Skip to content

Commit

Permalink
sql,opt: add support for inverted filter with JSON and Array columns
Browse files Browse the repository at this point in the history
This commit adds support for planning and executing an inverted index
scan + invertedFilter when there is a contains (@>) predicate on a JSON or
array column that has an inverted index. Similar to spatial columns, the
optimizer may be able to plan an inverted index scan + invertedFilter
for JSON and Array columns even if the filter condition is complex, with
multiple predicates combined with AND and OR expressions.

Informs cockroachdb#47340

Release note (performance improvement): The optimizer now supports using
an inverted index on JSON or Array columns for a wider variety of filter
predicates. Previously, inverted index usage was only supported for simple
predicates (e.g., `WHERE a @> '{"foo": "bar"}'), but now more complicated
predicates are supported by combining simple contains (@>) expressions with
AND and OR (e.g., `WHERE a @> '{"foo": "bar"}' OR a @> '{"foo": "baz"}'`).
An inverted index will be used if it is available on the filtered column and
the optimizer expects it to be more efficient than any alternative plan.
This may result in performance improvements for queries involving JSON and
Array columns.
  • Loading branch information
rytaft committed Nov 18, 2020
1 parent a416c60 commit 9a747ce
Show file tree
Hide file tree
Showing 25 changed files with 1,959 additions and 491 deletions.
270 changes: 270 additions & 0 deletions pkg/sql/logictest/testdata/logic_test/inverted_filter_json_array
Original file line number Diff line number Diff line change
@@ -0,0 +1,270 @@
# Test cases for using invertedFilterer on an inverted json or array index.

statement ok
CREATE TABLE json_tab (
a INT PRIMARY KEY,
b JSONB
)

statement ok
CREATE INVERTED INDEX foo_inv ON json_tab(b)

statement ok
CREATE TABLE array_tab (
a INT PRIMARY KEY,
b INT[]
)

statement ok
CREATE INVERTED INDEX foo_inv ON array_tab(b)

statement ok
INSERT INTO json_tab VALUES
(1, '{"a": "b"}'),
(2, '[1,2,3,4, "foo"]'),
(3, '{"a": {"b": "c"}}'),
(4, '{"a": {"b": [1]}}'),
(5, '{"a": {"b": [1, [2]]}}'),
(6, '{"a": {"b": [[2]]}}'),
(7, '{"a": "b", "c": "d"}'),
(8, '{"a": {"b":true}}'),
(9, '{"a": {"b":false}}'),
(10, '"a"'),
(11, 'null'),
(12, 'true'),
(13, 'false'),
(14, '1'),
(15, '1.23'),
(16, '[{"a": {"b": [1, [2]]}}, "d"]'),
(17, '{}'),
(18, '[]'),
(19, '["a", "a"]'),
(20, '[{"a": "a"}, {"a": "a"}]'),
(21, '[[[["a"]]], [[["a"]]]]'),
(22, '[1,2,3,1]'),
(23, '{"a": 123.123}'),
(24, '{"a": 123.123000}'),
(25, '{"a": [{}]}'),
(26, '[[], {}]'),
(27, '[true, false, null, 1.23, "a"]'),
(28, '{"a": {}}'),
(29, NULL),
(30, '{"a": []}'),
(31, '{"a": {"b": "c", "d": "e"}, "f": "g"}'),
(32, '{"a": [1]}'),
(33, '[1, "bar"]'),
(34, '{"a": 1}'),
(35, '[1]'),
(36, '[2]'),
(37, '[[1]]'),
(38, '[[2]]'),
(39, '["a"]'),
(40, '{"a": [[]]}'),
(41, '[[1, 2]]'),
(42, '[[1], [2]]'),
(43, '[{"a": "b", "c": "d"}]'),
(44, '[{"a": "b"}, {"c": "d"}]')

# Filter with a scalar.
query IT
SELECT * FROM json_tab WHERE b @> '1' ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
14 1
22 [1, 2, 3, 1]
33 [1, "bar"]
35 [1]

query I
SELECT a FROM json_tab WHERE b @> '1' ORDER BY a
----
2
14
22
33
35

# Filter with a fully-specified array. This should use a zigzag join.
query IT
SELECT * FROM json_tab WHERE b @> '[1, 2]' ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
22 [1, 2, 3, 1]

# Filter with fully-specified arrays. This should use an inverted filter.
query I
SELECT a FROM json_tab WHERE b @> '[1, 2]' OR b @> '[3, 4]' ORDER BY a
----
2
22

# Filter with a path ending in an empty object.
query IT
SELECT * FROM json_tab WHERE b @> '{"a": {}}' ORDER BY a
----
3 {"a": {"b": "c"}}
4 {"a": {"b": [1]}}
5 {"a": {"b": [1, [2]]}}
6 {"a": {"b": [[2]]}}
8 {"a": {"b": true}}
9 {"a": {"b": false}}
28 {"a": {}}
31 {"a": {"b": "c", "d": "e"}, "f": "g"}

query I
SELECT a FROM json_tab WHERE b @> '{"a": {}}' ORDER BY a
----
3
4
5
6
8
9
28
31

# Filter with a path ending in an empty array.
query IT
SELECT * FROM json_tab WHERE b @> '{"a": []}' ORDER BY a
----
25 {"a": [{}]}
30 {"a": []}
32 {"a": [1]}
40 {"a": [[]]}

query I
SELECT a FROM json_tab WHERE b @> '{"a": []}' ORDER BY a
----
25
30
32
40

# Filter with a nested array. This index expression is not tight.
# This should use a zigzag join.
query IT
SELECT * FROM json_tab WHERE b @> '[[1, 2]]' ORDER BY a
----
41 [[1, 2]]

# Filter with a nested array. This index expression is not tight.
# This should use an inverted filter
query I
SELECT a FROM json_tab WHERE b @> '[[1, 2]]' OR b @> '[[3, 4]]' ORDER BY a
----
41

# Combine predicates with AND. Should have the same output as b @> '[1, 2]'.
# This should use a zigzag join.
query I
SELECT a FROM json_tab WHERE b @> '[1]' AND b @> '[2]' ORDER BY a
----
2
22

# Combine predicates with OR.
query IT
SELECT * FROM json_tab@foo_inv WHERE b @> '[1]' OR b @> '[2]' ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
22 [1, 2, 3, 1]
33 [1, "bar"]
35 [1]
36 [2]

query I
SELECT a FROM json_tab@foo_inv WHERE b @> '[1]' OR b @> '[2]' ORDER BY a
----
2
22
33
35
36

# More complex combination.
query IT
SELECT * FROM json_tab
WHERE (b @> '[1]'::json OR b @> '[2]'::json) AND (b @> '3'::json OR b @> '"bar"'::json)
ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
22 [1, 2, 3, 1]
33 [1, "bar"]

# Combined with non-JSON predicates.
query IT
SELECT * FROM json_tab WHERE b @> '[1]' AND a % 2 = 0 ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
22 [1, 2, 3, 1]

# The split disjunction rule allows us to use the index for this query.
query IT
SELECT * FROM json_tab WHERE b @> '[1]' OR a = 44 ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
22 [1, 2, 3, 1]
33 [1, "bar"]
35 [1]
44 [{"a": "b"}, {"c": "d"}]

# We cannot use the index for this query.
query IT
SELECT * FROM json_tab WHERE b @> '[1]' OR sqrt(a::decimal) = 2 ORDER BY a
----
2 [1, 2, 3, 4, "foo"]
4 {"a": {"b": [1]}}
22 [1, 2, 3, 1]
33 [1, "bar"]
35 [1]

statement ok
INSERT INTO array_tab VALUES
(1, '{}'),
(2, '{1}'),
(3, '{2}'),
(4, '{1, 2}'),
(5, '{1, 3}'),
(6, '{1, 2, 3, 4}')

# Array operations.
query I
SELECT a FROM array_tab@foo_inv WHERE b @> '{}' ORDER BY a
----
1
2
3
4
5
6

# This should use a zigzag join.
query IT
SELECT * FROM array_tab@foo_inv WHERE b @> '{1, 2}' ORDER BY a
----
4 {1,2}
6 {1,2,3,4}

# Combined with non-Array predicates.
query IT
SELECT * FROM array_tab WHERE b @> '{1}' AND a % 2 = 0 ORDER BY a
----
2 {1}
4 {1,2}
6 {1,2,3,4}

# The split disjunction rule allows us to use the index for this query.
query IT
SELECT * FROM array_tab WHERE b @> '{1}' OR a = 1 ORDER BY a
----
1 {}
2 {1}
4 {1,2}
5 {1,3}
6 {1,2,3,4}

# We cannot use the index for this query.
query IT
SELECT * FROM array_tab WHERE (b @> '{2}' AND a = 3) OR b[0] = a ORDER BY a
----
3 {2}
Loading

0 comments on commit 9a747ce

Please sign in to comment.