v35-0002-Maintain-the-replication-slot-in-logical-launche.patch

From 69781f596d88e28ebad1cafbf10ec1a562bbfd52 Mon Sep 17 00:00:00 2001
From: Hou Zhijie <houzj.fnst@cn.fujitsu.com>
Date: Tue, 3 Jun 2025 11:40:30 +0800
Subject: [PATCH v35 2/3] Maintain the replication slot in logical launcher to
 retain dead tuples

This patch enables the logical replication launcher to create and maintain a
replication slot named pg_conflict_detection.

The launcher periodically collects the oldest_nonremovable_xid from all apply
workers. It then computes the minimum transaction ID and advances the xmin
value of the replication slot if it precedes the computed value.
---
 doc/src/sgml/config.sgml                      |   2 +
 doc/src/sgml/func.sgml                        |  16 +-
 doc/src/sgml/protocol.sgml                    |   2 +
 doc/src/sgml/ref/create_subscription.sgml     |   4 +-
 src/backend/access/transam/xlogrecovery.c     |   2 +-
 src/backend/commands/subscriptioncmds.c       |   2 +-
 src/backend/replication/logical/launcher.c    | 197 +++++++++++++++++-
 .../replication/logical/reorderbuffer.c       |   2 +-
 src/backend/replication/logical/worker.c      |   3 +
 src/backend/replication/slot.c                |  36 +++-
 src/include/replication/logicallauncher.h     |   1 +
 src/include/replication/slot.h                |  11 +-
 12 files changed, 265 insertions(+), 13 deletions(-)

diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index 021153b2a5f..20b74d92006 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -4961,6 +4961,8 @@ ANY <replaceable class="parameter">num_sync</replaceable> ( <replaceable class="
           new setting.
           This setting has no effect if <varname>primary_conninfo</varname> is not
           set or the server is not in standby mode.
+          The name cannot be <literal>pg_conflict_detection</literal> as it is
+          reserved for the conflict detection slot.
          </para>
         </listitem>
        </varlistentry>
diff --git a/doc/src/sgml/func.sgml b/doc/src/sgml/func.sgml
index c67688cbf5f..1fd0f908554 100644
--- a/doc/src/sgml/func.sgml
+++ b/doc/src/sgml/func.sgml
@@ -29539,7 +29539,9 @@ postgres=# SELECT '0/0'::pg_lsn + pd.segment_number * ps.setting::int + :offset
        </para>
        <para>
         Creates a new physical replication slot named
-        <parameter>slot_name</parameter>. The optional second parameter,
+        <parameter>slot_name</parameter>. The name cannot be
+        <literal>pg_conflict_detection</literal> as it is reserved for the
+        conflict detection slot. The optional second parameter,
         when <literal>true</literal>, specifies that the <acronym>LSN</acronym> for this
         replication slot be reserved immediately; otherwise
         the <acronym>LSN</acronym> is reserved on first connection from a streaming
@@ -29583,7 +29585,9 @@ postgres=# SELECT '0/0'::pg_lsn + pd.segment_number * ps.setting::int + :offset
        <para>
         Creates a new logical (decoding) replication slot named
         <parameter>slot_name</parameter> using the output plugin
-        <parameter>plugin</parameter>. The optional third
+        <parameter>plugin</parameter>. The name cannot be
+        <literal>pg_conflict_detection</literal> as it is reserved for
+        the conflict detection slot. The optional third
         parameter, <parameter>temporary</parameter>, when set to true, specifies that
         the slot should not be permanently stored to disk and is only meant
         for use by the current session. Temporary slots are also
@@ -29613,6 +29617,8 @@ postgres=# SELECT '0/0'::pg_lsn + pd.segment_number * ps.setting::int + :offset
        <para>
         Copies an existing physical replication slot named <parameter>src_slot_name</parameter>
         to a physical replication slot named <parameter>dst_slot_name</parameter>.
+        The new slot name cannot be <literal>pg_conflict_detection</literal>,
+        as it is reserved for the conflict detection.
         The copied physical slot starts to reserve WAL from the same <acronym>LSN</acronym> as the
         source slot.
         <parameter>temporary</parameter> is optional. If <parameter>temporary</parameter>
@@ -29635,8 +29641,10 @@ postgres=# SELECT '0/0'::pg_lsn + pd.segment_number * ps.setting::int + :offset
         Copies an existing logical replication slot
         named <parameter>src_slot_name</parameter> to a logical replication
         slot named <parameter>dst_slot_name</parameter>, optionally changing
-        the output plugin and persistence.  The copied logical slot starts
-        from the same <acronym>LSN</acronym> as the source logical slot.  Both
+        the output plugin and persistence.  The name cannot be
+        <literal>pg_conflict_detection</literal> as it is reserved for
+        the conflict detection.  The copied logical slot starts from the same
+        <acronym>LSN</acronym> as the source logical slot.  Both
         <parameter>temporary</parameter> and <parameter>plugin</parameter> are
         optional; if they are omitted, the values of the source slot are used.
         The <literal>failover</literal> option of the source logical slot
diff --git a/doc/src/sgml/protocol.sgml b/doc/src/sgml/protocol.sgml
index ed56df85ca7..cbd36641161 100644
--- a/doc/src/sgml/protocol.sgml
+++ b/doc/src/sgml/protocol.sgml
@@ -2225,6 +2225,8 @@ psql "dbname=postgres replication=database" -c "IDENTIFY_SYSTEM;"
          <para>
           The name of the slot to create. Must be a valid replication slot
           name (see <xref linkend="streaming-replication-slots-manipulation"/>).
+          The name cannot be <literal>pg_conflict_detection</literal> as it
+          is reserved for the conflict detection.
          </para>
         </listitem>
        </varlistentry>
diff --git a/doc/src/sgml/ref/create_subscription.sgml b/doc/src/sgml/ref/create_subscription.sgml
index 57dec28a5df..37fd40252a3 100644
--- a/doc/src/sgml/ref/create_subscription.sgml
+++ b/doc/src/sgml/ref/create_subscription.sgml
@@ -169,7 +169,9 @@ CREATE SUBSCRIPTION <replaceable class="parameter">subscription_name</replaceabl
         <listitem>
          <para>
           Name of the publisher's replication slot to use.  The default is
-          to use the name of the subscription for the slot name.
+          to use the name of the subscription for the slot name. The name cannot
+          be <literal>pg_conflict_detection</literal> as it is reserved for the
+          conflict detection.
          </para>
 
          <para>
diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c
index 6ce979f2d8b..2dcda37bc77 100644
--- a/src/backend/access/transam/xlogrecovery.c
+++ b/src/backend/access/transam/xlogrecovery.c
@@ -4760,7 +4760,7 @@ bool
 check_primary_slot_name(char **newval, void **extra, GucSource source)
 {
 	if (*newval && strcmp(*newval, "") != 0 &&
-		!ReplicationSlotValidateName(*newval, WARNING))
+		!ReplicationSlotValidateName(*newval, false, WARNING))
 		return false;
 
 	return true;
diff --git a/src/backend/commands/subscriptioncmds.c b/src/backend/commands/subscriptioncmds.c
index 4aec73bcc6b..46d4e65da97 100644
--- a/src/backend/commands/subscriptioncmds.c
+++ b/src/backend/commands/subscriptioncmds.c
@@ -210,7 +210,7 @@ parse_subscription_options(ParseState *pstate, List *stmt_options,
 			if (strcmp(opts->slot_name, "none") == 0)
 				opts->slot_name = NULL;
 			else
-				ReplicationSlotValidateName(opts->slot_name, ERROR);
+				ReplicationSlotValidateName(opts->slot_name, false, ERROR);
 		}
 		else if (IsSet(supported_opts, SUBOPT_COPY_DATA) &&
 				 strcmp(defel->defname, "copy_data") == 0)
diff --git a/src/backend/replication/logical/launcher.c b/src/backend/replication/logical/launcher.c
index 67bacc0f28c..9b12b2900e6 100644
--- a/src/backend/replication/logical/launcher.c
+++ b/src/backend/replication/logical/launcher.c
@@ -32,6 +32,7 @@
 #include "postmaster/interrupt.h"
 #include "replication/logicallauncher.h"
 #include "replication/origin.h"
+#include "replication/slot.h"
 #include "replication/walreceiver.h"
 #include "replication/worker_internal.h"
 #include "storage/ipc.h"
@@ -90,8 +91,12 @@ static dshash_table *last_start_times = NULL;
 
 static bool on_commit_launcher_wakeup = false;
 
+/*
+ * Whether the slot used to retain dead tuples for conflict detection has been
+ * dropped.
+ */
+static bool conflict_slot_dropped = false;
 
-static void ApplyLauncherWakeup(void);
 static void logicalrep_launcher_onexit(int code, Datum arg);
 static void logicalrep_worker_onexit(int code, Datum arg);
 static void logicalrep_worker_detach(void);
@@ -100,6 +105,12 @@ 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,
+										 bool *can_advance_xmin);
+static void create_conflict_slot_if_not_exists(void);
+static void advance_conflict_slot_xmin(TransactionId new_xmin);
+static void drop_conflict_slot_if_exists(void);
 
 
 /*
@@ -1106,7 +1117,10 @@ ApplyLauncherWakeupAtCommit(void)
 		on_commit_launcher_wakeup = true;
 }
 
-static void
+/*
+ * Wakeup the launcher immediately.
+ */
+void
 ApplyLauncherWakeup(void)
 {
 	if (LogicalRepCtx->launcher_pid != 0)
@@ -1147,6 +1161,8 @@ ApplyLauncherMain(Datum main_arg)
 		MemoryContext subctx;
 		MemoryContext oldctx;
 		long		wait_time = DEFAULT_NAPTIME_PER_CYCLE;
+		bool		can_advance_xmin = true;
+		TransactionId xmin = InvalidTransactionId;
 
 		CHECK_FOR_INTERRUPTS();
 
@@ -1167,12 +1183,17 @@ ApplyLauncherMain(Datum main_arg)
 			long		elapsed;
 
 			if (!sub->enabled)
+			{
+				compute_min_nonremovable_xid(NULL, &xmin, &can_advance_xmin);
 				continue;
+			}
 
 			LWLockAcquire(LogicalRepWorkerLock, LW_SHARED);
 			w = logicalrep_worker_find(sub->oid, InvalidOid, false);
 			LWLockRelease(LogicalRepWorkerLock);
 
+			compute_min_nonremovable_xid(w, &xmin, &can_advance_xmin);
+
 			if (w != NULL)
 				continue;		/* worker is running already */
 
@@ -1207,6 +1228,16 @@ ApplyLauncherMain(Datum main_arg)
 			}
 		}
 
+		/*
+		 * Maintain the xmin value of the replication slot for conflict
+		 * detection if needed. Otherwise, drop the slot if we're no longer
+		 * retaining information useful for conflict detection.
+		 */
+		if (!sublist)
+			drop_conflict_slot_if_exists();
+		else if (can_advance_xmin)
+			advance_conflict_slot_xmin(xmin);
+
 		/* Switch back to original memory context. */
 		MemoryContextSwitchTo(oldctx);
 		/* Clean the temporary memory. */
@@ -1234,6 +1265,168 @@ ApplyLauncherMain(Datum main_arg)
 	/* Not reachable */
 }
 
+/*
+ * Compute the minimum non-removable transaction ID from all apply workers.
+ * Store the result in *xmin.
+ *
+ * If the slot cannot be advanced during this cycle, due to either a disabled
+ * subscription or an inactive worker, *can_advance_xmin is set to false.
+ */
+static void
+compute_min_nonremovable_xid(LogicalRepWorker *worker, TransactionId *xmin,
+							 bool *can_advance_xmin)
+{
+	if (!*can_advance_xmin)
+		return;
+
+	if (worker)
+	{
+		TransactionId nonremovable_xid;
+
+		/*
+		 * Assume the replication slot for conflict detection is created
+		 * before the worker starts.
+		 */
+		Assert(MyReplicationSlot);
+
+		SpinLockAcquire(&worker->relmutex);
+		nonremovable_xid = worker->oldest_nonremovable_xid;
+		SpinLockRelease(&worker->relmutex);
+
+		/*
+		 * Stop advancing xmin if an invalid non-removable transaction ID is
+		 * found, otherwise update xmin.
+		 */
+		if (!TransactionIdIsValid(nonremovable_xid))
+			*can_advance_xmin = false;
+		else if (!TransactionIdIsValid(*xmin) ||
+				 TransactionIdPrecedes(nonremovable_xid, *xmin))
+			*xmin = nonremovable_xid;
+	}
+	else
+	{
+		/*
+		 * Create a replication slot to retain information (e.g., dead tuples,
+		 * commit timestamps, and origins) useful for conflict detection if
+		 * any subscription requests it.
+		 *
+		 * The slot is created before starting the apply worker to prevent it
+		 * from unnecessarily maintaining its oldest_nonremovable_xid. It is
+		 * created even for a disabled subscription to ensure information is
+		 * available for detecting conflicts during the application of remote
+		 * changes that occur before the subscription is enabled.
+		 */
+		create_conflict_slot_if_not_exists();
+
+		/*
+		 * Only collect xmin when all workers for subscriptions are running.
+		 */
+		*can_advance_xmin = false;
+	}
+}
+
+/*
+ * Create and acquire the replication slot used to retain information for
+ * conflict detection, if not yet.
+ */
+static void
+create_conflict_slot_if_not_exists(void)
+{
+	TransactionId xmin_horizon;
+
+	/* Exit early if the replication slot is already created and acquired */
+	if (MyReplicationSlot)
+		return;
+
+	/* If the replication slot exists, acquire it and exit */
+	if (SearchNamedReplicationSlot(CONFLICT_DETECTION_SLOT, true))
+	{
+		ReplicationSlotAcquire(CONFLICT_DETECTION_SLOT, true, false);
+		return;
+	}
+
+	ReplicationSlotCreate(CONFLICT_DETECTION_SLOT, false,
+						  RS_PERSISTENT, false, false, false);
+
+	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+
+	xmin_horizon = GetOldestSafeDecodingTransactionId(false);
+
+	SpinLockAcquire(&MyReplicationSlot->mutex);
+	MyReplicationSlot->effective_xmin = xmin_horizon;
+	MyReplicationSlot->data.xmin = xmin_horizon;
+	SpinLockRelease(&MyReplicationSlot->mutex);
+
+	ReplicationSlotsComputeRequiredXmin(true);
+
+	LWLockRelease(ProcArrayLock);
+
+	/* Write this slot to disk */
+	ReplicationSlotMarkDirty();
+	ReplicationSlotSave();
+
+	conflict_slot_dropped = false;
+}
+
+/*
+ * Attempt to advance the xmin value of the replication slot used to retain
+ * dead tuples for conflict detection.
+ */
+static void
+advance_conflict_slot_xmin(TransactionId new_xmin)
+{
+	Assert(MyReplicationSlot);
+	Assert(TransactionIdIsValid(new_xmin));
+	Assert(TransactionIdPrecedesOrEquals(MyReplicationSlot->data.xmin,
+										 new_xmin));
+
+	SpinLockAcquire(&MyReplicationSlot->mutex);
+	MyReplicationSlot->data.xmin = new_xmin;
+	SpinLockRelease(&MyReplicationSlot->mutex);
+
+	/* first write new xmin to disk, so we know what's up after a crash */
+
+	ReplicationSlotMarkDirty();
+	ReplicationSlotSave();
+	elog(DEBUG1, "updated xmin: %u", MyReplicationSlot->data.xmin);
+
+	/*
+	 * Now the new xmin is safely on disk, we can let the global value
+	 * advance. We do not take ProcArrayLock or similar since we only advance
+	 * xmin here and there's not much harm done by a concurrent computation
+	 * missing that.
+	 */
+	SpinLockAcquire(&MyReplicationSlot->mutex);
+	MyReplicationSlot->effective_xmin = MyReplicationSlot->data.xmin;
+	SpinLockRelease(&MyReplicationSlot->mutex);
+
+	ReplicationSlotsComputeRequiredXmin(false);
+
+	return;
+}
+
+/*
+ * Drop the replication slot used to retain dead tuples for conflict detection,
+ * if it exists.
+ */
+static void
+drop_conflict_slot_if_exists(void)
+{
+	/*
+	 * Avoid the overhead of scanning shared memory for a replication slot
+	 * that is known to have been dropped.
+	 */
+	if (conflict_slot_dropped)
+		return;
+
+	if (MyReplicationSlot)
+		ReplicationSlotDropAcquired();
+	else if (SearchNamedReplicationSlot(CONFLICT_DETECTION_SLOT, true))
+		ReplicationSlotDrop(CONFLICT_DETECTION_SLOT, true);
+
+	conflict_slot_dropped = true;
+}
+
 /*
  * Is current process the logical replication launcher?
  */
diff --git a/src/backend/replication/logical/reorderbuffer.c b/src/backend/replication/logical/reorderbuffer.c
index 67655111875..85239f6c316 100644
--- a/src/backend/replication/logical/reorderbuffer.c
+++ b/src/backend/replication/logical/reorderbuffer.c
@@ -4787,7 +4787,7 @@ StartupReorderBuffer(void)
 			continue;
 
 		/* if it cannot be a slot, skip the directory */
-		if (!ReplicationSlotValidateName(logical_de->d_name, DEBUG2))
+		if (!ReplicationSlotValidateName(logical_de->d_name, true, DEBUG2))
 			continue;
 
 		/*
diff --git a/src/backend/replication/logical/worker.c b/src/backend/replication/logical/worker.c
index 4cd9efb8c94..c864f7f52b0 100644
--- a/src/backend/replication/logical/worker.c
+++ b/src/backend/replication/logical/worker.c
@@ -4386,6 +4386,9 @@ wait_for_local_flush(RetainConflictInfoData *rci_data)
 		 LSN_FORMAT_ARGS(rci_data->remote_lsn),
 		 rci_data->candidate_xid);
 
+	/* Notify launcher to update the xmin of the conflict slot */
+	ApplyLauncherWakeup();
+
 	/*
 	 * Reset all data fields except those used to determine the timing for the
 	 * next round of transaction ID advancement. We can even use
diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c
index 600b87fa9cb..41adf71c4c8 100644
--- a/src/backend/replication/slot.c
+++ b/src/backend/replication/slot.c
@@ -47,6 +47,7 @@
 #include "miscadmin.h"
 #include "pgstat.h"
 #include "postmaster/interrupt.h"
+#include "replication/logicallauncher.h"
 #include "replication/slotsync.h"
 #include "replication/slot.h"
 #include "replication/walsender_private.h"
@@ -172,6 +173,7 @@ static SyncStandbySlotsConfigData *synchronized_standby_slots_config;
 static XLogRecPtr ss_oldest_flush_lsn = InvalidXLogRecPtr;
 
 static void ReplicationSlotShmemExit(int code, Datum arg);
+static bool IsReservedSlotName(const char *name);
 static void ReplicationSlotDropPtr(ReplicationSlot *slot);
 
 /* internal persistency functions */
@@ -258,13 +260,17 @@ ReplicationSlotShmemExit(int code, Datum arg)
 /*
  * Check whether the passed slot name is valid and report errors at elevel.
  *
+ * An error will be reported for a reserved replication slot name if
+ * allow_reserved_name is set to false.
+ *
  * Slot names may consist out of [a-z0-9_]{1,NAMEDATALEN-1} which should allow
  * the name to be used as a directory name on every supported OS.
  *
  * Returns whether the directory name is valid or not if elevel < ERROR.
  */
 bool
-ReplicationSlotValidateName(const char *name, int elevel)
+ReplicationSlotValidateName(const char *name, bool allow_reserved_name,
+							int elevel)
 {
 	const char *cp;
 
@@ -300,9 +306,31 @@ ReplicationSlotValidateName(const char *name, int elevel)
 			return false;
 		}
 	}
+
+	if (!allow_reserved_name && IsReservedSlotName(name))
+	{
+		ereport(elevel,
+				errcode(ERRCODE_RESERVED_NAME),
+				errmsg("replication slot name \"%s\" is reserved",
+					   name),
+				errdetail("The name \"%s\" is reserved for the conflict detection slot.",
+						  CONFLICT_DETECTION_SLOT));
+
+		return false;
+	}
+
 	return true;
 }
 
+/*
+ * Return true if the replication slot name is "pg_conflict_detection".
+ */
+static bool
+IsReservedSlotName(const char *name)
+{
+	return (strcmp(name, CONFLICT_DETECTION_SLOT) == 0);
+}
+
 /*
  * Create a new replication slot and mark it as used by this backend.
  *
@@ -330,7 +358,11 @@ ReplicationSlotCreate(const char *name, bool db_specific,
 
 	Assert(MyReplicationSlot == NULL);
 
-	ReplicationSlotValidateName(name, ERROR);
+	/*
+	 * The logical launcher might be creating an internal slot, so using a
+	 * reserved name is allowed in this case.
+	 */
+	ReplicationSlotValidateName(name, IsLogicalLauncher(), ERROR);
 
 	if (failover)
 	{
diff --git a/src/include/replication/logicallauncher.h b/src/include/replication/logicallauncher.h
index 82b202f3305..7b29f1814db 100644
--- a/src/include/replication/logicallauncher.h
+++ b/src/include/replication/logicallauncher.h
@@ -25,6 +25,7 @@ extern void ApplyLauncherShmemInit(void);
 extern void ApplyLauncherForgetWorkerStartTime(Oid subid);
 
 extern void ApplyLauncherWakeupAtCommit(void);
+extern void ApplyLauncherWakeup(void);
 extern void AtEOXact_ApplyLauncher(bool isCommit);
 
 extern bool IsLogicalLauncher(void);
diff --git a/src/include/replication/slot.h b/src/include/replication/slot.h
index eb0b93b1114..e03e123a2ff 100644
--- a/src/include/replication/slot.h
+++ b/src/include/replication/slot.h
@@ -20,6 +20,13 @@
 /* directory to store replication slot data in */
 #define PG_REPLSLOT_DIR     "pg_replslot"
 
+/*
+ * The reserved name for a replication slot used to retain dead tuples for
+ * conflict detection in logical replication. See
+ * maybe_advance_nonremovable_xid() for detail.
+ */
+#define CONFLICT_DETECTION_SLOT "pg_conflict_detection"
+
 /*
  * Behaviour of replication slots, upon release or crash.
  *
@@ -284,7 +291,9 @@ extern void ReplicationSlotMarkDirty(void);
 
 /* misc stuff */
 extern void ReplicationSlotInitialize(void);
-extern bool ReplicationSlotValidateName(const char *name, int elevel);
+extern bool ReplicationSlotValidateName(const char *name,
+										bool allow_reserved_name,
+										int elevel);
 extern void ReplicationSlotReserveWal(void);
 extern void ReplicationSlotsComputeRequiredXmin(bool already_locked);
 extern void ReplicationSlotsComputeRequiredLSN(void);
-- 
2.30.0.windows.2