Allow citus_*_size on index related to a distributed table (#7271)

I just enhanced the existing code to check if the relation is an index
belonging to a distributed table.
If so the shardId is appended to relation (index) name and the *_size
function are executed as before.

There is a change in an extern function:
  `extern StringInfo GenerateSizeQueryOnMultiplePlacements(...)`
It's possible to create a new function and deprecate this one later if
compatibility is an issue.

Fixes https://github.com/citusdata/citus/issues/6496.

DESCRIPTION: Allows using Citus size functions on distributed tables
indexes.

---------

Co-authored-by: Onur Tirtir <onurcantirtir@gmail.com>

src/backend/distributed/metadata/metadata_utility.c

@@ -24,6 +24,7 @@
 #include "access/sysattr.h"
 #include "access/xact.h"
 #include "catalog/dependency.h"
+#include "catalog/index.h"
 #include "catalog/indexing.h"
 #include "catalog/pg_authid.h"
 #include "catalog/pg_constraint.h"
@@ -88,11 +89,11 @@ static uint64 * AllocateUint64(uint64 value);
 static void RecordDistributedRelationDependencies(Oid distributedRelationId);
 static GroupShardPlacement * TupleToGroupShardPlacement(TupleDesc tupleDesc,
 														HeapTuple heapTuple);
-static bool DistributedTableSize(Oid relationId, SizeQueryType sizeQueryType,
+static bool DistributedRelationSize(Oid relationId, SizeQueryType sizeQueryType,
-								 bool failOnError, uint64 *tableSize);
+									bool failOnError, uint64 *relationSize);
-static bool DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
+static bool DistributedRelationSizeOnWorker(WorkerNode *workerNode, Oid relationId,
-										 SizeQueryType sizeQueryType, bool failOnError,
+											SizeQueryType sizeQueryType, bool failOnError,
-										 uint64 *tableSize);
+											uint64 *relationSize);
 static List * ShardIntervalsOnWorkerGroup(WorkerNode *workerNode, Oid relationId);
 static char * GenerateShardIdNameValuesForShardList(List *shardIntervalList,
 													bool firstValue);
@@ -282,7 +283,7 @@ citus_shard_sizes(PG_FUNCTION_ARGS)
 
 
 /*
- * citus_total_relation_size accepts a table name and returns a distributed table
+ * citus_total_relation_size accepts a distributed table name and returns a distributed table
  * and its indexes' total relation size.
  */
 Datum
@@ -294,20 +295,20 @@ citus_total_relation_size(PG_FUNCTION_ARGS)
 	bool failOnError = PG_GETARG_BOOL(1);
 
 	SizeQueryType sizeQueryType = TOTAL_RELATION_SIZE;
-	uint64 tableSize = 0;
+	uint64 relationSize = 0;
 
-	if (!DistributedTableSize(relationId, sizeQueryType, failOnError, &tableSize))
+	if (!DistributedRelationSize(relationId, sizeQueryType, failOnError, &relationSize))
 	{
 		Assert(!failOnError);
 		PG_RETURN_NULL();
 	}
 
-	PG_RETURN_INT64(tableSize);
+	PG_RETURN_INT64(relationSize);
 }
 
 
 /*
- * citus_table_size accepts a table name and returns a distributed table's total
+ * citus_table_size accepts a distributed table name and returns a distributed table's total
  * relation size.
  */
 Datum
@@ -318,21 +319,24 @@ citus_table_size(PG_FUNCTION_ARGS)
 	Oid relationId = PG_GETARG_OID(0);
 	bool failOnError = true;
 	SizeQueryType sizeQueryType = TABLE_SIZE;
-	uint64 tableSize = 0;
+	uint64 relationSize = 0;
 
-	if (!DistributedTableSize(relationId, sizeQueryType, failOnError, &tableSize))
+	/* We do not check if relation is really a table, like PostgreSQL is doing. */
+	if (!DistributedRelationSize(relationId, sizeQueryType, failOnError, &relationSize))
 	{
 		Assert(!failOnError);
 		PG_RETURN_NULL();
 	}
 
-	PG_RETURN_INT64(tableSize);
+	PG_RETURN_INT64(relationSize);
 }
 
 
 /*
- * citus_relation_size accept a table name and returns a relation's 'main'
+ * citus_relation_size accept a distributed relation name and returns a relation's 'main'
  * fork's size.
+ *
+ * Input relation is allowed to be an index on a distributed table too.
  */
 Datum
 citus_relation_size(PG_FUNCTION_ARGS)
@@ -344,7 +348,7 @@ citus_relation_size(PG_FUNCTION_ARGS)
 	SizeQueryType sizeQueryType = RELATION_SIZE;
 	uint64 relationSize = 0;
 
-	if (!DistributedTableSize(relationId, sizeQueryType, failOnError, &relationSize))
+	if (!DistributedRelationSize(relationId, sizeQueryType, failOnError, &relationSize))
 	{
 		Assert(!failOnError);
 		PG_RETURN_NULL();
@@ -506,13 +510,16 @@ ReceiveShardIdAndSizeResults(List *connectionList, Tuplestorestate *tupleStore,
 
 
 /*
- * DistributedTableSize is helper function for each kind of citus size functions.
+ * DistributedRelationSize is helper function for each kind of citus size
- * It first checks whether the table is distributed and size query can be run on
+ * functions. It first checks whether the relation is a distributed table or an
- * it. Connection to each node has to be established to get the size of the table.
+ * index belonging to a distributed table and size query can be run on it.
+ * Connection to each node has to be established to get the size of the
+ * relation.
+ * Input relation is allowed to be an index on a distributed table too.
  */
 static bool
-DistributedTableSize(Oid relationId, SizeQueryType sizeQueryType, bool failOnError,
+DistributedRelationSize(Oid relationId, SizeQueryType sizeQueryType,
-					 uint64 *tableSize)
+						bool failOnError, uint64 *relationSize)
 {
 	int logLevel = WARNING;
 
@@ -538,7 +545,7 @@ DistributedTableSize(Oid relationId, SizeQueryType sizeQueryType, bool failOnErr
 	if (relation == NULL)
 	{
 		ereport(logLevel,
-				(errmsg("could not compute table size: relation does not exist")));
+				(errmsg("could not compute relation size: relation does not exist")));
 
 		return false;
 	}
@@ -553,8 +560,9 @@ DistributedTableSize(Oid relationId, SizeQueryType sizeQueryType, bool failOnErr
 	{
 		uint64 relationSizeOnNode = 0;
 
-		bool gotSize = DistributedTableSizeOnWorker(workerNode, relationId, sizeQueryType,
+		bool gotSize = DistributedRelationSizeOnWorker(workerNode, relationId,
-													failOnError, &relationSizeOnNode);
+													   sizeQueryType,
+													   failOnError, &relationSizeOnNode);
 		if (!gotSize)
 		{
 			return false;
@@ -563,21 +571,22 @@ DistributedTableSize(Oid relationId, SizeQueryType sizeQueryType, bool failOnErr
 		sumOfSizes += relationSizeOnNode;
 	}
 
-	*tableSize = sumOfSizes;
+	*relationSize = sumOfSizes;
 
 	return true;
 }
 
 
 /*
- * DistributedTableSizeOnWorker gets the workerNode and relationId to calculate
+ * DistributedRelationSizeOnWorker gets the workerNode and relationId to calculate
  * size of that relation on the given workerNode by summing up the size of each
  * shard placement.
+ * Input relation is allowed to be an index on a distributed table too.
  */
 static bool
-DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
+DistributedRelationSizeOnWorker(WorkerNode *workerNode, Oid relationId,
-							 SizeQueryType sizeQueryType,
+								SizeQueryType sizeQueryType,
-							 bool failOnError, uint64 *tableSize)
+								bool failOnError, uint64 *relationSize)
 {
 	int logLevel = WARNING;
 
@@ -591,6 +600,17 @@ DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
 	uint32 connectionFlag = 0;
 	PGresult *result = NULL;
 
+	/* if the relation is an index, update relationId and define indexId */
+	Oid indexId = InvalidOid;
+	Oid relKind = get_rel_relkind(relationId);
+	if (relKind == RELKIND_INDEX || relKind == RELKIND_PARTITIONED_INDEX)
+	{
+		indexId = relationId;
+
+		bool missingOk = false;
+		relationId = IndexGetRelation(indexId, missingOk);
+	}
+
 	List *shardIntervalsOnNode = ShardIntervalsOnWorkerGroup(workerNode, relationId);
 
 	/*
@@ -598,21 +618,22 @@ DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
 	 * But citus size functions shouldn't include them, like PG.
 	 */
 	bool optimizePartitionCalculations = false;
-	StringInfo tableSizeQuery = GenerateSizeQueryOnMultiplePlacements(
+	StringInfo relationSizeQuery = GenerateSizeQueryOnMultiplePlacements(
 		shardIntervalsOnNode,
+		indexId,
 		sizeQueryType,
 		optimizePartitionCalculations);
 
 	MultiConnection *connection = GetNodeConnection(connectionFlag, workerNodeName,
 													workerNodePort);
-	int queryResult = ExecuteOptionalRemoteCommand(connection, tableSizeQuery->data,
+	int queryResult = ExecuteOptionalRemoteCommand(connection, relationSizeQuery->data,
 												   &result);
 
 	if (queryResult != 0)
 	{
 		ereport(logLevel, (errcode(ERRCODE_CONNECTION_FAILURE),
 						   errmsg("could not connect to %s:%d to get size of "
-								  "table \"%s\"",
+								  "relation \"%s\"",
 								  workerNodeName, workerNodePort,
 								  get_rel_name(relationId))));
 
@@ -626,19 +647,19 @@ DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
 		ClearResults(connection, failOnError);
 
 		ereport(logLevel, (errcode(ERRCODE_CONNECTION_FAILURE),
-						   errmsg("cannot parse size of table \"%s\" from %s:%d",
+						   errmsg("cannot parse size of relation \"%s\" from %s:%d",
 								  get_rel_name(relationId), workerNodeName,
 								  workerNodePort)));
 
 		return false;
 	}
 
-	StringInfo tableSizeStringInfo = (StringInfo) linitial(sizeList);
+	StringInfo relationSizeStringInfo = (StringInfo) linitial(sizeList);
-	char *tableSizeString = tableSizeStringInfo->data;
+	char *relationSizeString = relationSizeStringInfo->data;
 
-	if (strlen(tableSizeString) > 0)
+	if (strlen(relationSizeString) > 0)
 	{
-		*tableSize = SafeStringToUint64(tableSizeString);
+		*relationSize = SafeStringToUint64(relationSizeString);
 	}
 	else
 	{
@@ -647,7 +668,7 @@ DistributedTableSizeOnWorker(WorkerNode *workerNode, Oid relationId,
 		 * being executed. For this case we get an empty string as table size.
 		 * We can take that as zero to prevent any unnecessary errors.
 		 */
-		*tableSize = 0;
+		*relationSize = 0;
 	}
 
 	PQclear(result);
@@ -732,7 +753,7 @@ ShardIntervalsOnWorkerGroup(WorkerNode *workerNode, Oid relationId)
 
 /*
  * GenerateSizeQueryOnMultiplePlacements generates a select size query to get
- * size of multiple tables. Note that, different size functions supported by PG
+ * size of multiple relations. Note that, different size functions supported by PG
  * are also supported by this function changing the size query type given as the
  * last parameter to function. Depending on the sizeQueryType enum parameter, the
  * generated query will call one of the functions: pg_relation_size,
@@ -740,9 +761,13 @@ ShardIntervalsOnWorkerGroup(WorkerNode *workerNode, Oid relationId)
  * This function uses UDFs named worker_partitioned_*_size for partitioned tables,
  * if the parameter optimizePartitionCalculations is true. The UDF to be called is
  * determined by the parameter sizeQueryType.
+ *
+ * indexId is provided if we're interested in the size of an index, not the whole
+ * table.
  */
 StringInfo
 GenerateSizeQueryOnMultiplePlacements(List *shardIntervalList,
+									  Oid indexId,
 									  SizeQueryType sizeQueryType,
 									  bool optimizePartitionCalculations)
 {
@@ -766,16 +791,20 @@ GenerateSizeQueryOnMultiplePlacements(List *shardIntervalList,
 			 */
 			continue;
 		}
+
+		/* we need to build the shard relation name, being an index or table */
+		Oid objectId = OidIsValid(indexId) ? indexId : shardInterval->relationId;
+
 		uint64 shardId = shardInterval->shardId;
-		Oid schemaId = get_rel_namespace(shardInterval->relationId);
+		Oid schemaId = get_rel_namespace(objectId);
 		char *schemaName = get_namespace_name(schemaId);
-		char *shardName = get_rel_name(shardInterval->relationId);
+		char *shardName = get_rel_name(objectId);
 		AppendShardIdToName(&shardName, shardId);
 
 		char *shardQualifiedName = quote_qualified_identifier(schemaName, shardName);
 		char *quotedShardName = quote_literal_cstr(shardQualifiedName);
 
-		/* for partitoned tables, we will call worker_partitioned_... size functions */
+		/* for partitioned tables, we will call worker_partitioned_... size functions */
 		if (optimizePartitionCalculations && PartitionedTable(shardInterval->relationId))
 		{
 			partitionedShardNames = lappend(partitionedShardNames, quotedShardName);
@@ -1010,7 +1039,7 @@ AppendShardIdNameValues(StringInfo selectQuery, ShardInterval *shardInterval)
 
 
 /*
- * ErrorIfNotSuitableToGetSize determines whether the table is suitable to find
+ * ErrorIfNotSuitableToGetSize determines whether the relation is suitable to find
  * its' size with internal functions.
  */
 static void
@@ -1018,11 +1047,32 @@ ErrorIfNotSuitableToGetSize(Oid relationId)
 {
 	if (!IsCitusTable(relationId))
 	{
-		char *relationName = get_rel_name(relationId);
+		Oid relKind = get_rel_relkind(relationId);
-		char *escapedQueryString = quote_literal_cstr(relationName);
+		if (relKind != RELKIND_INDEX && relKind != RELKIND_PARTITIONED_INDEX)
-		ereport(ERROR, (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
+		{
-						errmsg("cannot calculate the size because relation %s is not "
+			char *relationName = get_rel_name(relationId);
-							   "distributed", escapedQueryString)));
+			char *escapedRelationName = quote_literal_cstr(relationName);
+			ereport(ERROR, (errcode(ERRCODE_INVALID_TABLE_DEFINITION),
+							errmsg(
+								"cannot calculate the size because relation %s "
+								"is not distributed",
+								escapedRelationName)));
+		}
+		bool missingOk = false;
+		Oid indexId = relationId;
+		relationId = IndexGetRelation(relationId, missingOk);
+		if (!IsCitusTable(relationId))
+		{
+			char *tableName = get_rel_name(relationId);
+			char *escapedTableName = quote_literal_cstr(tableName);
+			char *indexName = get_rel_name(indexId);
+			char *escapedIndexName = quote_literal_cstr(indexName);
+			ereport(ERROR, (errcode(ERRCODE_INVALID_OBJECT_DEFINITION),
+							errmsg(
+								"cannot calculate the size because table %s for "
+								"index %s is not distributed",
+								escapedTableName, escapedIndexName)));
+		}
 	}
 }
 

src/backend/distributed/operations/shard_transfer.c

@@ -792,7 +792,12 @@ ShardListSizeInBytes(List *shardList, char *workerNodeName, uint32
 
 	/* we skip child tables of a partitioned table if this boolean variable is true */
 	bool optimizePartitionCalculations = true;
+
+	/* we're interested in whole table, not a particular index */
+	Oid indexId = InvalidOid;
+
 	StringInfo tableSizeQuery = GenerateSizeQueryOnMultiplePlacements(shardList,
+																	  indexId,
 																	  TOTAL_RELATION_SIZE,
 																	  optimizePartitionCalculations);
 

src/include/distributed/metadata_utility.h

@@ -342,6 +342,7 @@ extern void LookupTaskPlacementHostAndPort(ShardPlacement *taskPlacement, char *
 										   int *nodePort);
 extern bool IsDummyPlacement(ShardPlacement *taskPlacement);
 extern StringInfo GenerateSizeQueryOnMultiplePlacements(List *shardIntervalList,
+														Oid indexId,
 														SizeQueryType sizeQueryType,
 														bool optimizePartitionCalculations);
 extern List * RemoveCoordinatorPlacementIfNotSingleNode(List *placementList);

src/test/regress/citus_tests/run_test.py

@@ -175,6 +175,7 @@ DEPS = {
     ),
     "grant_on_schema_propagation": TestDeps("minimal_schedule"),
     "propagate_extension_commands": TestDeps("minimal_schedule"),
+    "multi_size_queries": TestDeps("base_schedule", ["multi_copy"]),
 }
 
 

src/test/regress/expected/isolation_drop_vs_all.out

@@ -226,7 +226,7 @@ step s1-drop: DROP TABLE drop_hash;
 step s2-table-size: SELECT citus_total_relation_size('drop_hash'); <waiting ...>
 step s1-commit: COMMIT;
 step s2-table-size: <... completed>
-ERROR:  could not compute table size: relation does not exist
+ERROR:  could not compute relation size: relation does not exist
 step s2-commit: COMMIT;
 step s1-select-count: SELECT COUNT(*) FROM drop_hash;
 ERROR:  relation "drop_hash" does not exist

src/test/regress/expected/multi_cluster_management.out

@@ -1258,3 +1258,9 @@ SELECT bool_and(hasmetadata) AND bool_and(metadatasynced) FROM pg_dist_node WHER
  t
 (1 row)
 
+-- Grant all on public schema to public
+--
+-- That's the default on Postgres versions < 15 and we want to
+-- keep permissions compatible accross versions, in regression
+-- tests.
+GRANT ALL ON SCHEMA public TO PUBLIC;

src/test/regress/expected/multi_size_queries.out

@@ -7,19 +7,25 @@
 SET citus.next_shard_id TO 1390000;
 -- Tests with invalid relation IDs
 SELECT citus_table_size(1);
-ERROR:  could not compute table size: relation does not exist
+ERROR:  could not compute relation size: relation does not exist
 SELECT citus_relation_size(1);
-ERROR:  could not compute table size: relation does not exist
+ERROR:  could not compute relation size: relation does not exist
 SELECT citus_total_relation_size(1);
-ERROR:  could not compute table size: relation does not exist
+ERROR:  could not compute relation size: relation does not exist
 -- Tests with non-distributed table
-CREATE TABLE non_distributed_table (x int);
+CREATE TABLE non_distributed_table (x int primary key);
 SELECT citus_table_size('non_distributed_table');
 ERROR:  cannot calculate the size because relation 'non_distributed_table' is not distributed
 SELECT citus_relation_size('non_distributed_table');
 ERROR:  cannot calculate the size because relation 'non_distributed_table' is not distributed
 SELECT citus_total_relation_size('non_distributed_table');
 ERROR:  cannot calculate the size because relation 'non_distributed_table' is not distributed
+SELECT citus_table_size('non_distributed_table_pkey');
+ERROR:  cannot calculate the size because table 'non_distributed_table' for index 'non_distributed_table_pkey' is not distributed
+SELECT citus_relation_size('non_distributed_table_pkey');
+ERROR:  cannot calculate the size because table 'non_distributed_table' for index 'non_distributed_table_pkey' is not distributed
+SELECT citus_total_relation_size('non_distributed_table_pkey');
+ERROR:  cannot calculate the size because table 'non_distributed_table' for index 'non_distributed_table_pkey' is not distributed
 DROP TABLE non_distributed_table;
 -- fix broken placements via disabling the node
 SET client_min_messages TO ERROR;
@@ -31,24 +37,70 @@ SELECT replicate_table_shards('lineitem_hash_part', shard_replication_factor:=2,
 
 -- Tests on distributed table with replication factor > 1
 VACUUM (FULL) lineitem_hash_part;
-SELECT citus_table_size('lineitem_hash_part');
+SELECT citus_relation_size('lineitem_hash_part') <= citus_table_size('lineitem_hash_part');
- citus_table_size
+ ?column?
 ---------------------------------------------------------------------
-          3801088
+ t
 (1 row)
 
-SELECT citus_relation_size('lineitem_hash_part');
+SELECT citus_table_size('lineitem_hash_part') <= citus_total_relation_size('lineitem_hash_part');
- citus_relation_size
+ ?column?
 ---------------------------------------------------------------------
-             3801088
+ t
 (1 row)
 
-SELECT citus_total_relation_size('lineitem_hash_part');
+SELECT citus_relation_size('lineitem_hash_part') > 0;
- citus_total_relation_size
+ ?column?
 ---------------------------------------------------------------------
-                   3801088
+ t
 (1 row)
 
+CREATE INDEX lineitem_hash_part_idx ON lineitem_hash_part(l_orderkey);
+VACUUM (FULL) lineitem_hash_part;
+SELECT citus_relation_size('lineitem_hash_part') <= citus_table_size('lineitem_hash_part');
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_table_size('lineitem_hash_part') <= citus_total_relation_size('lineitem_hash_part');
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_relation_size('lineitem_hash_part') > 0;
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_relation_size('lineitem_hash_part_idx') <= citus_table_size('lineitem_hash_part_idx');
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_table_size('lineitem_hash_part_idx') <= citus_total_relation_size('lineitem_hash_part_idx');
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_relation_size('lineitem_hash_part_idx') > 0;
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+SELECT citus_total_relation_size('lineitem_hash_part') >=
+       citus_table_size('lineitem_hash_part') + citus_table_size('lineitem_hash_part_idx');
+ ?column?
+---------------------------------------------------------------------
+ t
+(1 row)
+
+DROP INDEX lineitem_hash_part_idx;
 VACUUM (FULL) customer_copy_hash;
 -- Tests on distributed tables with streaming replication.
 SELECT citus_table_size('customer_copy_hash');
@@ -72,10 +124,10 @@ SELECT citus_total_relation_size('customer_copy_hash');
 -- Make sure we can get multiple sizes in a single query
 SELECT citus_table_size('customer_copy_hash'),
        citus_table_size('customer_copy_hash'),
-       citus_table_size('supplier');
+       citus_table_size('customer_copy_hash');
  citus_table_size | citus_table_size | citus_table_size
 ---------------------------------------------------------------------
-           548864 |           548864 |           655360
+           548864 |           548864 |           548864
 (1 row)
 
 CREATE INDEX index_1 on customer_copy_hash(c_custkey);
@@ -99,6 +151,24 @@ SELECT citus_total_relation_size('customer_copy_hash');
                    2646016
 (1 row)
 
+SELECT citus_table_size('index_1');
+ citus_table_size
+---------------------------------------------------------------------
+          1048576
+(1 row)
+
+SELECT citus_relation_size('index_1');
+ citus_relation_size
+---------------------------------------------------------------------
+             1048576
+(1 row)
+
+SELECT citus_total_relation_size('index_1');
+ citus_total_relation_size
+---------------------------------------------------------------------
+                   1048576
+(1 row)
+
 -- Tests on reference table
 VACUUM (FULL) supplier;
 SELECT citus_table_size('supplier');
@@ -139,6 +209,74 @@ SELECT citus_total_relation_size('supplier');
                     688128
 (1 row)
 
+SELECT citus_table_size('index_2');
+ citus_table_size
+---------------------------------------------------------------------
+           122880
+(1 row)
+
+SELECT citus_relation_size('index_2');
+ citus_relation_size
+---------------------------------------------------------------------
+              122880
+(1 row)
+
+SELECT citus_total_relation_size('index_2');
+ citus_total_relation_size
+---------------------------------------------------------------------
+                    122880
+(1 row)
+
+-- Test on partitioned table
+CREATE TABLE split_me (dist_col int, partition_col timestamp) PARTITION BY RANGE (partition_col);
+CREATE INDEX ON split_me(dist_col);
+-- create 2 partitions
+CREATE TABLE m PARTITION OF split_me FOR VALUES FROM ('2018-01-01') TO ('2019-01-01');
+CREATE TABLE e PARTITION OF split_me FOR VALUES FROM ('2019-01-01') TO ('2020-01-01');
+INSERT INTO split_me SELECT 1, '2018-01-01'::timestamp + i * interval '1 day' FROM generate_series(1, 360) i;
+INSERT INTO split_me SELECT 2, '2019-01-01'::timestamp + i * interval '1 day' FROM generate_series(1, 180) i;
+-- before citus
+SELECT citus_relation_size('split_me');
+ERROR:  cannot calculate the size because relation 'split_me' is not distributed
+SELECT citus_relation_size('split_me_dist_col_idx');
+ERROR:  cannot calculate the size because table 'split_me' for index 'split_me_dist_col_idx' is not distributed
+SELECT citus_relation_size('m');
+ERROR:  cannot calculate the size because relation 'm' is not distributed
+SELECT citus_relation_size('m_dist_col_idx');
+ERROR:  cannot calculate the size because table 'm' for index 'm_dist_col_idx' is not distributed
+-- distribute the table(s)
+SELECT create_distributed_table('split_me', 'dist_col');
+ create_distributed_table
+---------------------------------------------------------------------
+
+(1 row)
+
+-- after citus
+SELECT citus_relation_size('split_me');
+ citus_relation_size
+---------------------------------------------------------------------
+                   0
+(1 row)
+
+SELECT citus_relation_size('split_me_dist_col_idx');
+ citus_relation_size
+---------------------------------------------------------------------
+                   0
+(1 row)
+
+SELECT citus_relation_size('m');
+ citus_relation_size
+---------------------------------------------------------------------
+               32768
+(1 row)
+
+SELECT citus_relation_size('m_dist_col_idx');
+ citus_relation_size
+---------------------------------------------------------------------
+               81920
+(1 row)
+
+DROP TABLE split_me;
 -- Test inside the transaction
 BEGIN;
 ALTER TABLE supplier ALTER COLUMN s_suppkey SET NOT NULL;

src/test/regress/sql/multi_cluster_management.sql

@@ -530,3 +530,10 @@ RESET citus.metadata_sync_mode;
 
 -- verify that at the end of this file, all primary nodes have metadata synced
 SELECT bool_and(hasmetadata) AND bool_and(metadatasynced) FROM pg_dist_node WHERE isactive = 't' and noderole = 'primary';
+
+-- Grant all on public schema to public
+--
+-- That's the default on Postgres versions < 15 and we want to
+-- keep permissions compatible accross versions, in regression
+-- tests.
+GRANT ALL ON SCHEMA public TO PUBLIC;

src/test/regress/sql/multi_size_queries.sql

@@ -13,10 +13,15 @@ SELECT citus_relation_size(1);
 SELECT citus_total_relation_size(1);
 
 -- Tests with non-distributed table
-CREATE TABLE non_distributed_table (x int);
+CREATE TABLE non_distributed_table (x int primary key);
+
 SELECT citus_table_size('non_distributed_table');
 SELECT citus_relation_size('non_distributed_table');
 SELECT citus_total_relation_size('non_distributed_table');
+
+SELECT citus_table_size('non_distributed_table_pkey');
+SELECT citus_relation_size('non_distributed_table_pkey');
+SELECT citus_total_relation_size('non_distributed_table_pkey');
 DROP TABLE non_distributed_table;
 
 -- fix broken placements via disabling the node
@@ -26,9 +31,25 @@ SELECT replicate_table_shards('lineitem_hash_part', shard_replication_factor:=2,
 -- Tests on distributed table with replication factor > 1
 VACUUM (FULL) lineitem_hash_part;
 
-SELECT citus_table_size('lineitem_hash_part');
+SELECT citus_relation_size('lineitem_hash_part') <= citus_table_size('lineitem_hash_part');
-SELECT citus_relation_size('lineitem_hash_part');
+SELECT citus_table_size('lineitem_hash_part') <= citus_total_relation_size('lineitem_hash_part');
-SELECT citus_total_relation_size('lineitem_hash_part');
+SELECT citus_relation_size('lineitem_hash_part') > 0;
+
+CREATE INDEX lineitem_hash_part_idx ON lineitem_hash_part(l_orderkey);
+VACUUM (FULL) lineitem_hash_part;
+
+SELECT citus_relation_size('lineitem_hash_part') <= citus_table_size('lineitem_hash_part');
+SELECT citus_table_size('lineitem_hash_part') <= citus_total_relation_size('lineitem_hash_part');
+SELECT citus_relation_size('lineitem_hash_part') > 0;
+
+SELECT citus_relation_size('lineitem_hash_part_idx') <= citus_table_size('lineitem_hash_part_idx');
+SELECT citus_table_size('lineitem_hash_part_idx') <= citus_total_relation_size('lineitem_hash_part_idx');
+SELECT citus_relation_size('lineitem_hash_part_idx') > 0;
+
+SELECT citus_total_relation_size('lineitem_hash_part') >=
+       citus_table_size('lineitem_hash_part') + citus_table_size('lineitem_hash_part_idx');
+
+DROP INDEX lineitem_hash_part_idx;
 
 VACUUM (FULL) customer_copy_hash;
 
@@ -40,7 +61,7 @@ SELECT citus_total_relation_size('customer_copy_hash');
 -- Make sure we can get multiple sizes in a single query
 SELECT citus_table_size('customer_copy_hash'),
        citus_table_size('customer_copy_hash'),
-       citus_table_size('supplier');
+       citus_table_size('customer_copy_hash');
 
 CREATE INDEX index_1 on customer_copy_hash(c_custkey);
 VACUUM (FULL) customer_copy_hash;
@@ -50,6 +71,10 @@ SELECT citus_table_size('customer_copy_hash');
 SELECT citus_relation_size('customer_copy_hash');
 SELECT citus_total_relation_size('customer_copy_hash');
 
+SELECT citus_table_size('index_1');
+SELECT citus_relation_size('index_1');
+SELECT citus_total_relation_size('index_1');
+
 -- Tests on reference table
 VACUUM (FULL) supplier;
 
@@ -64,6 +89,38 @@ SELECT citus_table_size('supplier');
 SELECT citus_relation_size('supplier');
 SELECT citus_total_relation_size('supplier');
 
+SELECT citus_table_size('index_2');
+SELECT citus_relation_size('index_2');
+SELECT citus_total_relation_size('index_2');
+
+-- Test on partitioned table
+CREATE TABLE split_me (dist_col int, partition_col timestamp) PARTITION BY RANGE (partition_col);
+CREATE INDEX ON split_me(dist_col);
+
+-- create 2 partitions
+CREATE TABLE m PARTITION OF split_me FOR VALUES FROM ('2018-01-01') TO ('2019-01-01');
+CREATE TABLE e PARTITION OF split_me FOR VALUES FROM ('2019-01-01') TO ('2020-01-01');
+
+INSERT INTO split_me SELECT 1, '2018-01-01'::timestamp + i * interval '1 day' FROM generate_series(1, 360) i;
+INSERT INTO split_me SELECT 2, '2019-01-01'::timestamp + i * interval '1 day' FROM generate_series(1, 180) i;
+
+-- before citus
+SELECT citus_relation_size('split_me');
+SELECT citus_relation_size('split_me_dist_col_idx');
+SELECT citus_relation_size('m');
+SELECT citus_relation_size('m_dist_col_idx');
+
+-- distribute the table(s)
+SELECT create_distributed_table('split_me', 'dist_col');
+
+-- after citus
+SELECT citus_relation_size('split_me');
+SELECT citus_relation_size('split_me_dist_col_idx');
+SELECT citus_relation_size('m');
+SELECT citus_relation_size('m_dist_col_idx');
+
+DROP TABLE split_me;
+
 -- Test inside the transaction
 BEGIN;
 ALTER TABLE supplier ALTER COLUMN s_suppkey SET NOT NULL;