactionCommand();
+
 		pgstat_report_stat(false);
 
 		store_flush_position(commit_data.end_lsn);
@@ -742,6 +743,359 @@ apply_handle_commit(StringInfo s)
 }
 
 /*
+ * Called from apply_handle_prepare to handle a PREPARE TRANSACTION.
+ */
+static void
+apply_handle_prepare_txn(LogicalRepPrepareData * prepare_data)
+{
+	Assert(prepare_data->prepare_lsn == remote_final_lsn);
+
+	/* The synchronization worker runs in single transaction. */
+	if (IsTransactionState() && !am_tablesync_worker())
+	{
+		/* End the earlier transaction and start a new one */
+		BeginTransactionBlock();
+		CommitTransactionCommand();
+		StartTransactionCommand();
+
+		/*
+		 * Update origin state so we can restart streaming from correct
+		 * position in case of crash.
+		 */
+		replorigin_session_origin_lsn = prepare_data->end_lsn;
+		replorigin_session_origin_timestamp = prepare_data->preparetime;
+
+		PrepareTransactionBlock(prepare_data->gid);
+		CommitTransactionCommand();
+		pgstat_report_stat(false);
+
+		store_flush_position(prepare_data->end_lsn);
+	}
+	else
+	{
+		/* Process any invalidation messages that might have accumulated. */
+		AcceptInvalidationMessages();
+		maybe_reread_subscription();
+	}
+
+	in_remote_transaction = false;
+
+	/* Process any tables that are being synchronized in parallel. */
+	process_syncing_tables(prepare_data->end_lsn);
+
+	pgstat_report_activity(STATE_IDLE, NULL);
+}
+
+/*
+ * Called from apply_handle_prepare to handle a COMMIT PREPARED of a previously
+ * PREPARED transaction.
+ */
+static void
+apply_handle_commit_prepared_txn(LogicalRepPrepareData * prepare_data)
+{
+	/* there is no transaction when COMMIT PREPARED is called */
+	ensure_transaction();
+
+	/*
+	 * Update origin state so we can restart streaming from correct position
+	 * in case of crash.
+	 */
+	replorigin_session_origin_lsn = prepare_data->end_lsn;
+	replorigin_session_origin_timestamp = prepare_data->preparetime;
+
+	FinishPreparedTransaction(prepare_data->gid, true);
+	CommitTransactionCommand();
+	pgstat_report_stat(false);
+
+	store_flush_position(prepare_data->end_lsn);
+	in_remote_transaction = false;
+
+	/* Process any tables that are being synchronized in parallel. */
+	process_syncing_tables(prepare_data->end_lsn);
+
+	pgstat_report_activity(STATE_IDLE, NULL);
+}
+
+/*
+ * Called from apply_handle_prepare to handle a ROLLBACK PREPARED of a previously
+ * PREPARED TRANSACTION.
+ */
+static void
+apply_handle_rollback_prepared_txn(LogicalRepPrepareData * prepare_data)
+{
+	/*
+	 * Update origin state so we can restart streaming from correct position
+	 * in case of crash.
+	 */
+	replorigin_session_origin_lsn = prepare_data->end_lsn;
+	replorigin_session_origin_timestamp = prepare_data->preparetime;
+
+	/*
+	 * During logical decoding, on the apply side, it's possible that a
+	 * prepared transaction got aborted while decoding. In that case, we stop
+	 * the decoding and abort the transaction immediately. However the
+	 * ROLLBACK prepared processing still reaches the subscriber. In that case
+	 * it's ok to have a missing gid
+	 */
+	if (LookupGXact(prepare_data->gid))
+	{
+		/* there is no transaction when ABORT/ROLLBACK PREPARED is called */
+		ensure_transaction();
+		FinishPreparedTransaction(prepare_data->gid, false);
+		CommitTransactionCommand();
+	}
+
+	pgstat_report_stat(false);
+
+	store_flush_position(prepare_data->end_lsn);
+	in_remote_transaction = false;
+
+	/* Process any tables that are being synchronized in parallel. */
+	process_syncing_tables(prepare_data->end_lsn);
+
+	pgstat_report_activity(STATE_IDLE, NULL);
+}
+
+/*
+ * Handle PREPARE message.
+ */
+static void
+apply_handle_prepare(StringInfo s)
+{
+	LogicalRepPrepareData prepare_data;
+
+	logicalrep_read_prepare(s, &prepare_data);
+
+	switch (prepare_data.prepare_type)
+	{
+		case LOGICALREP_IS_PREPARE:
+			apply_handle_prepare_txn(&prepare_data);
+			break;
+
+		case LOGICALREP_IS_COMMIT_PREPARED:
+			apply_handle_commit_prepared_txn(&prepare_data);
+			break;
+
+		case LOGICALREP_IS_ROLLBACK_PREPARED:
+			apply_handle_rollback_prepared_txn(&prepare_data);
+			break;
+
+		default:
+			ereport(ERROR,
+					(errcode(ERRCODE_PROTOCOL_VIOLATION),
+					 errmsg("unexpected type of prepare message: %d",
+							prepare_data.prepare_type)));
+	}
+}
+
+/*
+ * Handle STREAM PREPARE message.
+ */
+static void
+apply_handle_stream_prepare(StringInfo s)
+
+{
+	StringInfoData s2;
+	int			nchanges;
+	char		path[MAXPGPATH];
+	char	   *buffer = NULL;
+	bool		found;
+	StreamXidHash *ent;
+	MemoryContext oldcxt;
+	BufFile    *fd;
+	LogicalRepPrepareData prepare_data;
+	TransactionId xid;
+
+	xid = logicalrep_read_stream_prepare(s, &prepare_data);
+
+	/* This should be a PREPARE. COMMIT PREPARED and ROLLBACK PREPARED should
+	 * result in non-streaming APIs being called.
+	 */
+	Assert(prepare_data.prepare_type == LOGICALREP_IS_PREPARE);
+
+
+#ifdef DEBUG_STREAM_2PC
+	elog(LOG, "worker: apply_handle_stream_prepare");
+#endif
+
+	/*
+	 * FIXME - Following condition was in apply_handle_prepare_txn except I found  it was ALWAYS IsTransactionState() == false
+	 * The synchronization worker runs in single transaction. *
+	if (IsTransactionState() && !am_tablesync_worker())
+	*/
+	if (!am_tablesync_worker())
+	{
+		/*
+		 * ==================================================================================================
+		 * The following chunk of code is largely cut/paste from the existing apply_handle_prepare_commit_txn
+		 * which was handling the non-two-phase streaming commit by applying the operations of the spooled file.
+		 *
+		 * Differences are:
+		 * - Here the xid is known already because apply_handle_stream_prepare already called
+		 *   locicalrep_read_stream_prepare
+		 *
+		 * TODO - This is possible candidate for refactoring since so much of it is the same.
+		 * ==================================================================================================
+		 */
+
+		Assert(!in_streamed_transaction);
+
+		elog(DEBUG1, "received prepare for streamed transaction %u", xid);
+
+		ensure_transaction();
+
+		/*
+		 * Allocate file handle and memory required to process all the messages in
+		 * TopTransactionContext to avoid them getting reset after each message is
+		 * processed.
+		 */
+		oldcxt = MemoryContextSwitchTo(TopTransactionContext);
+
+		/* open the spool file for the committed transaction */
+		changes_filename(path, MyLogicalRepWorker->subid, xid);
+		elog(DEBUG1, "replaying changes from file \"%s\"", path);
+#ifdef DEBUG_STREAM_2PC
+		elog(LOG, "worker: replaying changes from file \"%s\"", path);
+#endif
+		ent = (StreamXidHash *) hash_search(xidhash,
+											(void *) &xid,
+											HASH_FIND,
+											&found);
+		Assert(found);
+		fd = BufFileOpenShared(ent->stream_fileset, path, O_RDONLY);
+
+		buffer = palloc(BLCKSZ);
+		initStringInfo(&s2);
+
+		MemoryContextSwitchTo(oldcxt);
+
+		remote_final_lsn = prepare_data.prepare_lsn;
+
+		/*
+		 * Make sure the handle apply_dispatch methods are aware we're in a remote
+		 * transaction.
+		 */
+		in_remote_transaction = true;
+		pgstat_report_activity(STATE_RUNNING, NULL);
+
+		/*
+		 * Read the entries one by one and pass them through the same logic as in
+		 * apply_dispatch.
+		 */
+		nchanges = 0;
+		while (true)
+		{
+			int			nbytes;
+			int			len;
+
+			CHECK_FOR_INTERRUPTS();
+
+			/* read length of the on-disk record */
+			nbytes = BufFileRead(fd, &len, sizeof(len));
+
+			/* have we reached end of the file? */
+			if (nbytes == 0)
+				break;
+
+			/* do we have a correct length? */
+			if (nbytes != sizeof(len))
+				ereport(ERROR,
+						(errcode_for_file_access(),
+						 errmsg("could not read from streaming transaction's changes file \"%s\": %m",
+								path)));
+
+			Assert(len > 0);
+
+			/* make sure we have sufficiently large buffer */
+			buffer = repalloc(buffer, len);
+
+			/* and finally read the data into the buffer */
+			if (BufFileRead(fd, buffer, len) != len)
+				ereport(ERROR,
+						(errcode_for_file_access(),
+						 errmsg("could not read from streaming transaction's changes file \"%s\": %m",
+								path)));
+
+			/* copy the buffer to the stringinfo and call apply_dispatch */
+			resetStringInfo(&s2);
+			appendBinaryStringInfo(&s2, buffer, len);
+
+			/* Ensure we are reading the data into our memory context. */
+			oldcxt = MemoryContextSwitchTo(ApplyMessageContext);
+
+			apply_dispatch(&s2);
+
+			MemoryContextReset(ApplyMessageContext);
+
+			MemoryContextSwitchTo(oldcxt);
+
+			nchanges++;
+
+			if (nchanges % 1000 == 0)
+				elog(DEBUG1, "replayed %d changes from file '%s'",
+					 nchanges, path);
+		}
+
+		BufFileClose(fd);
+
+		pfree(buffer);
+		pfree(s2.data);
+
+		/*
+		 * ==================================================================================================
+		 * The following chunk of code is cut/paste from the existing apply_handle_prepare_txn
+		 * which was handling the two-phase prepare of the non-streamed tx
+		 * ==================================================================================================
+		 */
+#ifdef DEBUG_STREAM_2PC
+		elog(LOG, "worker: call PrepareTransactionBlock()");
+#endif
+		BeginTransactionBlock();
+		CommitTransactionCommand();
+		StartTransactionCommand();
+
+		/*
+		 * Update origin state so we can restart streaming from correct position
+		 * in case of crash.
+		 */
+		replorigin_session_origin_lsn = prepare_data.end_lsn;
+		replorigin_session_origin_timestamp = prepare_data.preparetime;
+
+		PrepareTransactionBlock(prepare_data.gid);
+		/* End of copied prepare code */
+
+		CommitTransactionCommand();
+		pgstat_report_stat(false);
+
+		store_flush_position(prepare_data.end_lsn);
+
+		elog(DEBUG1, "replayed %d (all) changes from file \"%s\"", nchanges, path);
+#ifdef DEBUG_STREAM_2PC
+		elog(LOG, "worker: replayed %d (all) changes from file \"%s\"", nchanges, path);
+#endif
+
+		/* ============================================================================================== */
+	}
+	else
+	{
+		/* Process any invalidation messages that might have accumulated. */
+		AcceptInvalidationMessages();
+		maybe_reread_subscription();
+	}
+
+	in_remote_transaction = false;
+
+	/* Process any tables that are being synchronized in parallel. */
+	process_syncing_tables(prepare_data.end_lsn);
+
+	/* unlink the files with serialized changes and subxact info */
+	/* FIXME - OK to do this here (outside of if/else). */
+	stream_cleanup_files(MyLogicalRepWorker->subid, xid);
+
+	pgstat_report_activity(STATE_IDLE, NULL);
+}
+
+/*
  * Handle ORIGIN message.
  *
  * TODO, support tracking of multiple origins
@@ -1905,10 +2259,14 @@ apply_dispatch(StringInfo s)
 		case 'B':
 			apply_handle_begin(s);
 			break;
-			/* COMMIT */
+			/* COMMIT/ABORT */
 		case 'C':
 			apply_handle_commit(s);
 			break;
+			/* PREPARE and [COMMIT|ROLLBACK] PREPARED */
+		case 'P':
+			apply_handle_prepare(s);
+			break;
 			/* INSERT */
 		case 'I':
 			apply_handle_insert(s);
@@ -1953,6 +2311,10 @@ apply_dispatch(StringInfo s)
 		case 'c':
 			apply_handle_stream_commit(s);
 			break;
+			/* STREAM PREPARE */
+		case 'p':
+			apply_handle_stream_prepare(s);
+			break;
 		default:
 			ereport(ERROR,
 					(errcode(ERRCODE_PROTOCOL_VIOLATION),
diff --git a/src/backend/replication/pgoutput/pgoutput.c b/src/backend/replication/pgoutput/pgoutput.c
index 9c997ae..32fe5c5 100644
--- a/src/backend/replication/pgoutput/pgoutput.c
+++ b/src/backend/replication/pgoutput/pgoutput.c
@@ -47,6 +47,12 @@ static void pgoutput_truncate(LogicalDecodingContext *ctx,
 							  ReorderBufferChange *change);
 static bool pgoutput_origin_filter(LogicalDecodingContext *ctx,
 								   RepOriginId origin_id);
+static void pgoutput_prepare_txn(LogicalDecodingContext *ctx,
+								 ReorderBufferTXN *txn, XLogRecPtr prepare_lsn);
+static void pgoutput_commit_prepared_txn(LogicalDecodingContext *ctx,
+										 ReorderBufferTXN *txn, XLogRecPtr prepare_lsn);
+static void pgoutput_rollback_prepared_txn(LogicalDecodingContext *ctx,
+										   ReorderBufferTXN *txn, XLogRecPtr prepare_lsn);
 static void pgoutput_stream_start(struct LogicalDecodingContext *ctx,
 								  ReorderBufferTXN *txn);
 static void pgoutput_stream_stop(struct LogicalDecodingContext *ctx,
@@ -57,7 +63,8 @@ static void pgoutput_stream_abort(struct LogicalDecodingContext *ctx,
 static void pgoutput_stream_commit(struct LogicalDecodingContext *ctx,
 								   ReorderBufferTXN *txn,
 								   XLogRecPtr commit_lsn);
-
+static void pgoutput_stream_prepare_txn(LogicalDecodingContext *ctx,
+										ReorderBufferTXN *txn, XLogRecPtr prepare_lsn);
 static bool publications_valid;
 static bool in_streaming;
 
@@ -143,6 +150,10 @@ _PG_output_plugin_init(OutputPluginCallbacks *cb)
 	cb->change_cb = pgoutput_change;
 	cb->truncate_cb = pgoutput_truncate;
 	cb->commit_cb = pgoutput_commit_txn;
+
+	cb->prepare_cb = pgoutput_prepare_txn;
+	cb->commit_prepared_cb = pgoutput_commit_prepared_txn;
+	cb->rollback_prepared_cb = pgoutput_rollback_prepared_txn;
 	cb->filter_by_origin_cb = pgoutput_origin_filter;
 	cb->shutdown_cb = pgoutput_shutdown;
 
@@ -153,6 +164,8 @@ _PG_output_plugin_init(OutputPluginCallbacks *cb)
 	cb->stream_commit_cb = pgoutput_stream_commit;
 	cb->stream_change_cb = pgoutput_change;
 	cb->stream_truncate_cb = pgoutput_truncate;
+	/* transaction streaming - two-phase commit */
+	cb->stream_prepare_cb = pgoutput_stream_prepare_txn;
 }
 
 static void
@@ -378,6 +391,48 @@ pgoutput_commit_txn(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 }
 
 /*
+ * PREPARE callback
+ */
+static void
+pgoutput_prepare_txn(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
+					 XLogRecPtr prepare_lsn)
+{
+	OutputPluginUpdateProgress(ctx);
+
+	OutputPluginPrepareWrite(ctx, true);
+	logicalrep_write_prepare(ctx->out, txn, prepare_lsn);
+	OutputPluginWrite(ctx, true);
+}
+
+/*
+ * COMMIT PREPARED callback
+ */
+static void
+pgoutput_commit_prepared_txn(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
+							 XLogRecPtr prepare_lsn)
+{
+	OutputPluginUpdateProgress(ctx);
+
+	OutputPluginPrepareWrite(ctx, true);
+	logicalrep_write_prepare(ctx->out, txn, prepare_lsn);
+	OutputPluginWrite(ctx, true);
+}
+
+/*
+ * ROLLBACK PREPARED callback
+ */
+static void
+pgoutput_rollback_prepared_txn(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
+							XLogRecPtr prepare_lsn)
+{
+	OutputPluginUpdateProgress(ctx);
+
+	OutputPluginPrepareWrite(ctx, true);
+	logicalrep_write_prepare(ctx->out, txn, prepare_lsn);
+	OutputPluginWrite(ctx, true);
+}
+
+/*
  * Write the current schema of the relation and its ancestor (if any) if not
  * done yet.
  */
@@ -857,6 +912,28 @@ pgoutput_stream_commit(struct LogicalDecodingContext *ctx,
 }
 
 /*
+ * PREPARE callback (for streaming two-phase commit).
+ *
+ * Notify the downstream to prepare the transaction.
+ */
+static void
+pgoutput_stream_prepare_txn(LogicalDecodingContext *ctx,
+							ReorderBufferTXN *txn,
+							XLogRecPtr prepare_lsn)
+{
+#ifdef DEBUG_STREAM_2PC
+	elog(LOG, "pgoutput: pgoutput_stream_prepare_txn");
+#endif
+
+	Assert(rbtxn_is_streamed(txn));
+
+	OutputPluginUpdateProgress(ctx);
+	OutputPluginPrepareWrite(ctx, true);
+	logicalrep_write_stream_prepare(ctx->out, txn, prepare_lsn);
+	OutputPluginWrite(ctx, true);
+}
+
+/*
  * Initialize the relation schema sync cache for a decoding session.
  *
  * The hash table is destroyed at the end of a decoding session. While
diff --git a/src/include/access/twophase.h b/src/include/access/twophase.h
index 2ca71c3..b2628ea 100644
--- a/src/include/access/twophase.h
+++ b/src/include/access/twophase.h
@@ -44,6 +44,7 @@ extern GlobalTransaction MarkAsPreparing(TransactionId xid, const char *gid,
 extern void StartPrepare(GlobalTransaction gxact);
 extern void EndPrepare(GlobalTransaction gxact);
 extern bool StandbyTransactionIdIsPrepared(TransactionId xid);
+extern bool LookupGXact(const char *gid);
 
 extern TransactionId PrescanPreparedTransactions(TransactionId **xids_p,
 												 int *nxids_p);
diff --git a/src/include/replication/logicalproto.h b/src/include/replication/logicalproto.h
index 0c2cda2..ae59c04 100644
--- a/src/include/replication/logicalproto.h
+++ b/src/include/replication/logicalproto.h
@@ -87,6 +87,7 @@ typedef struct LogicalRepBeginData
 	TransactionId xid;
 } LogicalRepBeginData;
 
+/* Commit (and abort) information */
 typedef struct LogicalRepCommitData
 {
 	XLogRecPtr	commit_lsn;
@@ -94,6 +95,28 @@ typedef struct LogicalRepCommitData
 	TimestampTz committime;
 } LogicalRepCommitData;
 
+
+/* Prepare protocol information */
+typedef struct LogicalRepPrepareData
+{
+	uint8		prepare_type;
+	XLogRecPtr	prepare_lsn;
+	XLogRecPtr	end_lsn;
+	TimestampTz preparetime;
+	char		gid[GIDSIZE];
+} LogicalRepPrepareData;
+
+/* types of the prepare protocol message */
+#define LOGICALREP_IS_PREPARE			0x01
+#define LOGICALREP_IS_COMMIT_PREPARED	0x02
+#define LOGICALREP_IS_ROLLBACK_PREPARED	0x04
+
+/* prepare can be exactly one of PREPARE, [COMMIT|ROLLBACK] PREPARED*/
+#define PrepareFlagsAreValid(flags) \
+	(((flags) == LOGICALREP_IS_PREPARE) || \
+	 ((flags) == LOGICALREP_IS_COMMIT_PREPARED) || \
+	 ((flags) == LOGICALREP_IS_ROLLBACK_PREPARED))
+
 extern void logicalrep_write_begin(StringInfo out, ReorderBufferTXN *txn);
 extern void logicalrep_read_begin(StringInfo in,
 								  LogicalRepBeginData *begin_data);
@@ -101,6 +124,10 @@ extern void logicalrep_write_commit(StringInfo out, ReorderBufferTXN *txn,
 									XLogRecPtr commit_lsn);
 extern void logicalrep_read_commit(StringInfo in,
 								   LogicalRepCommitData *commit_data);
+extern void logicalrep_write_prepare(StringInfo out, ReorderBufferTXN *txn,
+									  XLogRecPtr prepare_lsn);
+extern void logicalrep_read_prepare(StringInfo in,
+									LogicalRepPrepareData * prepare_data);
 extern void logicalrep_write_origin(StringInfo out, const char *origin,
 									XLogRecPtr origin_lsn);
 extern char *logicalrep_read_origin(StringInfo in, XLogRecPtr *origin_lsn);
@@ -144,4 +171,16 @@ extern void logicalrep_write_stream_abort(StringInfo out, TransactionId xid,
 extern void logicalrep_read_stream_abort(StringInfo in, TransactionId *xid,
 										 TransactionId *subxid);
 
+/*
+ * FIXME - Uncomment this to see more lgging for streamed two-phase transactions.
+ *
+ * #define DEBUG_STREAM_2PC
+ */
+
+extern void logicalrep_write_stream_prepare(StringInfo out, ReorderBufferTXN *txn,
+											XLogRecPtr prepare_lsn);
+extern TransactionId logicalrep_read_stream_prepare(StringInfo in,
+													LogicalRepPrepareData *prepare_data);
+
+
 #endif							/* LOGICAL_PROTO_H */
diff --git a/src/test/subscription/t/020_twophase.pl b/src/test/subscription/t/020_twophase.pl
new file mode 100644
index 0000000..3feb2c3
--- /dev/null
+++ b/src/test/subscription/t/020_twophase.pl
@@ -0,0 +1,163 @@
+# logical replication of 2PC test
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More tests => 12;
+
+# Initialize publisher node
+my $node_publisher = get_new_node('publisher');
+$node_publisher->init(allows_streaming => 'logical');
+$node_publisher->append_conf(
+        'postgresql.conf', qq(
+        max_prepared_transactions = 10
+        ));
+$node_publisher->start;
+
+# Create subscriber node
+my $node_subscriber = get_new_node('subscriber');
+$node_subscriber->init(allows_streaming => 'logical');
+$node_subscriber->append_conf(
+        'postgresql.conf', qq(max_prepared_transactions = 10));
+$node_subscriber->start;
+
+# Create some pre-existing content on publisher
+$node_publisher->safe_psql('postgres', "CREATE TABLE tab_full (a int PRIMARY KEY)");
+$node_publisher->safe_psql('postgres',
+	"INSERT INTO tab_full SELECT generate_series(1,10)");
+$node_publisher->safe_psql('postgres', "CREATE TABLE tab_full2 (x text)");
+$node_publisher->safe_psql('postgres',
+	"INSERT INTO tab_full2 VALUES ('a'), ('b'), ('b')");
+
+# Setup structure on subscriber
+$node_subscriber->safe_psql('postgres', "CREATE TABLE tab_full (a int PRIMARY KEY)");
+$node_subscriber->safe_psql('postgres', "CREATE TABLE tab_full2 (x text)");
+
+# Setup logical replication
+my $publisher_connstr = $node_publisher->connstr . ' dbname=postgres';
+$node_publisher->safe_psql('postgres', "CREATE PUBLICATION tap_pub");
+$node_publisher->safe_psql('postgres',
+"ALTER PUBLICATION tap_pub ADD TABLE tab_full, tab_full2"
+);
+
+my $appname = 'tap_sub';
+$node_subscriber->safe_psql('postgres',
+"CREATE SUBSCRIPTION tap_sub CONNECTION '$publisher_connstr application_name=$appname' PUBLICATION tap_pub"
+);
+
+# Wait for subscriber to finish initialization
+my $caughtup_query =
+"SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$appname';";
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# Also wait for initial table sync to finish
+my $synced_query =
+"SELECT count(1) = 0 FROM pg_subscription_rel WHERE srsubstate NOT IN ('r', 's');";
+$node_subscriber->poll_query_until('postgres', $synced_query)
+  or die "Timed out while waiting for subscriber to synchronize data";
+
+# check that 2PC gets replicated to subscriber
+$node_publisher->safe_psql('postgres',
+	"BEGIN;INSERT INTO tab_full VALUES (11);PREPARE TRANSACTION 'test_prepared_tab_full';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+my $result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab_full';");
+   is($result, qq(1), 'transaction is prepared on subscriber');
+
+# check that 2PC gets committed on subscriber
+$node_publisher->safe_psql('postgres',
+	"COMMIT PREPARED 'test_prepared_tab_full';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is committed on subscriber
+$result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM tab_full where a = 11;");
+   is($result, qq(1), 'Row inserted via 2PC has committed on subscriber');
+$result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab_full';");
+   is($result, qq(0), 'transaction is committed on subscriber');
+
+# check that 2PC gets replicated to subscriber
+$node_publisher->safe_psql('postgres',
+	"BEGIN;INSERT INTO tab_full VALUES (12);PREPARE TRANSACTION 'test_prepared_tab_full';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab_full';");
+   is($result, qq(1), 'transaction is prepared on subscriber');
+
+# check that 2PC gets aborted on subscriber
+$node_publisher->safe_psql('postgres',
+	"ROLLBACK PREPARED 'test_prepared_tab_full';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is aborted on subscriber
+$result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM tab_full where a = 12;");
+   is($result, qq(0), 'Row inserted via 2PC is not present on subscriber');
+
+$result =
+   $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab_full';");
+   is($result, qq(0), 'transaction is aborted on subscriber');
+
+# Check that commit prepared is decoded properly on crash restart
+$node_publisher->safe_psql('postgres', "
+    BEGIN;
+    INSERT INTO tab_full VALUES (12);
+    INSERT INTO tab_full VALUES (13);
+    PREPARE TRANSACTION 'test_prepared_tab';");
+$node_subscriber->stop('immediate');
+$node_publisher->stop('immediate');
+$node_publisher->start;
+$node_subscriber->start;
+
+# commit post the restart
+$node_publisher->safe_psql('postgres', "COMMIT PREPARED 'test_prepared_tab';");
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+  or die "Timed out while waiting for subscriber to catch up";
+
+# check inserts are visible
+$result = $node_subscriber->safe_psql('postgres', "SELECT count(*) FROM tab_full where a IN (12,13);");
+is($result, qq(2), 'Rows inserted via 2PC are visible on the subscriber');
+
+# TODO add test cases involving DDL. This can be added after we add functionality
+# to replicate DDL changes to subscriber.
+
+# check all the cleanup
+$node_subscriber->safe_psql('postgres', "DROP SUBSCRIPTION tap_sub");
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_subscription");
+is($result, qq(0), 'check subscription was dropped on subscriber');
+
+$result = $node_publisher->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_slots");
+is($result, qq(0), 'check replication slot was dropped on publisher');
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_subscription_rel");
+is($result, qq(0),
+	'check subscription relation status was dropped on subscriber');
+
+$result = $node_publisher->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_slots");
+is($result, qq(0), 'check replication slot was dropped on publisher');
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_origin");
+is($result, qq(0), 'check replication origin was dropped on subscriber');
+
+$node_subscriber->stop('fast');
+$node_publisher->stop('fast');
diff --git a/src/test/subscription/t/021_twophase_streaming.pl b/src/test/subscription/t/021_twophase_streaming.pl
new file mode 100644
index 0000000..391eb17
--- /dev/null
+++ b/src/test/subscription/t/021_twophase_streaming.pl
@@ -0,0 +1,366 @@
+# logical replication of 2PC test
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More tests => 23;
+
+# Initialize publisher node
+my $node_publisher = get_new_node('publisher');
+$node_publisher->init(allows_streaming => 'logical');
+$node_publisher->append_conf('postgresql.conf', qq(max_prepared_transactions = 10));
+$node_publisher->append_conf('postgresql.conf', qq(logical_decoding_work_mem = 64kB));
+$node_publisher->start;
+
+# Create subscriber node
+my $node_subscriber = get_new_node('subscriber');
+$node_subscriber->init(allows_streaming => 'logical');
+$node_subscriber->append_conf('postgresql.conf', qq(max_prepared_transactions = 10));
+$node_subscriber->start;
+
+# Create some pre-existing content on publisher (uses same DDL as 015_stream test)
+$node_publisher->safe_psql('postgres',
+	"CREATE TABLE test_tab (a int primary key, b varchar)");
+$node_publisher->safe_psql('postgres',
+	"INSERT INTO test_tab VALUES (1, 'foo'), (2, 'bar')");
+
+# Setup structure on subscriber (columns a and b are compatible with same table name on publisher)
+$node_subscriber->safe_psql('postgres',
+	"CREATE TABLE test_tab (a int primary key, b text, c timestamptz DEFAULT now(), d bigint DEFAULT 999)");
+
+# Setup logical replication (streaming = on)
+my $publisher_connstr = $node_publisher->connstr . ' dbname=postgres';
+$node_publisher->safe_psql('postgres', "CREATE PUBLICATION tap_pub FOR TABLE test_tab");
+
+my $appname = 'tap_sub';
+$node_subscriber->safe_psql('postgres',
+	"CREATE SUBSCRIPTION tap_sub CONNECTION '$publisher_connstr application_name=$appname' PUBLICATION tap_pub WITH (streaming = on)");
+
+# Wait for subscriber to finish initialization
+my $caughtup_query =
+	"SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$appname';";
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# Also wait for initial table sync to finish
+my $synced_query =
+	"SELECT count(1) = 0 FROM pg_subscription_rel WHERE srsubstate NOT IN ('r', 's');";
+$node_subscriber->poll_query_until('postgres', $synced_query)
+	or die "Timed out while waiting for subscriber to synchronize data";
+
+my $result =
+  $node_subscriber->safe_psql('postgres', "SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(2|2|2), 'check initial data was copied to subscriber');
+
+# --------------------------------------------------------------
+# 2PC PREPARE / COMMIT PREPARED test.
+#
+# Mass data is streamed as a 2PC transaction.
+# Then there is a commit prepared.
+# Expect all data is replicated on subscriber side after the commit.
+# --------------------------------------------------------------
+#
+# check that 2PC gets replicated to subscriber
+# Insert, update and delete enough rows to exceed the 64kB limit.
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(1), 'transaction is prepared on subscriber');
+
+# 2PC gets committed
+$node_publisher->safe_psql('postgres',
+	"COMMIT PREPARED 'test_prepared_tab';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is committed on subscriber
+$result =
+  $node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(3334|3334|3334), 'Rows inserted by 2PC have committed on subscriber, and extra columns contain local defaults');
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(0), 'transaction is committed on subscriber');
+
+# --------------------------------------------------------------
+# 2PC PREPARE / ROLLBACK PREPARED test.
+#
+# Table is deleted back to 2 rows which are replicated on subscriber.
+# Mass data is streamed using 2PC but then there is a rollback prepared.
+# Expect data rolls back leaving only the 2 rows on the subscriber.
+# --------------------------------------------------------------
+#
+# First, delete the data (delete will be replicated)
+# Then insert, update and delete enough rows to exceed the 64kB limit.
+$node_publisher->safe_psql('postgres', q{
+    DELETE FROM test_tab WHERE a > 2;});
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(1), 'transaction is prepared on subscriber');
+
+# 2PC gets aborted
+$node_publisher->safe_psql('postgres',
+	"ROLLBACK PREPARED 'test_prepared_tab';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is aborted on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(2|2|2), 'check initial data was copied to subscriber');
+
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(0), 'transaction is aborted on subscriber');
+
+# --------------------------------------------------------------
+# Check that 2PC commit prepared is decoded properly on crash restart.
+#
+# insert, update and delete enough rows to exceed the 64kB limit.
+# Then server crashes before the 2PC transaction is committed.
+# After servers are restarted the pending transaction is committed.
+# Expect all data is replicated on subscriber side after the commit.
+# --------------------------------------------------------------
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+$node_subscriber->stop('immediate');
+$node_publisher->stop('immediate');
+$node_publisher->start;
+$node_subscriber->start;
+
+# commit post the restart
+$node_publisher->safe_psql('postgres',
+	"COMMIT PREPARED 'test_prepared_tab';");
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check inserts are visible
+$result =
+  $node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(3334|3334|3334), 'Rows inserted by 2PC have committed on subscriber, and extra columns contain local defaults');
+
+# --------------------------------------------------------------
+# Do INSERT outside of 2PC but before ROLLBACK PREPARED.
+#
+# Table is deleted back to 2 rows which are replicated on subscriber.
+# Mass data is streamed using 2PC.
+# A single row INSERT is done which is outside of the 2PC transaction
+# Then there is a rollback prepared.
+# Expect 2PC data rolls back leaving only 3 rows on the subscriber.
+# --------------------------------------------------------------
+#
+# First, delete the data (delete will be replicated)
+# Then insert, update and delete enough rows to exceed the 64kB limit.
+$node_publisher->safe_psql('postgres', q{
+    DELETE FROM test_tab WHERE a > 2;});
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(1), 'transaction is prepared on subscriber');
+
+# Insert a different record (now we are outside of the 2PC tx)
+# FIXME - this works OK but if INSERT overlaps pk of a row participating in the 2PC then it hangs
+$node_publisher->safe_psql('postgres',
+	"INSERT INTO test_tab VALUES (99999, 'foobar')");
+
+# 2PC gets aborted
+$node_publisher->safe_psql('postgres',
+	"ROLLBACK PREPARED 'test_prepared_tab';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is aborted on subscriber
+# but the extra INSERT outside of the 2PC still happened
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(3|3|3), 'check the outside insert was copied to subscriber');
+
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(0), 'transaction is aborted on subscriber');
+
+# --------------------------------------------------------------
+# Do INSERT outside of 2PC but before COMMIT PREPARED.
+#
+# Table is deleted back to 2 rows which are replicated on subscriber.
+# Mass data is streamed using 2PC.
+# A single row INSERT is done which is outside of the 2PC transaction
+# Then there is a commit prepared.
+# Expect 2PC data + the extra row are on the subscriber.
+# --------------------------------------------------------------
+#
+# First, delete the data (delete will be replicated)
+# Then insert, update and delete enough rows to exceed the 64kB limit.
+$node_publisher->safe_psql('postgres', q{
+    DELETE FROM test_tab WHERE a > 2;});
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(1), 'transaction is prepared on subscriber');
+
+# Insert a different record (now we are outside of the 2PC tx)
+# FIXME - this works OK but if INSERT overlaps pk of a row participating in the 2PC then it hangs
+$node_publisher->safe_psql('postgres',
+	"INSERT INTO test_tab VALUES (99999, 'foobar')");
+
+# 2PC gets committed
+$node_publisher->safe_psql('postgres',
+	"COMMIT PREPARED 'test_prepared_tab';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is committed on subscriber
+$result =
+  $node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(3335|3335|3335), 'Rows inserted by 2PC (as well as outside insert) have committed on subscriber, and extra columns contain local defaults');
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(0), 'transaction is committed on subscriber');
+
+# --------------------------------------------------------------
+# Do DELETE outside of 2PC but before COMMIT PREPARED.
+#
+# Table is deleted back to 2 rows which are replicated on subscriber.
+# Mass data is streamed using 2PC.
+# A single row DELETE is done for one of the records of the 2PC transaction
+# Then there is a commit prepared.
+# Expect all the 2PC data rows on the subscriber (because delete would have failed).
+# --------------------------------------------------------------
+#
+# First, delete the data (delete will be replicated)
+# Then insert, update and delete enough rows to exceed the 64kB limit.
+$node_publisher->safe_psql('postgres', q{
+    DELETE FROM test_tab WHERE a > 2;});
+$node_publisher->safe_psql('postgres', q{
+	BEGIN;
+	INSERT INTO test_tab SELECT i, md5(i::text) FROM generate_series(3, 5000) s(i);
+	UPDATE test_tab SET b = md5(b) WHERE mod(a,2) = 0;
+	DELETE FROM test_tab WHERE mod(a,3) = 0;
+	PREPARE TRANSACTION 'test_prepared_tab';});
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is in prepared state on subscriber
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(1), 'transaction is prepared on subscriber');
+
+# DELETE one of the prepared 2PC records before they get committed (we are outside of the 2PC tx)
+$node_publisher->safe_psql('postgres',
+	"DELETE FROM test_tab WHERE a = 5");
+
+# 2PC gets committed
+$node_publisher->safe_psql('postgres',
+	"COMMIT PREPARED 'test_prepared_tab';");
+
+$node_publisher->poll_query_until('postgres', $caughtup_query)
+	or die "Timed out while waiting for subscriber to catch up";
+
+# check that transaction is committed on subscriber
+$result =
+  $node_subscriber->safe_psql('postgres',
+	"SELECT count(*), count(c), count(d = 999) FROM test_tab");
+is($result, qq(3334|3334|3334), 'Rows inserted by 2PC have committed on subscriber. Nothing was deleted');
+$result =
+  $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM test_tab WHERE a = 5");
+is($result, qq(1), 'The row we deleted before the commit till exists on subscriber.');
+$result =
+	$node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_prepared_xacts where gid = 'test_prepared_tab';");
+is($result, qq(0), 'transaction is committed on subscriber');
+
+# --------------------------------------------------------------
+
+# TODO add test cases involving DDL. This can be added after we add functionality
+# to replicate DDL changes to subscriber.
+
+# check all the cleanup
+$node_subscriber->safe_psql('postgres', "DROP SUBSCRIPTION tap_sub");
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_subscription");
+is($result, qq(0), 'check subscription was dropped on subscriber');
+
+$result = $node_publisher->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_slots");
+is($result, qq(0), 'check replication slot was dropped on publisher');
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_subscription_rel");
+is($result, qq(0),
+	'check subscription relation status was dropped on subscriber');
+
+$result = $node_publisher->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_slots");
+is($result, qq(0), 'check replication slot was dropped on publisher');
+
+$result = $node_subscriber->safe_psql('postgres',
+	"SELECT count(*) FROM pg_replication_origin");
+is($result, qq(0), 'check replication origin was dropped on subscriber');
+
+$node_subscriber->stop('fast');
+$node_publisher->stop('fast');
-- 
1.8.3.1