Add support for column lists in ANALYZE commands

Was easy enough to extend to include this, so did so.

src/backend/distributed/executor/multi_utility.c

@@ -114,6 +114,7 @@ static void ProcessVacuumStmt(VacuumStmt *vacuumStmt, const char *vacuumCommand)
 static bool IsSupportedDistributedVacuumStmt(Oid relationId, VacuumStmt *vacuumStmt);
 static List * VacuumTaskList(Oid relationId, VacuumStmt *vacuumStmt);
 static StringInfo DeparseVacuumStmtPrefix(VacuumStmt *vacuumStmt);
+static char * DeparseVacuumColumnNames(List *columnNameList);
 
 
 /* Local functions forward declarations for unsupported command checks */
@@ -988,13 +989,6 @@ IsSupportedDistributedVacuumStmt(Oid relationId, VacuumStmt *vacuumStmt)
 							   "distributed %s commands", stmtName)));
 	}
 
-	if (vacuumStmt->va_cols != NIL)
-	{
-		ereport(ERROR, (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-						errmsg("specifying a column list is currently unsupported in "
-							   "distributed %s commands", stmtName)));
-	}
-
 	return true;
 }
 
@@ -1012,6 +1006,7 @@ VacuumTaskList(Oid relationId, VacuumStmt *vacuumStmt)
 	uint64 jobId = INVALID_JOB_ID;
 	int taskId = 1;
 	StringInfo vacuumString = DeparseVacuumStmtPrefix(vacuumStmt);
+	const char *columnNames = DeparseVacuumColumnNames(vacuumStmt->va_cols);
 	const int vacuumPrefixLen = vacuumString->len;
 	Oid schemaId = get_rel_namespace(relationId);
 	char *schemaName = get_namespace_name(schemaId);
@@ -1037,6 +1032,7 @@ VacuumTaskList(Oid relationId, VacuumStmt *vacuumStmt)
 
 		vacuumString->len = vacuumPrefixLen;
 		appendStringInfoString(vacuumString, shardName);
+		appendStringInfoString(vacuumString, columnNames);
 
 		task = CitusMakeNode(Task);
 		task->jobId = jobId;
@@ -1086,9 +1082,8 @@ DeparseVacuumStmtPrefix(VacuumStmt *vacuumStmt)
 		vacuumFlags &= ~VACOPT_ANALYZE;
 	}
 
-	/* unsupported flags and options should have already been rejected */
+	/* unsupported flags should have already been rejected */
 	Assert((vacuumFlags & unsupportedFlags) == 0);
-	Assert(vacuumStmt->va_cols == NIL);
 
 	/* if no flags remain, exit early */
 	if (vacuumFlags == 0)
@@ -1129,6 +1124,40 @@ DeparseVacuumStmtPrefix(VacuumStmt *vacuumStmt)
 }
 
 
+/*
+ * DeparseVacuumColumnNames joins the list of strings using commas as a
+ * delimiter. The whole thing is placed in parenthesis and set off with a
+ * single space in order to facilitate appending it to the end of any VACUUM
+ * or ANALYZE command which uses explicit column names. If the provided list
+ * is empty, this function returns an empty string to keep the calling code
+ * simplest.
+ */
+static char *
+DeparseVacuumColumnNames(List *columnNameList)
+{
+	StringInfo columnNames = makeStringInfo();
+	ListCell *columnNameCell = NULL;
+
+	if (columnNameList == NIL)
+	{
+		return columnNames->data;
+	}
+
+	appendStringInfoString(columnNames, " (");
+
+	foreach(columnNameCell, columnNameList)
+	{
+		char *columnName = strVal(lfirst(columnNameCell));
+
+		appendStringInfo(columnNames, "%s,", columnName);
+	}
+
+	columnNames->data[columnNames->len - 1] = ')';
+
+	return columnNames->data;
+}
+
+
 /*
  * ErrorIfUnsupportedIndexStmt checks if the corresponding index statement is
  * supported for distributed tables and errors out if it is not.

src/test/regress/expected/multi_utilities.out

@@ -68,7 +68,7 @@ SELECT master_apply_delete_command('DELETE FROM sharded_table');
 DROP TABLE sharded_table;
 -- VACUUM tests
 -- create a table with a single shard (for convenience)
-CREATE TABLE dustbunnies (id integer, name text);
+CREATE TABLE dustbunnies (id integer, name text, age integer);
 SELECT master_create_distributed_table('dustbunnies', 'id', 'hash');
  master_create_distributed_table 
 ---------------------------------
@@ -82,7 +82,7 @@ SELECT master_create_worker_shards('dustbunnies', 1, 2);
 (1 row)
 
 -- add some data to the distributed table
-\copy dustbunnies from stdin with csv
+\copy dustbunnies (id, name) from stdin with csv
 -- following approach adapted from PostgreSQL's stats.sql file
 -- save relevant stat counter values in refreshable view
 \c - - - :worker_1_port
@@ -201,6 +201,31 @@ WHERE oid='dustbunnies_990002'::regclass;
  t
 (1 row)
 
+-- check there are no nulls in either column
+SELECT attname, null_frac FROM pg_stats
+WHERE tablename = 'dustbunnies_990002' ORDER BY attname;
+ attname | null_frac 
+---------+-----------
+ age     |         1
+ id      |         0
+ name    |         0
+(3 rows)
+
+-- add NULL values, then perform column-specific ANALYZE
+\c - - - :master_port
+INSERT INTO dustbunnies VALUES (6, NULL, NULL);
+ANALYZE dustbunnies (name);
+-- verify that name's NULL ratio is updated but age's is not
+\c - - - :worker_1_port
+SELECT attname, null_frac FROM pg_stats
+WHERE tablename = 'dustbunnies_990002' ORDER BY attname;
+ attname | null_frac 
+---------+-----------
+ age     |         1
+ id      |         0
+ name    |  0.166667
+(3 rows)
+
 \c - - - :master_port
 -- verify warning for unqualified VACUUM
 VACUUM;
@@ -212,11 +237,6 @@ VACUUM dustbunnies;
 WARNING:  not propagating VACUUM command to worker nodes
 HINT:  Set citus.enable_ddl_propagation to true in order to send targeted VACUUM commands to worker nodes.
 SET citus.enable_ddl_propagation to DEFAULT;
--- verify error messages for unsupported options
-VACUUM (ANALYZE) dustbunnies (id);
-ERROR:  specifying a column list is currently unsupported in distributed VACUUM commands
-ANALYZE dustbunnies (id);
-ERROR:  specifying a column list is currently unsupported in distributed ANALYZE commands
 -- TODO: support VERBOSE
 -- VACUUM VERBOSE dustbunnies;
 -- VACUUM (FULL, VERBOSE) dustbunnies;

src/test/regress/sql/multi_utilities.sql

@@ -44,12 +44,12 @@ DROP TABLE sharded_table;
 -- VACUUM tests
 
 -- create a table with a single shard (for convenience)
-CREATE TABLE dustbunnies (id integer, name text);
+CREATE TABLE dustbunnies (id integer, name text, age integer);
 SELECT master_create_distributed_table('dustbunnies', 'id', 'hash');
 SELECT master_create_worker_shards('dustbunnies', 1, 2);
 
 -- add some data to the distributed table
-\copy dustbunnies from stdin with csv
+\copy dustbunnies (id, name) from stdin with csv
 1,bugs
 2,babs
 3,buster
@@ -146,6 +146,20 @@ VACUUM (FREEZE) dustbunnies;
 SELECT relfrozenxid::text::integer > :frozenxid AS frozen_performed FROM pg_class
 WHERE oid='dustbunnies_990002'::regclass;
 
+-- check there are no nulls in either column
+SELECT attname, null_frac FROM pg_stats
+WHERE tablename = 'dustbunnies_990002' ORDER BY attname;
+
+-- add NULL values, then perform column-specific ANALYZE
+\c - - - :master_port
+INSERT INTO dustbunnies VALUES (6, NULL, NULL);
+ANALYZE dustbunnies (name);
+
+-- verify that name's NULL ratio is updated but age's is not
+\c - - - :worker_1_port
+SELECT attname, null_frac FROM pg_stats
+WHERE tablename = 'dustbunnies_990002' ORDER BY attname;
+
 \c - - - :master_port
 -- verify warning for unqualified VACUUM
 VACUUM;
@@ -155,10 +169,6 @@ SET citus.enable_ddl_propagation to false;
 VACUUM dustbunnies;
 SET citus.enable_ddl_propagation to DEFAULT;
 
--- verify error messages for unsupported options
-VACUUM (ANALYZE) dustbunnies (id);
-ANALYZE dustbunnies (id);
-
 -- TODO: support VERBOSE
 -- VACUUM VERBOSE dustbunnies;
 -- VACUUM (FULL, VERBOSE) dustbunnies;