Restore contsel/contjoinsel for containment & key-existence operators (#2356)

Makefile

@@ -183,7 +183,8 @@ REGRESS = scan \
           direct_field_access \
           security \
           reserved_keyword_alias \
-          agtype_jsonb_cast
+          agtype_jsonb_cast \
+          containment_selectivity
 
 ifneq ($(EXTRA_TESTS),)
   REGRESS += $(EXTRA_TESTS)

age--1.7.0--y.y.y.sql

@@ -762,3 +762,41 @@ CREATE FUNCTION ag_catalog._agehash_self_test()
     VOLATILE
     PARALLEL UNSAFE
 AS 'MODULE_PATHNAME';
+
+--
+-- Issue #2356: rebind containment and key-existence operators to the
+-- lightweight contsel / contjoinsel selectivity estimators.
+--
+-- @>, <@, @>>, <<@, ?, ?|, ?& on agtype were bound to matchingsel /
+-- matchingjoinsel. During planning matchingsel invokes the operator's
+-- underlying function (agtype_contains) once per pg_statistic MCV entry and
+-- histogram bin; with a realistic default_statistics_target that planning
+-- cost dominates simple point queries (the regression reported in #2356).
+--
+-- contsel / contjoinsel return fixed selectivity constants without calling the
+-- operator function, so planning is constant-time. This is a deliberate
+-- planning-speed vs. estimate-accuracy trade-off. Note it DIVERGES from
+-- PostgreSQL core, which keeps jsonb's @>, <@, ?, ?|, ?& on matchingsel /
+-- matchingjoinsel; it is an AGE-specific choice favoring workloads where these
+-- operators appear in selective point lookups.
+--
+ALTER OPERATOR ag_catalog.@>(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.<@(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.@>>(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.<<@(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, text)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, text[])
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, text[])
+    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, agtype)
+    SET (RESTRICT = contsel, JOIN = contjoinsel);

regress/expected/containment_selectivity.out

@@ -0,0 +1,245 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+ * KIND, either express or implied.  See the License for the
+ * specific language governing permissions and limitations
+ * under the License.
+ */
+/*
+ * Regression coverage for issue #2356:
+ *   The containment (@>, <@, @>>, <<@) and key-existence (?, ?|, ?&)
+ *   operators on agtype must be bound to the lightweight selectivity
+ *   helpers contsel / contjoinsel during planning. Earlier PG14+
+ *   branches used matchingsel / matchingjoinsel, which caused planning
+ *   to invoke agtype_contains() against pg_statistic MCVs and produced
+ *   a 30%+ planning-time regression on point queries (severe TPS drop
+ *   reported on the PG18 branch).
+ *
+ *   This test pins the bindings by querying pg_operator directly. If
+ *   someone re-introduces matchingsel here, the test diff is loud and
+ *   precise.
+ */
+LOAD 'age';
+SET search_path TO ag_catalog;
+-- Selectivity helpers for the four containment operators.
+SELECT o.oprname,
+       pg_catalog.format_type(o.oprleft,  NULL) AS lhs,
+       pg_catalog.format_type(o.oprright, NULL) AS rhs,
+       o.oprrest::text  AS restrict_fn,
+       o.oprjoin::text  AS join_fn
+FROM   pg_catalog.pg_operator o
+JOIN   pg_catalog.pg_namespace n ON n.oid = o.oprnamespace
+WHERE  n.nspname = 'ag_catalog'
+  AND  o.oprname IN ('@>', '<@', '@>>', '<<@')
+ORDER  BY o.oprname, lhs, rhs;
+ oprname |  lhs   |  rhs   | restrict_fn |   join_fn   
+---------+--------+--------+-------------+-------------
+ <<@     | agtype | agtype | contsel     | contjoinsel
+ <@      | agtype | agtype | contsel     | contjoinsel
+ @>      | agtype | agtype | contsel     | contjoinsel
+ @>>     | agtype | agtype | contsel     | contjoinsel
+(4 rows)
+
+-- Selectivity helpers for all key-existence operator overloads
+-- (right-hand side may be text, text[], or agtype).
+SELECT o.oprname,
+       pg_catalog.format_type(o.oprleft,  NULL) AS lhs,
+       pg_catalog.format_type(o.oprright, NULL) AS rhs,
+       o.oprrest::text  AS restrict_fn,
+       o.oprjoin::text  AS join_fn
+FROM   pg_catalog.pg_operator o
+JOIN   pg_catalog.pg_namespace n ON n.oid = o.oprnamespace
+WHERE  n.nspname = 'ag_catalog'
+  AND  o.oprname IN ('?', '?|', '?&')
+ORDER  BY o.oprname, lhs, rhs;
+ oprname |  lhs   |  rhs   | restrict_fn |   join_fn   
+---------+--------+--------+-------------+-------------
+ ?       | agtype | agtype | contsel     | contjoinsel
+ ?       | agtype | text   | contsel     | contjoinsel
+ ?&      | agtype | agtype | contsel     | contjoinsel
+ ?&      | agtype | text[] | contsel     | contjoinsel
+ ?|      | agtype | agtype | contsel     | contjoinsel
+ ?|      | agtype | text[] | contsel     | contjoinsel
+(6 rows)
+
+-- Scoped guard for issue #2356: assert that none of the specific containment
+-- and key-existence operators on agtype are bound to matchingsel /
+-- matchingjoinsel. We deliberately limit the check to these operator names
+-- (rather than every operator in ag_catalog) so unrelated operators that
+-- legitimately use matchingsel for their own semantics are not affected by
+-- this regression test.
+SELECT COUNT(*) AS leaked_matchingsel_bindings
+FROM   pg_catalog.pg_operator o
+JOIN   pg_catalog.pg_namespace n ON n.oid = o.oprnamespace
+WHERE  n.nspname = 'ag_catalog'
+  AND  o.oprname IN ('@>', '<@', '@>>', '<<@', '?', '?|', '?&')
+  AND  (o.oprrest::text  = 'matchingsel'
+        OR o.oprjoin::text = 'matchingjoinsel');
+ leaked_matchingsel_bindings 
+-----------------------------
+                           0
+(1 row)
+
+-- Smoke test: each operator still works functionally. Selectivity binding
+-- only affects the planner; this guards against an inadvertent operator
+-- removal as part of any future cleanup.
+SELECT '{"a":1,"b":2}'::agtype @>  '{"a":1}'::agtype             AS contains_yes;
+ contains_yes 
+--------------
+ t
+(1 row)
+
+SELECT '{"a":1}'::agtype       <@  '{"a":1,"b":2}'::agtype       AS contained_yes;
+ contained_yes 
+---------------
+ t
+(1 row)
+
+SELECT '{"a":{"b":1}}'::agtype @>> '{"a":{"b":1}}'::agtype       AS top_contains_yes;
+ top_contains_yes 
+------------------
+ t
+(1 row)
+
+SELECT '{"a":{"b":1}}'::agtype <<@ '{"a":{"b":1}}'::agtype       AS top_contained_yes;
+ top_contained_yes 
+-------------------
+ t
+(1 row)
+
+SELECT '{"a":1}'::agtype       ?   'a'::text                     AS exists_text_yes;
+ exists_text_yes 
+-----------------
+ t
+(1 row)
+
+SELECT '{"a":1}'::agtype       ?   '"a"'::agtype                 AS exists_agtype_yes;
+ exists_agtype_yes 
+-------------------
+ t
+(1 row)
+
+SELECT '{"a":1,"b":2}'::agtype ?|  ARRAY['a','c']                AS exists_any_text_yes;
+ exists_any_text_yes 
+---------------------
+ t
+(1 row)
+
+SELECT '{"a":1,"b":2}'::agtype ?|  '["a","c"]'::agtype           AS exists_any_agtype_yes;
+ exists_any_agtype_yes 
+-----------------------
+ t
+(1 row)
+
+SELECT '{"a":1,"b":2}'::agtype ?&  ARRAY['a','b']                AS exists_all_text_yes;
+ exists_all_text_yes 
+---------------------
+ t
+(1 row)
+
+SELECT '{"a":1,"b":2}'::agtype ?&  '["a","b"]'::agtype           AS exists_all_agtype_yes;
+ exists_all_agtype_yes 
+-----------------------
+ t
+(1 row)
+
+-- Upgrade-path assertion for issue #2356.
+--
+-- The checks above cover a FRESH install: contsel / contjoinsel come straight
+-- from agtype_operators.sql and agtype_exists.sql. Existing installs instead
+-- pick up the fix from the ALTER OPERATOR ... SET (RESTRICT, JOIN) block that
+-- age--1.7.0--y.y.y.sql ships and "ALTER EXTENSION age UPDATE" replays. Nothing
+-- above exercises that block, so a silent regression in it would go unnoticed.
+--
+-- We replay the shipped ALTER OPERATOR statements directly rather than running
+-- ALTER EXTENSION age UPDATE: the dev upgrade script targets the placeholder
+-- version "y.y.y" and is not a stable version-chain target inside the
+-- regression harness. The whole section runs in a transaction that is rolled
+-- back, so it observes the flip without permanently mutating the operator
+-- catalog (PostgreSQL DDL is transactional).
+BEGIN;
+-- Simulate a stale (pre-fix) install: force all ten overloads back onto
+-- matchingsel / matchingjoinsel.
+ALTER OPERATOR ag_catalog.@>(agtype, agtype)   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.<@(agtype, agtype)   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.@>>(agtype, agtype)  SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.<<@(agtype, agtype)  SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, text)      SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, agtype)    SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, text[])   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, agtype)   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, text[])   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, agtype)   SET (RESTRICT = matchingsel, JOIN = matchingjoinsel);
+-- Stale state: every overload now reports matchingsel / matchingjoinsel.
+SELECT o.oprname,
+       pg_catalog.format_type(o.oprleft,  NULL) AS lhs,
+       pg_catalog.format_type(o.oprright, NULL) AS rhs,
+       o.oprrest::text  AS restrict_fn,
+       o.oprjoin::text  AS join_fn
+FROM   pg_catalog.pg_operator o
+JOIN   pg_catalog.pg_namespace n ON n.oid = o.oprnamespace
+WHERE  n.nspname = 'ag_catalog'
+  AND  o.oprname IN ('@>', '<@', '@>>', '<<@', '?', '?|', '?&')
+ORDER  BY o.oprname, lhs, rhs;
+ oprname |  lhs   |  rhs   | restrict_fn |     join_fn     
+---------+--------+--------+-------------+-----------------
+ <<@     | agtype | agtype | matchingsel | matchingjoinsel
+ <@      | agtype | agtype | matchingsel | matchingjoinsel
+ ?       | agtype | agtype | matchingsel | matchingjoinsel
+ ?       | agtype | text   | matchingsel | matchingjoinsel
+ ?&      | agtype | agtype | matchingsel | matchingjoinsel
+ ?&      | agtype | text[] | matchingsel | matchingjoinsel
+ ?|      | agtype | agtype | matchingsel | matchingjoinsel
+ ?|      | agtype | text[] | matchingsel | matchingjoinsel
+ @>      | agtype | agtype | matchingsel | matchingjoinsel
+ @>>     | agtype | agtype | matchingsel | matchingjoinsel
+(10 rows)
+
+-- Replay the exact ALTER OPERATOR block shipped in age--1.7.0--y.y.y.sql.
+ALTER OPERATOR ag_catalog.@>(agtype, agtype)   SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.<@(agtype, agtype)   SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.@>>(agtype, agtype)  SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.<<@(agtype, agtype)  SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, text)      SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?(agtype, agtype)    SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, text[])   SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?|(agtype, agtype)   SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, text[])   SET (RESTRICT = contsel, JOIN = contjoinsel);
+ALTER OPERATOR ag_catalog.?&(agtype, agtype)   SET (RESTRICT = contsel, JOIN = contjoinsel);
+-- After the upgrade replay every overload is back on contsel / contjoinsel.
+SELECT o.oprname,
+       pg_catalog.format_type(o.oprleft,  NULL) AS lhs,
+       pg_catalog.format_type(o.oprright, NULL) AS rhs,
+       o.oprrest::text  AS restrict_fn,
+       o.oprjoin::text  AS join_fn
+FROM   pg_catalog.pg_operator o
+JOIN   pg_catalog.pg_namespace n ON n.oid = o.oprnamespace
+WHERE  n.nspname = 'ag_catalog'
+  AND  o.oprname IN ('@>', '<@', '@>>', '<<@', '?', '?|', '?&')
+ORDER  BY o.oprname, lhs, rhs;
+ oprname |  lhs   |  rhs   | restrict_fn |   join_fn   
+---------+--------+--------+-------------+-------------
+ <<@     | agtype | agtype | contsel     | contjoinsel
+ <@      | agtype | agtype | contsel     | contjoinsel
+ ?       | agtype | agtype | contsel     | contjoinsel
+ ?       | agtype | text   | contsel     | contjoinsel
+ ?&      | agtype | agtype | contsel     | contjoinsel
+ ?&      | agtype | text[] | contsel     | contjoinsel
+ ?|      | agtype | agtype | contsel     | contjoinsel
+ ?|      | agtype | text[] | contsel     | contjoinsel
+ @>      | agtype | agtype | contsel     | contjoinsel
+ @>>     | agtype | agtype | contsel     | contjoinsel
+(10 rows)
+
+ROLLBACK;