Make compression use the defaults functions

Previously, we create functions to calculate default order by and
segment by values. This PR makes those functions be used by default
when compression is enabled. We also added GUCs to disable those
functions or to use alternative functions for the defaults calculation.

.unreleased/pr_6696

@@ -0,0 +1 @@
+Implements: #6696 Improve defaults for compression segment_by and order_by

sql/compression_defaults.sql

@@ -45,10 +45,10 @@ BEGIN
     INNER JOIN
       pg_attribute a on (a.attnum = i.attnum AND a.attrelid = relation)
     --right now stats are from the hypertable itself. Use chunks in the future.
-    INNER JOIN pg_statistic s ON (s.staattnum = a.attnum and s.starelid = relation)
+    INNER JOIN pg_stats s ON (s.attname = a.attname and s.schemaname = _schema_name and s.tablename = _table_name)
     WHERE
       a.attname NOT IN (SELECT column_name FROM _timescaledb_catalog.dimension d WHERE d.hypertable_id = _hypertable_row.id)
-      AND s.stadistinct > 1
+      AND s.n_distinct > 1
     ORDER BY i.pos
     LIMIT 1;
 
@@ -75,10 +75,10 @@ BEGIN
     INNER JOIN
       pg_attribute a on (a.attnum = i.attnum AND a.attrelid = relation)
     --right now stats are from the hypertable itself. Use chunks in the future.
-    INNER JOIN pg_statistic s ON (s.staattnum = a.attnum and s.starelid = relation)
+    INNER JOIN pg_stats s ON (s.attname = a.attname and s.schemaname = _schema_name and s.tablename = _table_name)
     WHERE
       a.attname NOT IN (SELECT column_name FROM _timescaledb_catalog.dimension d WHERE d.hypertable_id = _hypertable_row.id)
-      AND s.stadistinct > 1
+      AND s.n_distinct > 1
     ORDER BY i.pos
     LIMIT 1;
 
@@ -106,10 +106,10 @@ BEGIN
     LEFT JOIN
       pg_catalog.pg_attrdef ad ON (ad.adrelid = relation AND ad.adnum = a.attnum)
     LEFT JOIN
-      pg_statistic s ON (s.staattnum = a.attnum and s.starelid = relation)
+      pg_stats s ON (s.attname = a.attname and s.schemaname = _schema_name and s.tablename = _table_name)
     WHERE
       a.attname NOT IN (SELECT column_name FROM _timescaledb_catalog.dimension d WHERE d.hypertable_id = _hypertable_row.id)
-      AND s.stadistinct is null
+      AND s.n_distinct is null
       AND a.attidentity = '' AND (ad.adbin IS NULL OR pg_get_expr(adbin, adrelid) not like 'nextval%')
     ORDER BY i.pos
     LIMIT 1;
@@ -141,10 +141,10 @@ BEGIN
     LEFT JOIN
       pg_catalog.pg_attrdef ad ON (ad.adrelid = relation AND ad.adnum = a.attnum)
     LEFT JOIN
-      pg_statistic s ON (s.staattnum = a.attnum and s.starelid = relation)
+      pg_stats s ON (s.attname = a.attname and s.schemaname = _schema_name and s.tablename = _table_name)
     WHERE
       a.attname NOT IN (SELECT column_name FROM _timescaledb_catalog.dimension d WHERE d.hypertable_id = _hypertable_row.id)
-      AND s.stadistinct is null
+      AND s.n_distinct is null
       AND a.attidentity = '' AND (ad.adbin IS NULL OR pg_get_expr(adbin, adrelid) not like 'nextval%')
     ORDER BY i.pos
     LIMIT 1;
@@ -285,12 +285,12 @@ BEGIN
 
     --add DESC to any dimensions
     SELECT
-      array_agg(
+      coalesce(array_agg(
       CASE WHEN d.column_name IS NULL THEN
-        a.colname
+        format('%I', a.colname)
       ELSE
-        a.colname || ' DESC'
-      END ORDER BY pos) INTO STRICT _orderby_clauses
+        format('%I DESC', a.colname)
+      END ORDER BY pos), array[]::text[]) INTO STRICT _orderby_clauses
     FROM unnest(_orderby_names) WITH ORDINALITY as a(colname, pos)
     LEFT JOIN _timescaledb_catalog.dimension d ON (d.column_name = a.colname AND d.hypertable_id = _hypertable_row.id);
 

src/compression_with_clause.c

@@ -176,8 +176,8 @@ throw_order_by_error(char *order_by)
 }
 
 /* compress_orderby is parsed same as order by in select queries */
-static OrderBySettings
-parse_order_collist(char *inpstr, Hypertable *hypertable)
+OrderBySettings
+ts_compress_parse_order_collist(char *inpstr, Hypertable *hypertable)
 {
 	StringInfoData buf;
 	List *parsed;
@@ -322,13 +322,9 @@ ts_compress_hypertable_parse_segment_by(WithClauseResult *parsed_options, Hypert
 OrderBySettings
 ts_compress_hypertable_parse_order_by(WithClauseResult *parsed_options, Hypertable *hypertable)
 {
-	if (parsed_options[CompressOrderBy].is_default == false)
-	{
-		Datum textarg = parsed_options[CompressOrderBy].parsed;
-		return parse_order_collist(TextDatumGetCString(textarg), hypertable);
-	}
-	else
-		return (OrderBySettings){ 0 };
+	Ensure(parsed_options[CompressOrderBy].is_default == false, "with clause is not default");
+	Datum textarg = parsed_options[CompressOrderBy].parsed;
+	return ts_compress_parse_order_collist(TextDatumGetCString(textarg), hypertable);
 }
 
 /* returns List of CompressedParsedCol

src/compression_with_clause.h

@@ -44,3 +44,5 @@ ts_compress_hypertable_parse_order_by(WithClauseResult *parsed_options, Hypertab
 extern TSDLLEXPORT Interval *
 ts_compress_hypertable_parse_chunk_time_interval(WithClauseResult *parsed_options,
 												 Hypertable *hypertable);
+extern TSDLLEXPORT OrderBySettings ts_compress_parse_order_collist(char *inpstr,
+																   Hypertable *hypertable);

src/guc.c

@@ -5,9 +5,13 @@
  */
 #include <postgres.h>
 #include <utils/guc.h>
+#include <utils/varlena.h>
+#include <utils/regproc.h>
+#include <parser/parse_func.h>
 #include <miscadmin.h>
 
 #include "guc.h"
+#include "extension.h"
 #include "license_guc.h"
 #include "config.h"
 #include "hypertable_cache.h"
@@ -76,6 +80,8 @@ TSDLLEXPORT bool ts_guc_enable_bulk_decompression = true;
 TSDLLEXPORT bool ts_guc_auto_sparse_indexes = true;
 TSDLLEXPORT int ts_guc_bgw_log_level = WARNING;
 TSDLLEXPORT bool ts_guc_enable_skip_scan = true;
+static char *ts_guc_default_segmentby_fn = NULL;
+static char *ts_guc_default_orderby_fn = NULL;
 /* default value of ts_guc_max_open_chunks_per_insert and ts_guc_max_cached_chunks_per_hypertable
  * will be set as their respective boot-value when the GUC mechanism starts up */
 int ts_guc_max_open_chunks_per_insert;
@@ -218,6 +224,90 @@ assign_max_open_chunks_per_insert_hook(int newval, void *extra)
 	validate_chunk_cache_sizes(ts_guc_max_cached_chunks_per_hypertable, newval);
 }
 
+static Oid
+get_segmentby_func(char *input_name)
+{
+	List *namelist = NIL;
+
+	if (strlen(input_name) == 0)
+	{
+		return InvalidOid;
+	}
+
+#if PG16_LT
+	namelist = stringToQualifiedNameList(input_name);
+#else
+	namelist = stringToQualifiedNameList(input_name, NULL);
+#endif
+	Oid argtyp[] = { REGCLASSOID };
+	return LookupFuncName(namelist, lengthof(argtyp), argtyp, true);
+}
+
+static bool
+check_segmentby_func(char **newval, void **extra, GucSource source)
+{
+	/* if the extension doesn't exist you can't check for the function, have to take it on faith */
+	if (ts_extension_is_loaded())
+	{
+		Oid segment_func_oid = get_segmentby_func(*newval);
+
+		if (strlen(*newval) > 0 && !OidIsValid(segment_func_oid))
+		{
+			GUC_check_errdetail("Function \"%s\" does not exist.", *newval);
+			return false;
+		}
+	}
+	return true;
+}
+
+Oid
+ts_guc_default_segmentby_fn_oid()
+{
+	return get_segmentby_func(ts_guc_default_segmentby_fn);
+}
+
+static Oid
+get_orderby_func(char *input_name)
+{
+	List *namelist = NIL;
+
+	if (strlen(input_name) == 0)
+	{
+		return InvalidOid;
+	}
+
+#if PG16_LT
+	namelist = stringToQualifiedNameList(input_name);
+#else
+	namelist = stringToQualifiedNameList(input_name, NULL);
+#endif
+	Oid argtyp[] = { REGCLASSOID, TEXTARRAYOID };
+	return LookupFuncName(namelist, lengthof(argtyp), argtyp, true);
+}
+
+static bool
+check_orderby_func(char **newval, void **extra, GucSource source)
+{
+	/* if the extension doesn't exist you can't check for the function, have to take it on faith */
+	if (ts_extension_is_loaded())
+	{
+		Oid func_oid = get_orderby_func(*newval);
+
+		if (strlen(*newval) > 0 && !OidIsValid(func_oid))
+		{
+			GUC_check_errdetail("Function \"%s\" does not exist.", *newval);
+			return false;
+		}
+	}
+	return true;
+}
+
+Oid
+ts_guc_default_orderby_fn_oid()
+{
+	return get_orderby_func(ts_guc_default_orderby_fn);
+}
+
 void
 _guc_init(void)
 {
@@ -570,6 +660,32 @@ _guc_init(void)
 							 NULL);
 #endif
 
+	DefineCustomStringVariable(/* name= */ MAKE_EXTOPTION("compression_segmentby_default_function"),
+							   /* short_desc= */ "Function that sets default segment_by",
+							   /* long_desc= */
+							   "Function to use for calculating default segment_by setting for "
+							   "compression",
+							   /* valueAddr= */ &ts_guc_default_segmentby_fn,
+							   /* Value= */ "_timescaledb_functions.get_segmentby_defaults",
+							   /* context= */ PGC_USERSET,
+							   /* flags= */ 0,
+							   /* check_hook= */ check_segmentby_func,
+							   /* assign_hook= */ NULL,
+							   /* show_hook= */ NULL);
+
+	DefineCustomStringVariable(/* name= */ MAKE_EXTOPTION("compression_orderby_default_function"),
+							   /* short_desc= */ "Function that sets default order_by",
+							   /* long_desc= */
+							   "Function to use for calculating default order_by setting for "
+							   "compression",
+							   /* valueAddr= */ &ts_guc_default_orderby_fn,
+							   /* Value= */ "_timescaledb_functions.get_orderby_defaults",
+							   /* context= */ PGC_USERSET,
+							   /* flags= */ 0,
+							   /* check_hook= */ check_orderby_func,
+							   /* assign_hook= */ NULL,
+							   /* show_hook= */ NULL);
+
 	DefineCustomStringVariable(/* name= */ MAKE_EXTOPTION("license"),
 							   /* short_desc= */ "TimescaleDB license type",
 							   /* long_desc= */ "Determines which features are enabled",

src/guc.h

@@ -94,3 +94,5 @@ typedef enum
 } FeatureFlagType;
 
 extern TSDLLEXPORT void ts_feature_flag_check(FeatureFlagType);
+extern TSDLLEXPORT Oid ts_guc_default_segmentby_fn_oid(void);
+extern TSDLLEXPORT Oid ts_guc_default_orderby_fn_oid(void);

src/ts_catalog/array_utils.c

@@ -89,6 +89,42 @@ ts_array_is_member(ArrayType *arr, const char *name)
 	return ret;
 }
 
+extern TSDLLEXPORT void
+ts_array_append_stringinfo(ArrayType *arr, StringInfo info)
+{
+	bool first = true;
+	Datum datum;
+	bool null;
+
+	if (!arr)
+		return;
+
+	Assert(ARR_NDIM(arr) <= 1);
+	Assert(arr->elemtype == TEXTOID);
+
+	ArrayIterator it = array_create_iterator(arr, 0, NULL);
+	while (array_iterate(it, &datum, &null))
+	{
+		Assert(!null);
+		/*
+		 * Our internal catalog arrays should either be NULL or
+		 * have non-NULL members. During normal operation it should
+		 * never have NULL members. If we have NULL members either
+		 * the catalog is corrupted or some catalog tampering has
+		 * happened.
+		 */
+		Ensure(!null, "array element was NULL");
+		if (!first)
+			appendStringInfoString(info, ", ");
+		else
+			first = false;
+
+		appendStringInfo(info, "%s", TextDatumGetCString(datum));
+	}
+
+	array_free_iterator(it);
+}
+
 extern TSDLLEXPORT int
 ts_array_position(ArrayType *arr, const char *name)
 {

src/ts_catalog/array_utils.h

@@ -20,6 +20,7 @@
 extern TSDLLEXPORT int ts_array_length(ArrayType *arr);
 extern TSDLLEXPORT bool ts_array_equal(ArrayType *left, ArrayType *right);
 extern TSDLLEXPORT bool ts_array_is_member(ArrayType *arr, const char *name);
+extern TSDLLEXPORT void ts_array_append_stringinfo(ArrayType *arr, StringInfo info);
 extern TSDLLEXPORT int ts_array_position(ArrayType *arr, const char *name);
 
 extern TSDLLEXPORT bool ts_array_get_element_bool(ArrayType *arr, int position);