agora inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH] fix "Preserve relfilenodes" for pg_largeobject and its index
415+ messages / 3 participants
[nested] [flat]

* [PATCH] fix "Preserve relfilenodes" for pg_largeobject and its index
@ 2022-07-02 05:48  Justin Pryzby <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Justin Pryzby @ 2022-07-02 05:48 UTC (permalink / raw)

An empty table is created by initdb, and the filenode was still pointing to
that, so pg_largeobject was empty after binary upgrades.

Test like:
| ./tmp_install/usr/local/pgsql/bin/initdb --no-sync -D pg15b1.dat -k
| ./tmp_install/usr/local/pgsql/bin/postgres -D pg15b1.dat -c logging_collector=no -p 5678 -k /tmp&
| psql -h /tmp postgres -p 5678 -c '\lo_import /etc/shells' -c 'VACUUM FULL pg_largeobject'
| rm -fr pg15b2.dat && ./tmp_install/usr/local/pgsql/bin/initdb --no-sync -k -D pg15b2.dat && ./tmp_install/usr/local/pgsql/bin/pg_upgrade -d pg15b1.dat -D pg15b2.dat -b ./tmp_install/usr/local/pgsql/bin
| ./tmp_install/usr/local/pgsql/bin/postgres -D pg15b2.dat -c logging_collector=no -p 5678 -k /tmp&

See also:
9a974cbcba005256a19991203583a94b4f9a21a9
https://www.postgresql.org/message-id/20210601140949.GC22012%40momjian.us
https://www.postgresql.org/message-id/20220701231413.GI13040%40telsasoft.com
---
 src/bin/pg_dump/pg_dump.c              | 37 ++++++++------
 src/bin/pg_upgrade/relfilenumber.c     | 68 +++++++++++++++++++++++---
 src/bin/pg_upgrade/t/002_pg_upgrade.pl |  2 +
 3 files changed, 86 insertions(+), 21 deletions(-)

diff --git a/src/bin/pg_dump/pg_dump.c b/src/bin/pg_dump/pg_dump.c
index 227cdca2e2d..06d436f2f34 100644
--- a/src/bin/pg_dump/pg_dump.c
+++ b/src/bin/pg_dump/pg_dump.c
@@ -3135,7 +3135,7 @@ dumpDatabase(Archive *fout)
 
 	/*
 	 * pg_largeobject comes from the old system intact, so set its
-	 * relfrozenxids and relminmxids.
+	 * relfrozenxids, relminmxids and relfilenode.
 	 */
 	if (dopt->binary_upgrade)
 	{
@@ -3143,34 +3143,41 @@ dumpDatabase(Archive *fout)
 		PQExpBuffer loFrozenQry = createPQExpBuffer();
 		PQExpBuffer loOutQry = createPQExpBuffer();
 		int			i_relfrozenxid,
+					i_relfilenode,
+					i_oid,
 					i_relminmxid;
 
 		/*
 		 * pg_largeobject
 		 */
 		if (fout->remoteVersion >= 90300)
-			appendPQExpBuffer(loFrozenQry, "SELECT relfrozenxid, relminmxid\n"
+			appendPQExpBuffer(loFrozenQry, "SELECT relfrozenxid, relminmxid, relfilenode, oid\n"
 							  "FROM pg_catalog.pg_class\n"
-							  "WHERE oid = %u;\n",
-							  LargeObjectRelationId);
+							  "WHERE oid IN (%u, %u);\n",
+							  LargeObjectRelationId, LargeObjectLOidPNIndexId);
 		else
-			appendPQExpBuffer(loFrozenQry, "SELECT relfrozenxid, 0 AS relminmxid\n"
+			appendPQExpBuffer(loFrozenQry, "SELECT relfrozenxid, 0 AS relminmxid, relfilenode, oid\n"
 							  "FROM pg_catalog.pg_class\n"
-							  "WHERE oid = %u;\n",
-							  LargeObjectRelationId);
+							  "WHERE oid IN (%u, %u);\n",
+							  LargeObjectRelationId, LargeObjectLOidPNIndexId);
 
-		lo_res = ExecuteSqlQueryForSingleRow(fout, loFrozenQry->data);
+		lo_res = ExecuteSqlQuery(fout, loFrozenQry->data, PGRES_TUPLES_OK);
 
 		i_relfrozenxid = PQfnumber(lo_res, "relfrozenxid");
 		i_relminmxid = PQfnumber(lo_res, "relminmxid");
+		i_relfilenode = PQfnumber(lo_res, "relfilenode");
+		i_oid = PQfnumber(lo_res, "oid");
+
+		appendPQExpBufferStr(loOutQry, "\n-- For binary upgrade, preserve values for pg_largeobject and its index\n");
+		for (int i = 0; i < PQntuples(lo_res); ++i)
+			appendPQExpBuffer(loOutQry, "UPDATE pg_catalog.pg_class\n"
+							  "SET relfrozenxid = '%u', relminmxid = '%u', relfilenode = '%u'\n"
+							  "WHERE oid = %u;\n",
+							  atooid(PQgetvalue(lo_res, i, i_relfrozenxid)),
+							  atooid(PQgetvalue(lo_res, i, i_relminmxid)),
+							  atooid(PQgetvalue(lo_res, i, i_relfilenode)),
+							  atooid(PQgetvalue(lo_res, i, i_oid)));
 
-		appendPQExpBufferStr(loOutQry, "\n-- For binary upgrade, set pg_largeobject relfrozenxid and relminmxid\n");
-		appendPQExpBuffer(loOutQry, "UPDATE pg_catalog.pg_class\n"
-						  "SET relfrozenxid = '%u', relminmxid = '%u'\n"
-						  "WHERE oid = %u;\n",
-						  atooid(PQgetvalue(lo_res, 0, i_relfrozenxid)),
-						  atooid(PQgetvalue(lo_res, 0, i_relminmxid)),
-						  LargeObjectRelationId);
 		ArchiveEntry(fout, nilCatalogId, createDumpId(),
 					 ARCHIVE_OPTS(.tag = "pg_largeobject",
 								  .description = "pg_largeobject",
diff --git a/src/bin/pg_upgrade/relfilenumber.c b/src/bin/pg_upgrade/relfilenumber.c
index b3ad8209eca..0dd649aa0f5 100644
--- a/src/bin/pg_upgrade/relfilenumber.c
+++ b/src/bin/pg_upgrade/relfilenumber.c
@@ -13,9 +13,13 @@
 
 #include "access/transam.h"
 #include "catalog/pg_class_d.h"
+#include "catalog/pg_largeobject_d.h"
 #include "pg_upgrade.h"
 
-static void transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace);
+static void transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace,
+	bool *has_lotable, bool *has_loindex);
+static void remove_stray_largeobject_files(DbInfo *new_db, bool has_lotable,
+		bool has_loindex);
 static void transfer_relfile(FileNameMap *map, const char *suffix, bool vm_must_add_frozenbit);
 
 
@@ -98,6 +102,8 @@ transfer_all_new_dbs(DbInfoArr *old_db_arr, DbInfoArr *new_db_arr,
 				   *new_db = NULL;
 		FileNameMap *mappings;
 		int			n_maps;
+		bool		has_lotable = false;
+		bool		has_loindex = false;
 
 		/*
 		 * Advance past any databases that exist in the new cluster but not in
@@ -117,22 +123,66 @@ transfer_all_new_dbs(DbInfoArr *old_db_arr, DbInfoArr *new_db_arr,
 
 		mappings = gen_db_file_maps(old_db, new_db, &n_maps, old_pgdata,
 									new_pgdata);
-		if (n_maps)
-		{
-			transfer_single_new_db(mappings, n_maps, old_tablespace);
-		}
+		transfer_single_new_db(mappings, n_maps, old_tablespace,
+			&has_lotable, &has_loindex);
+
+		/*
+		 * Remove the original pg_largeobject files created by initdb if they
+		 * exist in the new cluster with different relfilenodes.  This avoids
+		 * leaving behind stray files unassociated with any relation.
+		 */
+		remove_stray_largeobject_files(new_db, has_lotable, has_loindex);
+
 		/* We allocate something even for n_maps == 0 */
 		pg_free(mappings);
 	}
 }
 
+/*
+ * If there's no relation in the cluster whose relfilenode is the OID of the
+ * largeobject table/index, then remove the stray file.  It ought to have a
+ * single segment and fork.
+ */
+static void
+remove_stray_largeobject_files(DbInfo *new_db, bool has_lotable,
+		bool has_loindex)
+{
+	char		prefix[MAXPGPATH];
+	char		largeobj_file[MAXPGPATH];
+
+	if (new_db->db_tablespace[0] != '\0')
+		snprintf(prefix, sizeof(prefix), "%s/%s/%u",
+				new_db->db_tablespace, new_cluster.tablespace_suffix,
+				new_db->db_oid);
+	else
+		snprintf(prefix, sizeof(prefix), "%s/base/%u",
+				new_cluster.pgdata, new_db->db_oid);
+
+	if (!has_lotable)
+	{
+		snprintf(largeobj_file, sizeof(largeobj_file), "%s/%u",
+				prefix, LargeObjectRelationId);
+		if (unlink(largeobj_file) != 0)
+			pg_fatal("Unable to remove %s.\n", largeobj_file);
+	}
+
+	if (!has_loindex)
+	{
+		snprintf(largeobj_file, sizeof(largeobj_file), "%s/%u",
+				prefix, LargeObjectLOidPNIndexId);
+		if (unlink(largeobj_file) != 0)
+			pg_fatal("Unable to remove %s.\n", largeobj_file);
+	}
+}
+
 /*
  * transfer_single_new_db()
  *
  * create links for mappings stored in "maps" array.
  */
 static void
-transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace)
+transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace,
+	bool *has_lotable, bool *has_loindex)
 {
 	int			mapnum;
 	bool		vm_must_add_frozenbit = false;
@@ -146,6 +196,12 @@ transfer_single_new_db(FileNameMap *maps, int size, char *old_tablespace)
 
 	for (mapnum = 0; mapnum < size; mapnum++)
 	{
+		/* Keep track of whether a filenode matches the OID */
+		if (maps[mapnum].relfilenumber == LargeObjectRelationId)
+			*has_lotable = true;
+		if (maps[mapnum].relfilenumber == LargeObjectLOidPNIndexId)
+			*has_loindex = true;
+
 		if (old_tablespace == NULL ||
 			strcmp(maps[mapnum].old_tablespace, old_tablespace) == 0)
 		{
diff --git a/src/bin/pg_upgrade/t/002_pg_upgrade.pl b/src/bin/pg_upgrade/t/002_pg_upgrade.pl
index 2f9b13bf0ae..9ed48c4e06a 100644
--- a/src/bin/pg_upgrade/t/002_pg_upgrade.pl
+++ b/src/bin/pg_upgrade/t/002_pg_upgrade.pl
@@ -208,6 +208,8 @@ if (defined($ENV{oldinstall}))
 	}
 }
 
+$oldnode->safe_psql("regression", "VACUUM FULL pg_largeobject;");
+
 # In a VPATH build, we'll be started in the source directory, but we want
 # to run pg_upgrade in the build directory so that any files generated finish
 # in it, like delete_old_cluster.{sh,bat}.
-- 
2.17.1


--m+utxuhC6KVTvgNz--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH v31 09/11] Add support for min/max aggregates for IVM
@ 2023-05-31 11:58  Yugo Nagata <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Yugo Nagata @ 2023-05-31 11:58 UTC (permalink / raw)

Supporting min and max is more complicated than count, sum, or avg.

For an example of min, when tuples are inserted, the current min value
in the view and the min value in the inseteted tuples are compared,
then the smaller one is used as the latest min value. On the other
hand, when tuples are deleted, if the current min value in the view
equals to the min in the deleted tuples, we need re-computation the
latest min value from base tables. Otherwise, the current value in
the view remains.
---
 src/backend/commands/createas.c |  45 +++
 src/backend/commands/matview.c  | 644 +++++++++++++++++++++++++++++++-
 2 files changed, 680 insertions(+), 9 deletions(-)

diff --git a/src/backend/commands/createas.c b/src/backend/commands/createas.c
index aa8440b4e1..7e1627ad1b 100644
--- a/src/backend/commands/createas.c
+++ b/src/backend/commands/createas.c
@@ -1309,6 +1309,51 @@ check_aggregate_supports_ivm(Oid aggfnoid)
 		case F_AVG_FLOAT8:
 		case F_AVG_INTERVAL:
 
+		/* min */
+		case F_MIN_ANYARRAY:
+		case F_MIN_INT8:
+		case F_MIN_INT4:
+		case F_MIN_INT2:
+		case F_MIN_OID:
+		case F_MIN_FLOAT4:
+		case F_MIN_FLOAT8:
+		case F_MIN_DATE:
+		case F_MIN_TIME:
+		case F_MIN_TIMETZ:
+		case F_MIN_MONEY:
+		case F_MIN_TIMESTAMP:
+		case F_MIN_TIMESTAMPTZ:
+		case F_MIN_INTERVAL:
+		case F_MIN_TEXT:
+		case F_MIN_NUMERIC:
+		case F_MIN_BPCHAR:
+		case F_MIN_TID:
+		case F_MIN_ANYENUM:
+		case F_MIN_INET:
+		case F_MIN_PG_LSN:
+
+		/* max */
+		case F_MAX_ANYARRAY:
+		case F_MAX_INT8:
+		case F_MAX_INT4:
+		case F_MAX_INT2:
+		case F_MAX_OID:
+		case F_MAX_FLOAT4:
+		case F_MAX_FLOAT8:
+		case F_MAX_DATE:
+		case F_MAX_TIME:
+		case F_MAX_TIMETZ:
+		case F_MAX_MONEY:
+		case F_MAX_TIMESTAMP:
+		case F_MAX_TIMESTAMPTZ:
+		case F_MAX_INTERVAL:
+		case F_MAX_TEXT:
+		case F_MAX_NUMERIC:
+		case F_MAX_BPCHAR:
+		case F_MAX_TID:
+		case F_MAX_ANYENUM:
+		case F_MAX_INET:
+		case F_MAX_PG_LSN:
 			return true;
 
 		default:
diff --git a/src/backend/commands/matview.c b/src/backend/commands/matview.c
index 97406b28c9..0ff5e3922b 100644
--- a/src/backend/commands/matview.c
+++ b/src/backend/commands/matview.c
@@ -69,6 +69,34 @@ typedef struct
 
 #define MV_INIT_QUERYHASHSIZE	16
 
+/* MV query type codes */
+#define MV_PLAN_RECALC			1
+#define MV_PLAN_SET_VALUE		2
+
+/*
+ * MI_QueryKey
+ *
+ * The key identifying a prepared SPI plan in our query hashtable
+ */
+typedef struct MV_QueryKey
+{
+	Oid			matview_id;	/* OID of materialized view */
+	int32		query_type;	/* query type ID, see MV_PLAN_XXX above */
+} MV_QueryKey;
+
+/*
+ * MV_QueryHashEntry
+ *
+ * Hash entry for cached plans used to maintain materialized views.
+ */
+typedef struct MV_QueryHashEntry
+{
+	MV_QueryKey key;
+	SPIPlanPtr	plan;
+	SearchPathMatcher *search_path;        /* search_path used for parsing
+											 * and planning */
+} MV_QueryHashEntry;
+
 /*
  * MV_TriggerHashEntry
  *
@@ -105,6 +133,7 @@ typedef struct MV_TriggerTable
 	TupleTableSlot *slot;		/* for checking visibility in the pre-state table */
 } MV_TriggerTable;
 
+static HTAB *mv_query_cache = NULL;
 static HTAB *mv_trigger_info = NULL;
 
 static bool in_delta_calculation = false;
@@ -165,6 +194,9 @@ static void append_set_clause_for_sum(const char *resname, StringInfo buf_old,
 static void append_set_clause_for_avg(const char *resname, StringInfo buf_old,
 						  StringInfo buf_new, StringInfo aggs_list,
 						  const char *aggtype);
+static void append_set_clause_for_minmax(const char *resname, StringInfo buf_old,
+							 StringInfo buf_new, StringInfo aggs_list,
+							 bool is_min);
 static char *get_operation_string(IvmOp op, const char *col, const char *arg1, const char *arg2,
 					 const char* count_col, const char *castType);
 static char *get_null_condition_string(IvmOp op, const char *arg1, const char *arg2,
@@ -173,17 +205,30 @@ static void apply_old_delta(const char *matviewname, const char *deltaname_old,
 				List *keys);
 static void apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 				List *keys, StringInfo aggs_list, StringInfo aggs_set,
-				const char *count_colname);
+				List *minmax_list, List *is_min_list,
+				const char *count_colname,
+				SPITupleTable **tuptable_recalc, uint64 *num_recalc);
 static void apply_new_delta(const char *matviewname, const char *deltaname_new,
 				StringInfo target_list);
 static void apply_new_delta_with_count(const char *matviewname, const char* deltaname_new,
 				List *keys, StringInfo target_list, StringInfo aggs_set,
 				const char* count_colname);
 static char *get_matching_condition_string(List *keys);
+static char *get_returning_string(List *minmax_list, List *is_min_list, List *keys);
+static char *get_minmax_recalc_condition_string(List *minmax_list, List *is_min_list);
+static char *get_select_for_recalc_string(List *keys);
+static void recalc_and_set_values(SPITupleTable *tuptable_recalc, int64 num_tuples,
+					  List *namelist, List *keys, Relation matviewRel);
+static SPIPlanPtr get_plan_for_recalc(Oid matviewOid, List *namelist, List *keys, Oid *keyTypes);
+static SPIPlanPtr get_plan_for_set_values(Oid matviewOid, char *matviewname, List *namelist,
+						Oid *valTypes);
 static void generate_equal(StringInfo querybuf, Oid opttype,
 			   const char *leftop, const char *rightop);
 
 static void mv_InitHashTables(void);
+static SPIPlanPtr mv_FetchPreparedPlan(MV_QueryKey *key);
+static void mv_HashPreparedPlan(MV_QueryKey *key, SPIPlanPtr plan);
+static void mv_BuildQueryKey(MV_QueryKey *key, Oid matview_id, int32 query_type);
 static void clean_up_IVM_hash_entry(MV_TriggerHashEntry *entry, bool is_abort);
 
 /*
@@ -2122,6 +2167,8 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 	ListCell	*lc;
 	int			i;
 	List	   *keys = NIL;
+	List	   *minmax_list = NIL;
+	List	   *is_min_list = NIL;
 
 
 	/*
@@ -2203,6 +2250,17 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 				append_set_clause_for_avg(resname, aggs_set_old, aggs_set_new, aggs_list_buf,
 										  format_type_be(aggref->aggtype));
 
+			/* min/max */
+			else if (!strcmp(aggname, "min") || !strcmp(aggname, "max"))
+			{
+				bool	is_min = (!strcmp(aggname, "min"));
+
+				append_set_clause_for_minmax(resname, aggs_set_old, aggs_set_new, aggs_list_buf, is_min);
+
+				/* make a resname list of min and max aggregates */
+				minmax_list = lappend(minmax_list, resname);
+				is_min_list = lappend_int(is_min_list, is_min);
+			}
 			else
 				elog(ERROR, "unsupported aggregate function: %s", aggname);
 		}
@@ -2232,6 +2290,8 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 	if (old_tuplestores && tuplestore_tuple_count(old_tuplestores) > 0)
 	{
 		EphemeralNamedRelation enr = palloc(sizeof(EphemeralNamedRelationData));
+		SPITupleTable  *tuptable_recalc = NULL;
+		uint64			num_recalc;
 		int				rc;
 
 		/* convert tuplestores to ENR, and register for SPI */
@@ -2250,10 +2310,18 @@ apply_delta(Oid matviewOid, Tuplestorestate *old_tuplestores, Tuplestorestate *n
 			/* apply old delta and get rows to be recalculated */
 			apply_old_delta_with_count(matviewname, OLD_DELTA_ENRNAME,
 									   keys, aggs_list_buf, aggs_set_old,
-									   count_colname);
+									   minmax_list, is_min_list,
+									   count_colname, &tuptable_recalc, &num_recalc);
 		else
 			apply_old_delta(matviewname, OLD_DELTA_ENRNAME, keys);
 
+		/*
+		 * If we have min or max, we might have to recalculate aggregate values from base tables
+		 * on some tuples. TIDs and keys such tuples are returned as a result of the above query.
+		 */
+		if (minmax_list && tuptable_recalc)
+			recalc_and_set_values(tuptable_recalc, num_recalc, minmax_list, keys, matviewRel);
+
 	}
 	/* For tuple insertion */
 	if (new_tuplestores && tuplestore_tuple_count(new_tuplestores) > 0)
@@ -2445,6 +2513,70 @@ append_set_clause_for_avg(const char *resname, StringInfo buf_old,
 	);
 }
 
+/*
+ * append_set_clause_for_minmax
+ *
+ * Append SET clause string for min or max aggregation to given buffers.
+ * Also, append resnames required for calculating the aggregate value.
+ * is_min is true if this is min, false if not.
+ */
+static void
+append_set_clause_for_minmax(const char *resname, StringInfo buf_old,
+							 StringInfo buf_new, StringInfo aggs_list,
+							 bool is_min)
+{
+	char *count_col = IVM_colname("count", resname);
+
+	/* For tuple deletion */
+	if (buf_old)
+	{
+		/*
+		 * If the new value doesn't became NULL then use the value remaining
+		 * in the view although this will be recomputated afterwords.
+		 */
+		appendStringInfo(buf_old,
+			", %s = CASE WHEN %s THEN NULL ELSE %s END",
+			quote_qualified_identifier(NULL, resname),
+			get_null_condition_string(IVM_SUB, "mv", "t", count_col),
+			quote_qualified_identifier("mv", resname)
+		);
+		/* count = mv.count - t.count */
+		appendStringInfo(buf_old,
+			", %s = %s",
+			quote_qualified_identifier(NULL, count_col),
+			get_operation_string(IVM_SUB, count_col, "mv", "t", NULL, NULL)
+		);
+	}
+	/* For tuple insertion */
+	if (buf_new)
+	{
+		/*
+		 * min = LEAST(mv.min, diff.min)
+		 * max = GREATEST(mv.max, diff.max)
+		 */
+		appendStringInfo(buf_new,
+			", %s = CASE WHEN %s THEN NULL ELSE %s(%s,%s) END",
+			quote_qualified_identifier(NULL, resname),
+			get_null_condition_string(IVM_ADD, "mv", "diff", count_col),
+
+			is_min ? "LEAST" : "GREATEST",
+			quote_qualified_identifier("mv", resname),
+			quote_qualified_identifier("diff", resname)
+		);
+		/* count = mv.count + diff.count */
+		appendStringInfo(buf_new,
+			", %s = %s",
+			quote_qualified_identifier(NULL, count_col),
+			get_operation_string(IVM_ADD, count_col, "mv", "diff", NULL, NULL)
+		);
+	}
+
+	appendStringInfo(aggs_list, ", %s, %s",
+		quote_qualified_identifier("diff", resname),
+		quote_qualified_identifier("diff", IVM_colname("count", resname))
+	);
+}
+
 /*
  * get_operation_string
  *
@@ -2547,19 +2679,44 @@ get_null_condition_string(IvmOp op, const char *arg1, const char *arg2,
  * list to identify a tuple in the view. If the view has aggregates, this
  * requires strings representing resnames of aggregates and SET clause for
  * updating aggregate values.
+ *
+ * If the view has min or max aggregate, this requires a list of resnames of
+ * min/max aggregates and a list of boolean which represents which entries in
+ * minmax_list is min. These are necessary to check if we need to recalculate
+ * min or max aggregate values. In this case, this query returns TID and keys
+ * of tuples which need to be recalculated.  This result and the number of rows
+ * are stored in tuptables and num_recalc repectedly.
+ *
  */
 static void
 apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 				List *keys, StringInfo aggs_list, StringInfo aggs_set,
-				const char *count_colname)
+				List *minmax_list, List *is_min_list,
+				const char *count_colname,
+				SPITupleTable **tuptable_recalc, uint64 *num_recalc)
 {
 	StringInfoData	querybuf;
 	char   *match_cond;
+	char   *updt_returning = "";
+	char   *select_for_recalc = "SELECT";
 	bool	agg_without_groupby = (list_length(keys) == 0);
 
+	Assert(tuptable_recalc != NULL);
+	Assert(num_recalc != NULL);
+
 	/* build WHERE condition for searching tuples to be deleted */
 	match_cond = get_matching_condition_string(keys);
 
+	/*
+	 * We need a special RETURNING clause and SELECT statement for min/max to
+	 * check which tuple needs re-calculation from base tables.
+	 */
+	if (minmax_list)
+	{
+		updt_returning = get_returning_string(minmax_list, is_min_list, keys);
+		select_for_recalc = get_select_for_recalc_string(keys);
+	}
+
 	/* Search for matching tuples from the view and update or delete if found. */
 	initStringInfo(&querybuf);
 	appendStringInfo(&querybuf,
@@ -2574,10 +2731,11 @@ apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 						"UPDATE %s AS mv SET %s = mv.%s OPERATOR(pg_catalog.-) t.%s "
 											"%s"	/* SET clauses for aggregates */
 						"FROM t WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND NOT for_dlt "
-					")"
-					/* delete a tuple if this is to be deleted */
-					"DELETE FROM %s AS mv USING t "
-					"WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND for_dlt",
+						"%s"						/* RETURNING clause for recalc infomation */
+					"), dlt AS ("			/* delete a tuple if this is to be deleted */
+						"DELETE FROM %s AS mv USING t "
+						"WHERE mv.ctid OPERATOR(pg_catalog.=) t.ctid AND for_dlt"
+					") %s",							/* SELECT returning which tuples need to be recalculated */
 					count_colname,
 					count_colname, count_colname, (agg_without_groupby ? "false" : "true"),
 					(aggs_list != NULL ? aggs_list->data : ""),
@@ -2585,10 +2743,25 @@ apply_old_delta_with_count(const char *matviewname, const char *deltaname_old,
 					match_cond,
 					matviewname, count_colname, count_colname, count_colname,
 					(aggs_set != NULL ? aggs_set->data : ""),
-					matviewname);
+					updt_returning,
+					matviewname,
+					select_for_recalc);
 
-	if (SPI_exec(querybuf.data, 0) != SPI_OK_DELETE)
+	if (SPI_exec(querybuf.data, 0) != SPI_OK_SELECT)
 		elog(ERROR, "SPI_exec failed: %s", querybuf.data);
+
+
+	/* Return tuples to be recalculated. */
+	if (minmax_list)
+	{
+		*tuptable_recalc = SPI_tuptable;
+		*num_recalc = SPI_processed;
+	}
+	else
+	{
+		*tuptable_recalc = NULL;
+		*num_recalc = 0;
+	}
 }
 
 /*
@@ -2771,6 +2944,349 @@ get_matching_condition_string(List *keys)
 	return match_cond.data;
 }
 
+/*
+ * get_returning_string
+ *
+ * Build a string for RETURNING clause of UPDATE used in apply_old_delta_with_count.
+ * This clause returns ctid and a boolean value that indicates if we need to
+ * recalculate min or max value, for each updated row.
+ */
+static char *
+get_returning_string(List *minmax_list, List *is_min_list, List *keys)
+{
+	StringInfoData returning;
+	char		*recalc_cond;
+	ListCell	*lc;
+
+	Assert(minmax_list != NIL && is_min_list != NIL);
+	recalc_cond = get_minmax_recalc_condition_string(minmax_list, is_min_list);
+
+	initStringInfo(&returning);
+
+	appendStringInfo(&returning, "RETURNING mv.ctid AS tid, (%s) AS recalc", recalc_cond);
+	foreach (lc, keys)
+	{
+		Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+		char *resname = NameStr(attr->attname);
+		appendStringInfo(&returning, ", %s", quote_qualified_identifier("mv", resname));
+	}
+
+	return returning.data;
+}
+
+/*
+ * get_minmax_recalc_condition_string
+ *
+ * Build a predicate string for checking if any min/max aggregate
+ * value needs to be recalculated.
+ */
+static char *
+get_minmax_recalc_condition_string(List *minmax_list, List *is_min_list)
+{
+	StringInfoData recalc_cond;
+	ListCell	*lc1, *lc2;
+
+	initStringInfo(&recalc_cond);
+
+	Assert (list_length(minmax_list) == list_length(is_min_list));
+
+	forboth (lc1, minmax_list, lc2, is_min_list)
+	{
+		char   *resname = (char *) lfirst(lc1);
+		bool	is_min = (bool) lfirst_int(lc2);
+		char   *op_str = (is_min ? ">=" : "<=");
+
+		appendStringInfo(&recalc_cond, "%s OPERATOR(pg_catalog.%s) %s",
+			quote_qualified_identifier("mv", resname),
+			op_str,
+			quote_qualified_identifier("t", resname)
+		);
+
+		if (lnext(minmax_list, lc1))
+			appendStringInfo(&recalc_cond, " OR ");
+	}
+
+	return recalc_cond.data;
+}
+
+/*
+ * get_select_for_recalc_string
+ *
+ * Build a query to return tid and keys of tuples which need
+ * recalculation. This is used as the result of the query
+ * built by apply_old_delta.
+ */
+static char *
+get_select_for_recalc_string(List *keys)
+{
+	StringInfoData qry;
+	ListCell	*lc;
+
+	initStringInfo(&qry);
+
+	appendStringInfo(&qry, "SELECT tid");
+	foreach (lc, keys)
+	{
+		Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+		appendStringInfo(&qry, ", %s", NameStr(attr->attname));
+	}
+
+	appendStringInfo(&qry, " FROM updt WHERE recalc");
+
+	return qry.data;
+}
+
+/*
+ * recalc_and_set_values
+ *
+ * Recalculate tuples in a materialized from base tables and update these.
+ * The tuples which needs recalculation are specified by keys, and resnames
+ * of columns to be updated are specified by namelist. TIDs and key values
+ * are given by tuples in tuptable_recalc. Its first attribute must be TID
+ * and key values must be following this.
+ */
+static void
+recalc_and_set_values(SPITupleTable *tuptable_recalc, int64 num_tuples,
+					  List *namelist, List *keys, Relation matviewRel)
+{
+	TupleDesc   tupdesc_recalc = tuptable_recalc->tupdesc;
+	Oid		   *keyTypes = NULL, *types = NULL;
+	char	   *keyNulls = NULL, *nulls = NULL;
+	Datum	   *keyVals = NULL, *vals = NULL;
+	int			num_vals = list_length(namelist);
+	int			num_keys = list_length(keys);
+	uint64      i;
+	Oid			matviewOid;
+	char	   *matviewname;
+
+	matviewOid = RelationGetRelid(matviewRel);
+	matviewname = quote_qualified_identifier(get_namespace_name(RelationGetNamespace(matviewRel)),
+											 RelationGetRelationName(matviewRel));
+
+	/* If we have keys, initialize arrays for them. */
+	if (keys)
+	{
+		keyTypes = palloc(sizeof(Oid) * num_keys);
+		keyNulls = palloc(sizeof(char) * num_keys);
+		keyVals = palloc(sizeof(Datum) * num_keys);
+		/* a tuple contains keys to be recalculated and ctid to be updated*/
+		Assert(tupdesc_recalc->natts == num_keys + 1);
+
+		/* Types of key attributes  */
+		for (i = 0; i < num_keys; i++)
+			keyTypes[i] = TupleDescAttr(tupdesc_recalc, i + 1)->atttypid;
+	}
+
+	/* allocate memory for all attribute names and tid */
+	types = palloc(sizeof(Oid) * (num_vals + 1));
+	nulls = palloc(sizeof(char) * (num_vals + 1));
+	vals = palloc(sizeof(Datum) * (num_vals + 1));
+
+	/* For each tuple which needs recalculation */
+	for (i = 0; i < num_tuples; i++)
+	{
+		int j;
+		bool isnull;
+		SPIPlanPtr plan;
+		SPITupleTable *tuptable_newvals;
+		TupleDesc   tupdesc_newvals;
+
+		/* Set group key values as parameters if needed. */
+		if (keys)
+		{
+			for (j = 0; j < num_keys; j++)
+			{
+				keyVals[j] = SPI_getbinval(tuptable_recalc->vals[i], tupdesc_recalc, j + 2, &isnull);
+				if (isnull)
+					keyNulls[j] = 'n';
+				else
+					keyNulls[j] = ' ';
+			}
+		}
+
+		/*
+		 * Get recalculated values from base tables. The result must be
+		 * only one tuple thich contains the new values for specified keys.
+		 */
+		plan = get_plan_for_recalc(matviewOid, namelist, keys, keyTypes);
+		if (SPI_execute_plan(plan, keyVals, keyNulls, false, 0) != SPI_OK_SELECT)
+			elog(ERROR, "SPI_execute_plan");
+		if (SPI_processed != 1)
+			elog(ERROR, "SPI_execute_plan returned zero or more than one rows");
+
+		tuptable_newvals = SPI_tuptable;
+		tupdesc_newvals = tuptable_newvals->tupdesc;
+
+		Assert(tupdesc_newvals->natts == num_vals);
+
+		/* Set the new values as parameters */
+		for (j = 0; j < tupdesc_newvals->natts; j++)
+		{
+			if (i == 0)
+				types[j] = TupleDescAttr(tupdesc_newvals, j)->atttypid;
+
+			vals[j] = SPI_getbinval(tuptable_newvals->vals[0], tupdesc_newvals, j + 1, &isnull);
+			if (isnull)
+				nulls[j] = 'n';
+			else
+				nulls[j] = ' ';
+		}
+		/* Set TID of the view tuple to be updated as a parameter */
+		types[j] = TIDOID;
+		vals[j] = SPI_getbinval(tuptable_recalc->vals[i], tupdesc_recalc, 1, &isnull);
+		nulls[j] = ' ';
+
+		/* Update the view tuple to the new values */
+		plan = get_plan_for_set_values(matviewOid, matviewname, namelist, types);
+		if (SPI_execute_plan(plan, vals, nulls, false, 0) != SPI_OK_UPDATE)
+			elog(ERROR, "SPI_execute_plan");
+	}
+}
+
+
+/*
+ * get_plan_for_recalc
+ *
+ * Create or fetch a plan for recalculating value in the view's target list
+ * from base tables using the definition query of materialized view specified
+ * by matviewOid. namelist is a list of resnames of values to be recalculated.
+ *
+ * keys is a list of keys to identify tuples to be recalculated if this is not
+ * empty. KeyTypes is an array of types of keys.
+ */
+static SPIPlanPtr
+get_plan_for_recalc(Oid matviewOid, List *namelist, List *keys, Oid *keyTypes)
+{
+	MV_QueryKey hash_key;
+	SPIPlanPtr	plan;
+
+	/* Fetch or prepare a saved plan for the recalculation */
+	mv_BuildQueryKey(&hash_key, matviewOid, MV_PLAN_RECALC);
+	if ((plan = mv_FetchPreparedPlan(&hash_key)) == NULL)
+	{
+		ListCell	   *lc;
+		StringInfoData	str;
+		char   *viewdef;
+
+		/* get view definition of matview */
+		viewdef = text_to_cstring((text *) DatumGetPointer(
+					DirectFunctionCall1(pg_get_viewdef, ObjectIdGetDatum(matviewOid))));
+		/* get rid of trailing semi-colon */
+		viewdef[strlen(viewdef)-1] = '\0';
+
+		/*
+		 * Build a query string for recalculating values. This is like
+		 *
+		 *  SELECT x1, x2, x3, ... FROM ( ... view definition query ...) mv
+		 *   WHERE (key1, key2, ...) = ($1, $2, ...);
+		 */
+
+		initStringInfo(&str);
+		appendStringInfo(&str, "SELECT ");
+		foreach (lc, namelist)
+		{
+			appendStringInfo(&str, "%s", (char *) lfirst(lc));
+			if (lnext(namelist, lc))
+				appendStringInfoString(&str, ", ");
+		}
+		appendStringInfo(&str, " FROM (%s) mv", viewdef);
+
+		if (keys)
+		{
+			int		i = 1;
+			char	paramname[16];
+
+			appendStringInfo(&str, " WHERE (");
+			foreach (lc, keys)
+			{
+				Form_pg_attribute attr = (Form_pg_attribute) lfirst(lc);
+				char   *resname = NameStr(attr->attname);
+				Oid		typid = attr->atttypid;
+
+				sprintf(paramname, "$%d", i);
+				appendStringInfo(&str, "(");
+				generate_equal(&str, typid, resname, paramname);
+				appendStringInfo(&str, " OR (%s IS NULL AND %s IS NULL))",
+								 resname, paramname);
+
+				if (lnext(keys, lc))
+					appendStringInfoString(&str, " AND ");
+				i++;
+			}
+			appendStringInfo(&str, ")");
+		}
+		else
+			keyTypes = NULL;
+
+		plan = SPI_prepare(str.data, list_length(keys), keyTypes);
+		if (plan == NULL)
+			elog(ERROR, "SPI_prepare returned %s for %s", SPI_result_code_string(SPI_result), str.data);
+
+		SPI_keepplan(plan);
+		mv_HashPreparedPlan(&hash_key, plan);
+	}
+
+	return plan;
+}
+
+/*
+ * get_plan_for_set_values
+ *
+ * Create or fetch a plan for applying new values calculated by
+ * get_plan_for_recalc to a materialized view specified by matviewOid.
+ * matviewname is the name of the view.  namelist is a list of resnames
+ * of attributes to be updated, and valTypes is an array of types of the
+ * values.
+ */
+static SPIPlanPtr
+get_plan_for_set_values(Oid matviewOid, char *matviewname, List *namelist,
+						Oid *valTypes)
+{
+	MV_QueryKey	key;
+	SPIPlanPtr	plan;
+
+	/* Fetch or prepare a saved plan for the real check */
+	mv_BuildQueryKey(&key, matviewOid, MV_PLAN_SET_VALUE);
+	if ((plan = mv_FetchPreparedPlan(&key)) == NULL)
+	{
+		ListCell	  *lc;
+		StringInfoData str;
+		int		i;
+
+		/*
+		 * Build a query string for applying min/max values. This is like
+		 *
+		 *  UPDATE matviewname AS mv
+		 *   SET (x1, x2, x3, x4) = ($1, $2, $3, $4)
+		 *   WHERE ctid = $5;
+		 */
+
+		initStringInfo(&str);
+		appendStringInfo(&str, "UPDATE %s AS mv SET (", matviewname);
+		foreach (lc, namelist)
+		{
+			appendStringInfo(&str, "%s", (char *) lfirst(lc));
+			if (lnext(namelist, lc))
+				appendStringInfoString(&str, ", ");
+		}
+		appendStringInfo(&str, ") = ROW(");
+
+		for (i = 1; i <= list_length(namelist); i++)
+			appendStringInfo(&str, "%s$%d", (i==1 ? "" : ", "), i);
+
+		appendStringInfo(&str, ") WHERE ctid OPERATOR(pg_catalog.=) $%d", i);
+
+		plan = SPI_prepare(str.data, list_length(namelist) + 1, valTypes);
+		if (plan == NULL)
+			elog(ERROR, "SPI_prepare returned %s for %s", SPI_result_code_string(SPI_result), str.data);
+
+		SPI_keepplan(plan);
+		mv_HashPreparedPlan(&key, plan);
+	}
+
+	return plan;
+}
+
 /*
  * generate_equals
  *
@@ -2804,6 +3320,13 @@ mv_InitHashTables(void)
 {
 	HASHCTL		ctl;
 
+	memset(&ctl, 0, sizeof(ctl));
+	ctl.keysize = sizeof(MV_QueryKey);
+	ctl.entrysize = sizeof(MV_QueryHashEntry);
+	mv_query_cache = hash_create("MV query cache",
+								 MV_INIT_QUERYHASHSIZE,
+								 &ctl, HASH_ELEM | HASH_BLOBS);
+
 	memset(&ctl, 0, sizeof(ctl));
 	ctl.keysize = sizeof(Oid);
 	ctl.entrysize = sizeof(MV_TriggerHashEntry);
@@ -2812,6 +3335,109 @@ mv_InitHashTables(void)
 								 &ctl, HASH_ELEM | HASH_BLOBS);
 }
 
+/*
+ * mv_FetchPreparedPlan
+ */
+static SPIPlanPtr
+mv_FetchPreparedPlan(MV_QueryKey *key)
+{
+	MV_QueryHashEntry *entry;
+	SPIPlanPtr	plan;
+
+	/*
+	 * On the first call initialize the hashtable
+	 */
+	if (!mv_query_cache)
+		mv_InitHashTables();
+
+	/*
+	 * Lookup for the key
+	 */
+	entry = (MV_QueryHashEntry *) hash_search(mv_query_cache,
+											  (void *) key,
+											  HASH_FIND, NULL);
+	if (entry == NULL)
+		return NULL;
+
+	/*
+	 * Check whether the plan is still valid.  If it isn't, we don't want to
+	 * simply rely on plancache.c to regenerate it; rather we should start
+	 * from scratch and rebuild the query text too.  This is to cover cases
+	 * such as table/column renames.  We depend on the plancache machinery to
+	 * detect possible invalidations, though.
+	 *
+	 * CAUTION: this check is only trustworthy if the caller has already
+	 * locked both materialized views and base tables.
+	 *
+	 * Also, check whether the search_path is still the same as when we made it.
+	 * If it isn't, we need to rebuild the query text because the result of
+	 * pg_ivm_get_viewdef() will change.
+	 */
+	plan = entry->plan;
+	if (plan && SPI_plan_is_valid(plan) &&
+		SearchPathMatchesCurrentEnvironment(entry->search_path))
+		return plan;
+
+	/*
+	 * Otherwise we might as well flush the cached plan now, to free a little
+	 * memory space before we make a new one.
+	 */
+	if (plan)
+		SPI_freeplan(plan);
+	if (entry->search_path)
+		pfree(entry->search_path);
+
+	entry->plan = NULL;
+	entry->search_path = NULL;
+
+	return NULL;
+}
+
+/*
+ * mv_HashPreparedPlan
+ *
+ * Add another plan to our private SPI query plan hashtable.
+ */
+static void
+mv_HashPreparedPlan(MV_QueryKey *key, SPIPlanPtr plan)
+{
+	MV_QueryHashEntry *entry;
+	bool		found;
+
+	/*
+	 * On the first call initialize the hashtable
+	 */
+	if (!mv_query_cache)
+		mv_InitHashTables();
+
+	/*
+	 * Add the new plan.  We might be overwriting an entry previously found
+	 * invalid by mv_FetchPreparedPlan.
+	 */
+	entry = (MV_QueryHashEntry *) hash_search(mv_query_cache,
+											  (void *) key,
+											  HASH_ENTER, &found);
+	Assert(!found || entry->plan == NULL);
+	entry->plan = plan;
+	entry->search_path = GetSearchPathMatcher(TopMemoryContext);
+}
+
+/*
+ * mv_BuildQueryKey
+ *
+ * Construct a hashtable key for a prepared SPI plan for IVM.
+ */
+static void
+mv_BuildQueryKey(MV_QueryKey *key, Oid matview_id, int32 query_type)
+{
+	/*
+	 * We assume struct MV_QueryKey contains no padding bytes, else we'd need
+	 * to use memset to clear them.
+	 */
+	key->matview_id = matview_id;
+	key->query_type = query_type;
+}
+
 /*
  * AtAbort_IVM
  *
-- 
2.25.1


--Multipart=_Fri__29_Mar_2024_23_47_00_+0900_KGpmmDOIs1266Ib1
Content-Type: text/x-diff;
 name="v31-0010-Add-regression-tests-for-Incremental-View-Mainte.patch"
Content-Disposition: attachment;
 filename="v31-0010-Add-regression-tests-for-Incremental-View-Mainte.patch"
Content-Transfer-Encoding: 7bit



^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread

* [PATCH 5/5] Serialize decoded tuples without flattening.
@ 2026-02-27 18:01  Antonin Houska <[email protected]>
  0 siblings, 0 replies; 415+ messages in thread

From: Antonin Houska @ 2026-02-27 18:01 UTC (permalink / raw)

So far we used to flatten the tuples produced by logical decoding in order to
write them to a temporary file. That means that all the TOASTed attributes
ended up in a single chunk of memory. The problem is that even if individual
attributes do fit into the maximum chunk size (currently 1 GB), the whole
tuple is not guaranteed to fit.

The flattening might be a problem in the future anyway, since there's an
effort to restrict the impact of REPACK (CONCURRENTLY) on the VACUUM xmin
horizon. However, as the flattening assumes scans of the TOAST relaation, the
less we restrict VACUUM, the more likely it is that the TOAST attributes
needed for the flattening won't be available at decoding time.

(This patch also removes unnecessary field 'tupdesc' from RepackDecodingState
and allocates it in the appropriate memory context.)
---
 src/backend/commands/cluster.c                | 260 +++++++++++++++---
 .../pgoutput_repack/pgoutput_repack.c         | 114 ++++++--
 src/include/commands/cluster.h                |   5 +-
 .../expected/repack_toast.out                 |   1 +
 .../injection_points/specs/repack_toast.spec  |   7 +
 5 files changed, 318 insertions(+), 69 deletions(-)

diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index db710259eeb..e4de410e733 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -35,6 +35,7 @@
 #include "postgres.h"
 
 #include "access/amapi.h"
+#include "access/detoast.h"
 #include "access/heapam.h"
 #include "access/multixact.h"
 #include "access/relscan.h"
@@ -285,6 +286,10 @@ static void apply_concurrent_update(Relation rel, HeapTuple tup,
 									IndexInsertState *iistate,
 									TupleTableSlot *index_slot);
 static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
+static HeapTuple restore_tuple(BufFile *file, Relation relation,
+							   MemoryContext cxt);
+static HeapTuple adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+									   HeapTuple tup_src, MemoryContext cxt);
 static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
 								   HeapTuple tup_key,
 								   TupleTableSlot *ident_slot);
@@ -2574,6 +2579,7 @@ setup_logical_decoding(Oid relid)
 	LogicalDecodingContext *ctx;
 	NameData	slotname;
 	RepackDecodingState *dstate;
+	MemoryContext oldcxt;
 
 	/*
 	 * REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2624,14 +2630,6 @@ setup_logical_decoding(Oid relid)
 	 */
 	ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
 
-	/*
-	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
-	 * WAL.
-	 */
-	ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
-													   sizeof(ReadLocalXLogPageNoWaitPrivate));
-
-
 	/* Some WAL records should have been read. */
 	Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
 
@@ -2642,18 +2640,23 @@ setup_logical_decoding(Oid relid)
 	XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
 				wal_segment_size);
 
-	dstate = palloc0_object(RepackDecodingState);
-	dstate->relid = relid;
+	/* Our private state belongs to the decoding context. */
+	oldcxt = MemoryContextSwitchTo(ctx->context);
 
 	/*
-	 * Tuple descriptor may be needed to flatten a tuple before we write it to
-	 * a file. A copy is needed because the decoding worker invalidates system
-	 * caches before it starts to do the actual work.
+	 * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+	 * WAL.
 	 */
-	rel = table_open(relid, AccessShareLock);
-	dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
+	ctx->reader->private_data = palloc0_object(ReadLocalXLogPageNoWaitPrivate);
+	dstate = palloc0_object(RepackDecodingState);
+	MemoryContextSwitchTo(oldcxt);
+
+#ifdef	USE_ASSERT_CHECKING
+	dstate->relid = relid;
+#endif
 
 	/* Avoid logical decoding of other relations. */
+	rel = table_open(relid, AccessShareLock);
 	repacked_rel_locator = rel->rd_locator;
 	toastrelid = rel->rd_rel->reltoastrelid;
 	if (OidIsValid(toastrelid))
@@ -2822,10 +2825,10 @@ static void
 apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 {
 	char		kind;
-	uint32		t_len;
 	Relation	rel = dest->rel;
 	TupleTableSlot *index_slot,
 			   *ident_slot;
+	MemoryContext apply_cxt;
 	HeapTuple	tup_old = NULL;
 
 	/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
@@ -2835,6 +2838,14 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	/* A slot to fetch tuples from identity index. */
 	ident_slot = table_slot_create(rel, NULL);
 
+	/*
+	 * Use a separate memory context for the tuples and any memory needed to
+	 * process them.
+	 */
+	apply_cxt = AllocSetContextCreate(TopTransactionContext,
+									  "REPACK - apply",
+									  ALLOCSET_DEFAULT_SIZES);
+
 	while (true)
 	{
 		size_t		nread;
@@ -2849,14 +2860,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 			break;
 
 		/* Read the tuple. */
-		BufFileReadExact(file, &t_len, sizeof(t_len));
-		tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
-		tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
-		BufFileReadExact(file, tup->t_data, t_len);
-		tup->t_len = t_len;
-		ItemPointerSetInvalid(&tup->t_self);
-		tup->t_tableOid = RelationGetRelid(dest->rel);
-
+		tup = restore_tuple(file, rel, apply_cxt);
 		if (kind == CHANGE_UPDATE_OLD)
 		{
 			Assert(tup_old == NULL);
@@ -2868,7 +2872,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 
 			apply_concurrent_insert(rel, tup, dest->iistate, index_slot);
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
 		{
@@ -2892,18 +2896,28 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 				elog(ERROR, "failed to find target tuple");
 
 			if (kind == CHANGE_UPDATE_NEW)
+			{
+				/*
+				 * If 'tup' contains TOAST pointers, they point to the old
+				 * relation's toast. Copy the corresponding TOAST pointers for
+				 * the new relation from the existing tuple. (The fact that we
+				 * received a TOAST pointer here implies that the attribute
+				 * hasn't changed.)
+				 */
+				if (HeapTupleHasExternal(tup))
+					tup = adjust_toast_pointers(rel, tup, tup_exist,
+												apply_cxt);
+
 				apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
 										index_slot);
+			}
 			else
 				apply_concurrent_delete(rel, tup_exist);
 
 			if (tup_old != NULL)
-			{
-				pfree(tup_old);
 				tup_old = NULL;
-			}
 
-			pfree(tup);
+			MemoryContextReset(apply_cxt);
 		}
 		else
 			elog(ERROR, "unrecognized kind of change: %d", kind);
@@ -2920,6 +2934,7 @@ apply_concurrent_changes(BufFile *file, ChangeDest *dest)
 	}
 
 	/* Cleanup. */
+	MemoryContextDelete(apply_cxt);
 	ExecDropSingleTupleTableSlot(index_slot);
 	ExecDropSingleTupleTableSlot(ident_slot);
 }
@@ -3030,6 +3045,186 @@ apply_concurrent_delete(Relation rel, HeapTuple tup_target)
 	pgstat_progress_incr_param(PROGRESS_REPACK_HEAP_TUPLES_DELETED, 1);
 }
 
+/*
+ * Read tuple from file and store it in given memory context.
+ *
+ * External attributes are stored in separate memory chunks, in order to avoid
+ * exceeding MaxAllocSize - that could happen if the individual attributes are
+ * smaller than MaxAllocSize but the whole tuple is bigger.
+ */
+static HeapTuple
+restore_tuple(BufFile *file, Relation relation, MemoryContext cxt)
+{
+	uint32		t_len;
+	HeapTuple	tup;
+	MemoryContext oldcxt;
+	uint32		natt_ext;
+	List	   *attrs_ext = NIL;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	/* First, read the external attributes if there are some. */
+	BufFileReadExact(file, &natt_ext, sizeof(natt_ext));
+	for (int i = 0; i < natt_ext; i++)
+	{
+		varlena		varhdr;
+		char	   *ext_val;
+		Size		ext_val_size;
+
+		/*
+		 * Read the header into properly aligned memory before accessing it.
+		 */
+		BufFileReadExact(file, &varhdr, VARHDRSZ);
+		ext_val_size = VARSIZE_ANY(&varhdr);
+		ext_val = palloc0(ext_val_size);
+		memcpy(ext_val, &varhdr, VARHDRSZ);
+		BufFileReadExact(file, ext_val + VARHDRSZ, ext_val_size - VARHDRSZ);
+
+		attrs_ext = lappend(attrs_ext, ext_val);
+	}
+
+	/* Read the tuple. */
+	BufFileReadExact(file, &t_len, sizeof(t_len));
+	tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+	tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+	BufFileReadExact(file, tup->t_data, t_len);
+	tup->t_len = t_len;
+	ItemPointerSetInvalid(&tup->t_self);
+	tup->t_tableOid = RelationGetRelid(relation);
+
+	/*
+	 * If there are "external indirect" attributes, make the tuple reference
+	 * them.
+	 */
+	if (natt_ext > 0)
+	{
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+		ListCell   *lc;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tup, desc, attrs, isnull);
+
+		lc = list_head(attrs_ext);
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer,
+					   *new_datum;
+			varatt_indirect redirect_pointer;
+
+			/* Find the external attributes, like in store_change(). */
+			if (attr->attisdropped)
+				continue;
+			if (attr->attlen != -1)
+				continue;
+			if (isnull[i])
+				continue;
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+			if (!VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				continue;
+
+			/* Update the attribute. */
+			memset(&redirect_pointer, 0, sizeof(redirect_pointer));
+			redirect_pointer.pointer = lfirst(lc);
+
+			new_datum = (varlena *) palloc0(INDIRECT_POINTER_SIZE);
+			SET_VARTAG_EXTERNAL(new_datum, VARTAG_INDIRECT);
+			memcpy(VARDATA_EXTERNAL(new_datum), &redirect_pointer,
+				   sizeof(redirect_pointer));
+
+			attrs[i] = PointerGetDatum(new_datum);
+
+			lc = lnext(attrs_ext, lc);
+		}
+		Assert(lc == NULL);
+
+		tup = heap_form_tuple(desc, attrs, isnull);
+	}
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return tup;
+}
+
+/*
+ * Copy TOAST pointers from 'tup_src' to 'tup_dst' and return the modified
+ * 'tup_dst'. Use 'cxt' for all memory allocations.
+ */
+static HeapTuple
+adjust_toast_pointers(Relation relation, HeapTuple tup_dst,
+					  HeapTuple tup_src, MemoryContext cxt)
+{
+	TupleDesc	desc;
+	Datum	   *attrs_dst;
+	bool	   *isnull_dst;
+	Datum	   *attrs_src = NULL;
+	bool	   *isnull_src = NULL;
+	MemoryContext oldcxt;
+	HeapTuple	result;
+
+	oldcxt = MemoryContextSwitchTo(cxt);
+
+	desc = RelationGetDescr(relation);
+	attrs_dst = palloc0_array(Datum, desc->natts);
+	isnull_dst = palloc0_array(bool, desc->natts);
+
+	heap_deform_tuple(tup_dst, desc, attrs_dst, isnull_dst);
+
+	for (int i = 0; i < desc->natts; i++)
+	{
+		CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+		varlena    *varlena_dst,
+				   *varlena_src;
+
+		if (attr->attisdropped)
+			continue;
+		if (attr->attlen != -1)
+			continue;
+		if (isnull_dst[i])
+			continue;
+
+		varlena_dst = (varlena *) DatumGetPointer(attrs_dst[i]);
+		if (!VARATT_IS_EXTERNAL_ONDISK(varlena_dst))
+			continue;
+
+		/*
+		 * The first TOAST pointer? The caller can easily check if the tuple
+		 * has external attributes(s) in general, but is not supposed to check
+		 * specifically for TOAST.
+		 */
+		if (attrs_src == NULL)
+		{
+			attrs_src = palloc0_array(Datum, desc->natts);
+			isnull_src = palloc0_array(bool, desc->natts);
+
+			heap_deform_tuple(tup_src, desc, attrs_src, isnull_src);
+		}
+
+		varlena_src = (varlena *) DatumGetPointer(attrs_src[i]);
+
+		/*
+		 * Logical decoding only uses TOAST pointer for attributes that
+		 * haven't changed.
+		 */
+		Assert(!isnull_src[i]);
+		Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_src));
+
+		/* Copy the pointer. */
+		memcpy(VARDATA_EXTERNAL(varlena_dst), VARDATA_EXTERNAL(varlena_src),
+			   sizeof(varatt_external));
+	}
+	result = heap_form_tuple(desc, attrs_dst, isnull_dst);
+
+	MemoryContextSwitchTo(oldcxt);
+
+	return result;
+}
+
 /*
  * Find the tuple to be updated or deleted.
  *
@@ -3263,15 +3458,8 @@ free_index_insert_state(IndexInsertState *iistate)
 static void
 cleanup_logical_decoding(LogicalDecodingContext *ctx)
 {
-	RepackDecodingState *dstate;
-
-	dstate = (RepackDecodingState *) ctx->output_writer_private;
-
-	FreeTupleDesc(dstate->tupdesc);
 	FreeDecodingContext(ctx);
-
 	ReplicationSlotDropAcquired();
-	pfree(dstate);
 }
 
 /*
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index d0c464a98d5..8e22aeb163b 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -12,7 +12,7 @@
  */
 #include "postgres.h"
 
-#include "access/heaptoast.h"
+#include "access/detoast.h"
 #include "commands/cluster.h"
 #include "replication/snapbuild.h"
 #include "utils/memutils.h"
@@ -28,7 +28,7 @@ static void plugin_commit_txn(LogicalDecodingContext *ctx,
 							  ReorderBufferTXN *txn, XLogRecPtr commit_lsn);
 static void plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 						  Relation rel, ReorderBufferChange *change);
-static void store_change(LogicalDecodingContext *ctx,
+static void store_change(LogicalDecodingContext *ctx, Relation relation,
 						 ConcurrentChangeKind kind, HeapTuple tuple);
 
 void
@@ -97,9 +97,8 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
 
-	/* Only interested in one particular relation. */
-	if (relation->rd_id != dstate->relid)
-		return;
+	/* Changes of other relation should not have been decoded. */
+	Assert(RelationGetRelid(relation) == dstate->relid);
 
 	/* Decode entry depending on its type */
 	switch (change->action)
@@ -117,7 +116,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (newtuple == NULL)
 					elog(ERROR, "incomplete insert info.");
 
-				store_change(ctx, CHANGE_INSERT, newtuple);
+				store_change(ctx, relation, CHANGE_INSERT, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_UPDATE:
@@ -132,9 +131,9 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 					elog(ERROR, "incomplete update info.");
 
 				if (oldtuple != NULL)
-					store_change(ctx, CHANGE_UPDATE_OLD, oldtuple);
+					store_change(ctx, relation, CHANGE_UPDATE_OLD, oldtuple);
 
-				store_change(ctx, CHANGE_UPDATE_NEW, newtuple);
+				store_change(ctx, relation, CHANGE_UPDATE_NEW, newtuple);
 			}
 			break;
 		case REORDER_BUFFER_CHANGE_DELETE:
@@ -146,7 +145,7 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 				if (oldtuple == NULL)
 					elog(ERROR, "incomplete delete info.");
 
-				store_change(ctx, CHANGE_DELETE, oldtuple);
+				store_change(ctx, relation, CHANGE_DELETE, oldtuple);
 			}
 			break;
 		default:
@@ -164,39 +163,94 @@ plugin_change(LogicalDecodingContext *ctx, ReorderBufferTXN *txn,
 
 /* Store concurrent data change. */
 static void
-store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
-			 HeapTuple tuple)
+store_change(LogicalDecodingContext *ctx, Relation relation,
+			 ConcurrentChangeKind kind, HeapTuple tuple)
 {
 	RepackDecodingState *dstate;
+	BufFile    *file;
 	char		kind_byte = (char) kind;
-	bool		flattened = false;
+	List	   *attrs_ext = NIL;
+	uint32		natt_ext;
 
 	dstate = (RepackDecodingState *) ctx->output_writer_private;
+	file = dstate->file;
 
 	/* Store the change kind. */
-	BufFileWrite(dstate->file, &kind_byte, 1);
+	BufFileWrite(file, &kind_byte, 1);
 
 	/*
-	 * ReorderBufferCommit() stores the TOAST chunks in its private memory
-	 * context and frees them after having called apply_change().  Therefore
-	 * we need flat copy (including TOAST) that we eventually copy into the
-	 * memory context which is available to decode_concurrent_changes().
+	 * If the tuple contains "external indirect" attributes, we need to write
+	 * the contents to the file because we have no control over that memory.
 	 */
 	if (HeapTupleHasExternal(tuple))
 	{
-		/*
-		 * toast_flatten_tuple_to_datum() might be more convenient but we
-		 * don't want the decompression it does.
-		 */
-		tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
-		flattened = true;
+		TupleDesc	desc;
+		Datum	   *attrs;
+		bool	   *isnull;
+
+		desc = RelationGetDescr(relation);
+		attrs = palloc0_array(Datum, desc->natts);
+		isnull = palloc0_array(bool, desc->natts);
+
+		heap_deform_tuple(tuple, desc, attrs, isnull);
+
+		/* First, gather and count the "external indirect" attributes. */
+		for (int i = 0; i < desc->natts; i++)
+		{
+			CompactAttribute *attr = TupleDescCompactAttr(desc, i);
+			varlena    *varlena_pointer;
+
+			if (attr->attisdropped)
+				continue;
+
+			/* not a varlena datatype */
+			if (attr->attlen != -1)
+				continue;
+
+			/* no data */
+			if (isnull[i])
+				continue;
+
+			/* ok, we know we have a toast datum */
+			varlena_pointer = (varlena *) DatumGetPointer(attrs[i]);
+
+			if (!VARATT_IS_EXTERNAL(varlena_pointer))
+				continue;
+
+			if (VARATT_IS_EXTERNAL_INDIRECT(varlena_pointer))
+				attrs_ext = lappend(attrs_ext, varlena_pointer);
+			else
+			{
+				/*
+				 * Logical decoding should not produce "external expanded"
+				 * attributes (those actually should never appear on disk), so
+				 * only TOASTed attribute can be seen here.
+				 */
+				Assert(VARATT_IS_EXTERNAL_ONDISK(varlena_pointer));
+			}
+		}
+		natt_ext = list_length(attrs_ext);
 	}
-	/* Store the tuple size ... */
-	BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
-	/* ... and the tuple itself. */
-	BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
+	else
+		natt_ext = 0;
 
-	/* Free the flat copy if created above. */
-	if (flattened)
-		pfree(tuple);
+	/* Write the number of external attributes. */
+	BufFileWrite(file, &natt_ext, sizeof(natt_ext));
+	/* ... and the attributes themselves, if there are some. */
+	foreach_ptr(varlena, attr_val, attrs_ext)
+	{
+		varlena    *ext_val;
+		Size		ext_val_size;
+
+		ext_val = detoast_external_attr(attr_val);
+		ext_val_size = VARSIZE_ANY(ext_val);
+		BufFileWrite(file, ext_val, ext_val_size);
+		pfree(ext_val);
+	}
+	list_free(attrs_ext);
+
+	/* Finally write the tuple size ... */
+	BufFileWrite(file, &tuple->t_len, sizeof(tuple->t_len));
+	/* ... and the tuple itself. */
+	BufFileWrite(file, tuple->t_data, tuple->t_len);
 }
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 1b05d5d418b..a32813e8530 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -65,11 +65,10 @@ typedef enum
  */
 typedef struct RepackDecodingState
 {
+#ifdef	USE_ASSERT_CHECKING
 	/* The relation whose changes we're decoding. */
 	Oid			relid;
-
-	/* Tuple descriptor of the relation being processed. */
-	TupleDesc	tupdesc;
+#endif
 
 	/* The current output file. */
 	BufFile    *file;
diff --git a/src/test/modules/injection_points/expected/repack_toast.out b/src/test/modules/injection_points/expected/repack_toast.out
index 4f866a74e32..b56dde134f8 100644
--- a/src/test/modules/injection_points/expected/repack_toast.out
+++ b/src/test/modules/injection_points/expected/repack_toast.out
@@ -13,6 +13,7 @@ step change:
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 
 step check2: 
 	INSERT INTO relfilenodes(node)
diff --git a/src/test/modules/injection_points/specs/repack_toast.spec b/src/test/modules/injection_points/specs/repack_toast.spec
index b48abf21450..f84917daca2 100644
--- a/src/test/modules/injection_points/specs/repack_toast.spec
+++ b/src/test/modules/injection_points/specs/repack_toast.spec
@@ -74,10 +74,17 @@ teardown
 
 session s2
 step change
+# Separately test UPDATE of both plain ("i") and TOASTed ("j") attribute. In
+# the first case, the new tuple we get from reorderbuffer.c contains "j" as a
+# TOAST pointer, wich we need to update so it points to the new heap. In the
+# latter case, we receive "j" as "external indirect" value - here we test that
+# the decoding worker writes the tuple to a file correctly and that the
+# backend executing REPACK manages to restore it.
 {
 	UPDATE repack_test SET j=get_long_string() where i=2;
 	DELETE FROM repack_test WHERE i=3;
 	INSERT INTO repack_test(i, j) VALUES (4, get_long_string());
+	UPDATE repack_test SET i=3 where i=1;
 }
 # Check the table from the perspective of s2.
 step check2
-- 
2.47.3


--=-=-=--





^ permalink  raw  reply  [nested|flat] 415+ messages in thread