v66-0002-Resume-retaining-the-information-for-conflict-de.patch

From e2db073ab818d56e9fa587549674603265a3132f Mon Sep 17 00:00:00 2001
From: Zhijie Hou <houzj.fnst@fujitsu.com>
Date: Tue, 26 Aug 2025 18:55:57 +0800
Subject: [PATCH v66 2/3] Resume retaining the information for conflict
 detection

The patch allows the launcher to re-initialized invalidated slot, if at
least one apply worker has confirmed that the retention duration is now within
the max_retention_duration.
---
 doc/src/sgml/ref/create_subscription.sgml  |   7 +-
 src/backend/commands/subscriptioncmds.c    |   9 +-
 src/backend/replication/logical/launcher.c | 102 ++++++++---
 src/backend/replication/logical/worker.c   | 200 ++++++++++++++++++---
 4 files changed, 261 insertions(+), 57 deletions(-)

diff --git a/doc/src/sgml/ref/create_subscription.sgml b/doc/src/sgml/ref/create_subscription.sgml
index 478b8352164..2d2800a8b9a 100644
--- a/doc/src/sgml/ref/create_subscription.sgml
+++ b/doc/src/sgml/ref/create_subscription.sgml
@@ -540,8 +540,11 @@ CREATE SUBSCRIPTION <replaceable class="parameter">subscription_name</replaceabl
           retention duration exceeded the
           <literal>max_retention_duration</literal> set within the
           corresponding subscription. To re-enable retention manually, you can
-          disable <literal>retain_dead_tuples</literal> for all subscriptions and
-          re-enable it after confirming this replication slot has been dropped.
+          disable <literal>retain_dead_tuples</literal> and re-enable it.
+          Alternatively, the retention will be automatically resumed once at
+          least one apply worker confirms that the retention duration is within
+          the specified limit, or if a new subscription with retain_dead_tuples
+          enabled is created.
          </para>
          <para>
           Note that overall retention will not stop if other subscriptions
diff --git a/src/backend/commands/subscriptioncmds.c b/src/backend/commands/subscriptioncmds.c
index f9d92702434..5e737f97766 100644
--- a/src/backend/commands/subscriptioncmds.c
+++ b/src/backend/commands/subscriptioncmds.c
@@ -855,7 +855,14 @@ CreateSubscription(ParseState *pstate, CreateSubscriptionStmt *stmt,
 
 	pgstat_create_subscription(subid);
 
-	if (opts.enabled)
+	/*
+	 * If the subscription is enabled, notify the launcher to start the apply
+	 * worker.
+	 *
+	 * If the subscription has retain_dead_tuples enabled, notify the launcher
+	 * to create or resume the conflict detection slot.
+	 */
+	if (opts.enabled || opts.retaindeadtuples)
 		ApplyLauncherWakeupAtCommit();
 
 	ObjectAddressSet(myself, SubscriptionRelationId, subid);
diff --git a/src/backend/replication/logical/launcher.c b/src/backend/replication/logical/launcher.c
index c0211867881..73d2c22ae8b 100644
--- a/src/backend/replication/logical/launcher.c
+++ b/src/backend/replication/logical/launcher.c
@@ -102,9 +102,12 @@ static int	logicalrep_pa_worker_count(Oid subid);
 static void logicalrep_launcher_attach_dshmem(void);
 static void ApplyLauncherSetWorkerStartTime(Oid subid, TimestampTz start_time);
 static TimestampTz ApplyLauncherGetWorkerStartTime(Oid subid);
-static void compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin);
+static void compute_min_nonremovable_xid(LogicalRepWorker *worker,
+										 bool can_update_xmin,
+										 TransactionId *xmin);
 static bool acquire_conflict_slot_if_exists(void);
 static void update_conflict_slot_xmin(TransactionId new_xmin);
+static void init_conflict_slot_xmin(void);
 
 
 /*
@@ -1233,17 +1236,27 @@ ApplyLauncherMain(Datum main_arg)
 				 */
 				CreateConflictDetectionSlot();
 
-				/*
-				 * Can't advance xmin of the slot unless all the subscriptions
-				 * actively retaining dead tuples are enabled. This is
-				 * required to ensure that we don't advance the xmin of
-				 * CONFLICT_DETECTION_SLOT if one of the subscriptions is not
-				 * enabled. Otherwise, we won't be able to detect conflicts
-				 * reliably for such a subscription even though it has set the
-				 * retain_dead_tuples option.
-				 */
 				if (sub->retentionactive)
+				{
+					/*
+					 * Can't advance xmin of the slot unless all the
+					 * subscriptions actively retaining dead tuples are
+					 * enabled. This is required to ensure that we don't
+					 * advance the xmin of CONFLICT_DETECTION_SLOT if one of
+					 * the subscriptions is not enabled. Otherwise, we won't
+					 * be able to detect conflicts reliably for such a
+					 * subscription even though it has set the
+					 * retain_dead_tuples option.
+					 */
 					can_update_xmin &= sub->enabled;
+
+					/*
+					 * Initialize slot.xmin as a subscription resumes
+					 * retention of information useful for conflict detection.
+					 */
+					if (!TransactionIdIsValid(MyReplicationSlot->data.xmin))
+						init_conflict_slot_xmin();
+				}
 			}
 
 			if (!sub->enabled)
@@ -1260,10 +1273,8 @@ ApplyLauncherMain(Datum main_arg)
 				 * required for conflict detection among all running apply
 				 * workers that enables retain_dead_tuples.
 				 */
-				if (sub->retaindeadtuples &&
-					sub->retentionactive &&
-					can_update_xmin)
-					compute_min_nonremovable_xid(w, &xmin);
+				if (sub->retaindeadtuples && sub->retentionactive)
+					compute_min_nonremovable_xid(w, can_update_xmin, &xmin);
 
 				/* worker is running already */
 				continue;
@@ -1375,7 +1386,8 @@ ApplyLauncherMain(Datum main_arg)
  * in *xmin.
  */
 static void
-compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin)
+compute_min_nonremovable_xid(LogicalRepWorker *worker, bool can_update_xmin,
+							 TransactionId *xmin)
 {
 	TransactionId nonremovable_xid;
 
@@ -1391,7 +1403,27 @@ compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin)
 	nonremovable_xid = worker->oldest_nonremovable_xid;
 	SpinLockRelease(&worker->relmutex);
 
-	Assert(TransactionIdIsValid(nonremovable_xid));
+	/*
+	 * Assign slot.xmin to the apply worker's oldest_nonremovable_xid if the
+	 * latter is invalid. This ensures the apply worker continues to maintain
+	 * the oldest_nonremovable_xid (see resume_conflict_info_retention).
+	 */
+	if (!TransactionIdIsValid(nonremovable_xid))
+	{
+		nonremovable_xid = MyReplicationSlot->data.xmin;
+
+		SpinLockAcquire(&worker->relmutex);
+		worker->oldest_nonremovable_xid = nonremovable_xid;
+		SpinLockRelease(&worker->relmutex);
+
+		/* Notify the apply worker to start the next cycle of management */
+		LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
+		logicalrep_worker_wakeup_ptr(worker);
+		LWLockRelease(LogicalRepWorkerLock);
+	}
+
+	if (!can_update_xmin)
+		return;
 
 	if (!TransactionIdIsValid(*xmin) ||
 		TransactionIdPrecedes(nonremovable_xid, *xmin))
@@ -1452,23 +1484,15 @@ update_conflict_slot_xmin(TransactionId new_xmin)
 }
 
 /*
- * Create and acquire the replication slot used to retain information for
- * conflict detection, if not yet.
+ * Initialize the xmin for the conflict detection slot.
  */
-void
-CreateConflictDetectionSlot(void)
+static void
+init_conflict_slot_xmin(void)
 {
 	TransactionId xmin_horizon;
 
-	/* Exit early, if the replication slot is already created and acquired */
-	if (MyReplicationSlot)
-		return;
-
-	ereport(LOG,
-			errmsg("creating replication conflict detection slot"));
-
-	ReplicationSlotCreate(CONFLICT_DETECTION_SLOT, false, RS_PERSISTENT, false,
-						  false, false);
+	Assert(MyReplicationSlot &&
+		   !TransactionIdIsValid(MyReplicationSlot->data.xmin));
 
 	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
 
@@ -1488,6 +1512,26 @@ CreateConflictDetectionSlot(void)
 	ReplicationSlotSave();
 }
 
+/*
+ * Create and acquire the replication slot used to retain information for
+ * conflict detection, if not yet.
+ */
+void
+CreateConflictDetectionSlot(void)
+{
+	/* Exit early, if the replication slot is already created and acquired */
+	if (MyReplicationSlot)
+		return;
+
+	ereport(LOG,
+			errmsg("creating replication conflict detection slot"));
+
+	ReplicationSlotCreate(CONFLICT_DETECTION_SLOT, false, RS_PERSISTENT, false,
+						  false, false);
+
+	init_conflict_slot_xmin();
+}
+
 /*
  * Is current process the logical replication launcher?
  */
diff --git a/src/backend/replication/logical/worker.c b/src/backend/replication/logical/worker.c
index d378eb08c71..bc552ff036c 100644
--- a/src/backend/replication/logical/worker.c
+++ b/src/backend/replication/logical/worker.c
@@ -190,7 +190,7 @@
  * update_deleted is necessary, as the UPDATEs in remote transactions should be
  * ignored if their timestamp is earlier than that of the dead tuples.
  *
- * If max_retention_duration is defined, one additional phase is
+ * If max_retention_duration is defined, two additional phases are
  * involved:
  *
  * - RDT_STOP_CONFLICT_INFO_RETENTION:
@@ -200,6 +200,19 @@
  *   pg_subscription.subretentionactive is updated to false within a new
  *   transaction, and oldest_nonremovable_xid is set to InvalidTransactionId.
  *
+ * - RDT_RESUME_CONFLICT_INFO_RETENTION:
+ *   This phase is triggered when retention was previously stopped, and the time
+ *   required to advance the non-removable transaction ID in the
+ *   RDT_WAIT_FOR_LOCAL_FLUSH phase has decreased to within acceptable limits
+ *   (or if max_retention_duration is set to 0). During this phase,
+ *   pg_subscription.subretentionactive is updated to true within a new
+ *   transaction, and we wait for the launcher to initialize the
+ *   oldest_nonremovable_xid before proceeding to RDT_GET_CANDIDATE_XID phase.
+ *
+ * Note that the state could transition to RDT_RESUME_CONFLICT_INFO_RETENTION at
+ * any phase if the retention has been stopped, but
+ * max_retention_duration is now set to 0.
+ *
  * Note that advancing the non-removable transaction ID is not supported if the
  * publisher is also a physical standby. This is because the logical walsender
  * on the standby can only get the WAL replay position but there may be more
@@ -385,6 +398,7 @@ typedef enum
 	RDT_WAIT_FOR_PUBLISHER_STATUS,
 	RDT_WAIT_FOR_LOCAL_FLUSH,
 	RDT_STOP_CONFLICT_INFO_RETENTION,
+	RDT_RESUME_CONFLICT_INFO_RETENTION,
 } RetainDeadTuplesPhase;
 
 /*
@@ -572,6 +586,10 @@ static void wait_for_local_flush(RetainDeadTuplesData *rdt_data);
 static void stop_conflict_info_retention(RetainDeadTuplesData *rdt_data);
 static void reset_retention_data_fields(RetainDeadTuplesData *rdt_data);
 static bool should_stop_conflict_info_retention(RetainDeadTuplesData *rdt_data);
+static bool should_resume_retention_immediately(RetainDeadTuplesData *rdt_data,
+												bool status_received);
+static void resume_conflict_info_retention(RetainDeadTuplesData *rdt_data);
+static void update_retention_status(bool active);
 static void adjust_xid_advance_interval(RetainDeadTuplesData *rdt_data,
 										bool new_xid_found);
 
@@ -4337,6 +4355,13 @@ maybe_advance_nonremovable_xid(RetainDeadTuplesData *rdt_data,
 	if (!can_advance_nonremovable_xid(rdt_data))
 		return;
 
+	/*
+	 * Resume retention immediately if required. (See
+	 * should_resume_retention_immediately() for details).
+	 */
+	if (should_resume_retention_immediately(rdt_data, status_received))
+		rdt_data->phase = RDT_RESUME_CONFLICT_INFO_RETENTION;
+
 	process_rdt_phase_transition(rdt_data, status_received);
 }
 
@@ -4359,10 +4384,6 @@ can_advance_nonremovable_xid(RetainDeadTuplesData *rdt_data)
 	if (!MySubscription->retaindeadtuples)
 		return false;
 
-	/* No need to advance if we have already stopped retaining */
-	if (!TransactionIdIsValid(MyLogicalRepWorker->oldest_nonremovable_xid))
-		return false;
-
 	return true;
 }
 
@@ -4391,6 +4412,9 @@ process_rdt_phase_transition(RetainDeadTuplesData *rdt_data,
 		case RDT_STOP_CONFLICT_INFO_RETENTION:
 			stop_conflict_info_retention(rdt_data);
 			break;
+		case RDT_RESUME_CONFLICT_INFO_RETENTION:
+			resume_conflict_info_retention(rdt_data);
+			break;
 	}
 }
 
@@ -4647,6 +4671,18 @@ wait_for_local_flush(RetainDeadTuplesData *rdt_data)
 	if (last_flushpos < rdt_data->remote_lsn)
 		return;
 
+	/*
+	 * If conflict info retention was previously stopped due to a timeout, and
+	 * the time required to advance the non-removable transaction ID has now
+	 * decreased to within acceptable limits, resume the rentention.
+	 */
+	if (!TransactionIdIsValid(MyLogicalRepWorker->oldest_nonremovable_xid))
+	{
+		rdt_data->phase = RDT_RESUME_CONFLICT_INFO_RETENTION;
+		process_rdt_phase_transition(rdt_data, false);
+		return;
+	}
+
 	/*
 	 * Reaching here means the remote WAL position has been received, and all
 	 * transactions up to that position on the publisher have been applied and
@@ -4674,6 +4710,91 @@ wait_for_local_flush(RetainDeadTuplesData *rdt_data)
  */
 static void
 stop_conflict_info_retention(RetainDeadTuplesData *rdt_data)
+{
+	update_retention_status(false);
+
+	SpinLockAcquire(&MyLogicalRepWorker->relmutex);
+	MyLogicalRepWorker->oldest_nonremovable_xid = InvalidTransactionId;
+	SpinLockRelease(&MyLogicalRepWorker->relmutex);
+
+	ereport(LOG,
+			errmsg("logical replication worker for subscription \"%s\" has stopped retaining the information for detecting conflicts",
+				   MySubscription->name),
+			errdetail("The retention duration for information used in conflict detection has exceeded the maximum limit of %u ms.",
+					  MySubscription->maxretention),
+			errhint("You might need to increase \"%s\".",
+					"max_retention_duration"));
+
+	reset_retention_data_fields(rdt_data);
+
+	/* process the next phase */
+	process_rdt_phase_transition(rdt_data, false);
+}
+
+/*
+ * Workhorse for the RDT_RESUME_CONFLICT_INFO_RETENTION phase.
+ */
+static void
+resume_conflict_info_retention(RetainDeadTuplesData *rdt_data)
+{
+	TransactionId nonremovable_xid;
+
+	/* Update the pg_subscription.retentionactive if not yet */
+	if (!MySubscription->retentionactive)
+	{
+		update_retention_status(true);
+
+		ereport(LOG,
+				errmsg("logical replication worker for subscription \"%s\" will resume retaining the information for detecting conflicts",
+					   MySubscription->name),
+				MySubscription->maxretention
+				? errdetail("Retention of information used for conflict detection is now within the max_conflict_retention_duration of %u ms.",
+							MySubscription->maxretention)
+				: errdetail("Retention of information used for conflict detection is now indefinite."));
+	}
+
+	SpinLockAcquire(&MyLogicalRepWorker->relmutex);
+	nonremovable_xid = MyLogicalRepWorker->oldest_nonremovable_xid;
+	SpinLockRelease(&MyLogicalRepWorker->relmutex);
+
+	/*
+	 * Return if the launcher has not initialized oldest_nonremovable_xid.
+	 *
+	 * It might seem feasible to directly check the conflict detection
+	 * slot.xmin instead of relying on the launcher to assign the worker's
+	 * oldest_nonremovable_xid; however, that could lead to a race condition
+	 * where slot.xmin is set to InvalidTransactionId immediately after the
+	 * check. In such cases, oldest_nonremovable_xid would no longer be
+	 * protected by a replication slot and could become unreliable if a
+	 * wraparound occurs.
+	 */
+	if (!TransactionIdIsValid(nonremovable_xid))
+		return;
+
+	/*
+	 * Proceed to the next phase if either the launcher has initialized
+	 * slot.xmin and assigned it to oldest_nonremovable_xid, or retention has
+	 * not been stopped yet. The latter situation arises when transitioning
+	 * from the RDT_STOP_CONFLICT_INFO_RETENTION phase but subretentionactive
+	 * has not been updated due to the inability to start a new transaction
+	 * (see stop_conflict_info_retention).
+	 */
+	Assert(MySubscription->retentionactive);
+
+	reset_retention_data_fields(rdt_data);
+
+	/* process the next phase */
+	process_rdt_phase_transition(rdt_data, false);
+}
+
+/*
+ * Update pg_subscription.subretentionactive to the given value within a new
+ * transaction.
+ *
+ * Skip the update if currently within an existing transaction.
+ */
+static void
+update_retention_status(bool active)
 {
 	/*
 	 * Do not update the catalog during an active transaction. The transaction
@@ -4691,26 +4812,16 @@ stop_conflict_info_retention(RetainDeadTuplesData *rdt_data)
 	 */
 	PushActiveSnapshot(GetTransactionSnapshot());
 
-	/* Set pg_subscription.subretentionactive to false */
-	UpdateDeadTupleRetentionStatus(MySubscription->oid, false);
+	/* Update pg_subscription.subretentionactive */
+	UpdateDeadTupleRetentionStatus(MySubscription->oid, active);
 
 	PopActiveSnapshot();
 	CommitTransactionCommand();
 
-	SpinLockAcquire(&MyLogicalRepWorker->relmutex);
-	MyLogicalRepWorker->oldest_nonremovable_xid = InvalidTransactionId;
-	SpinLockRelease(&MyLogicalRepWorker->relmutex);
-
-	ereport(LOG,
-			errmsg("logical replication worker for subscription \"%s\" has stopped retaining the information for detecting conflicts",
-				   MySubscription->name),
-			errdetail("Retention of information used for conflict detection has exceeded max_retention_duration of %u ms.",
-					  MySubscription->maxretention));
-
-	/* Notify launcher to update the conflict slot */
+	/* Notify launcher to update the xmin of the conflict slot */
 	ApplyLauncherWakeup();
 
-	reset_retention_data_fields(rdt_data);
+	MySubscription->retentionactive = active;
 }
 
 /*
@@ -4740,9 +4851,8 @@ reset_retention_data_fields(RetainDeadTuplesData *rdt_data)
  * RDT_STOP_CONFLICT_INFO_RETENTION phase and return true. Otherwise, return
  * false.
  *
- * Note: Retention won't be resumed automatically. The user must manually
- * disable retain_dead_tuples and re-enable it after confirming that the
- * replication slot maintained by the launcher has been dropped.
+ * The retention will resume automatically if the worker has confirmed that the
+ * retention duration is now within the max_retention_duration.
  */
 static bool
 should_stop_conflict_info_retention(RetainDeadTuplesData *rdt_data)
@@ -4773,14 +4883,54 @@ should_stop_conflict_info_retention(RetainDeadTuplesData *rdt_data)
 									rdt_data->table_sync_wait_time))
 		return false;
 
-	rdt_data->phase = RDT_STOP_CONFLICT_INFO_RETENTION;
+	/* Stop retention if not yet */
+	if (TransactionIdIsValid(MyLogicalRepWorker->oldest_nonremovable_xid))
+	{
+		rdt_data->phase = RDT_STOP_CONFLICT_INFO_RETENTION;
 
-	/* process the next phase */
-	process_rdt_phase_transition(rdt_data, false);
+		/* process the next phase */
+		process_rdt_phase_transition(rdt_data, false);
+	}
+
+	reset_retention_data_fields(rdt_data);
 
 	return true;
 }
 
+/*
+ * Check whether retention should be resumed immediately if it has been
+ * previously stopped, but max_retention_duration is now set to 0.
+ */
+static bool
+should_resume_retention_immediately(RetainDeadTuplesData *rdt_data, bool status_received)
+{
+	/* Return false if retention is already being resumed */
+	if (rdt_data->phase == RDT_RESUME_CONFLICT_INFO_RETENTION)
+		return false;
+
+	/* Return false if max_retention_duration is not 0 */
+	if (MySubscription->maxretention)
+		return false;
+
+	/*
+	 * Do not resume when waiting for publisher status, as doing so may result
+	 * in the message being processed after the data and phase have been
+	 * reset, potentially causing it to be mistakenly identified as a new
+	 * message. This could lead to the premature advancement of
+	 * oldest_nonremovable_xid.
+	 */
+	if (rdt_data->phase == RDT_WAIT_FOR_PUBLISHER_STATUS &&
+		!status_received)
+		return false;
+
+	/*
+	 * Resume retention if we are in the process of stopping or have already
+	 * stopped retention.
+	 */
+	return rdt_data->phase == RDT_STOP_CONFLICT_INFO_RETENTION ||
+		!TransactionIdIsValid(MyLogicalRepWorker->oldest_nonremovable_xid);
+}
+
 /*
  * Adjust the interval for advancing non-removable transaction IDs.
  *
-- 
2.31.1