mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2026-10-28 16:18:23 +08:00
ManuelandClaude Opus 5.5
d350a49658
Materialize the input arrays in `Memory` tables instead of producing them with `arrayMap` inside every query, where that took about a third of the query time. Drop the baseline query, which does not call `arrayDistinct`, and the `SipHash` path query, which showed no change and was the slowest. Measured A/B against the same build with the previous commit reverted, `max_threads = 1`, minimum of 42 runs (new / old): - `Array(UInt64)`, all distinct: 0.68 (the old query showed 0.79) - `Array(UInt64)`, duplicate-heavy: 0.88 (old query: 0.92) - `Array(String)`: 0.81 (old query: 0.71) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
32 lines
1.7 KiB
XML
32 lines
1.7 KiB
XML
<test>
|
|
<!-- The input arrays are materialized in Memory tables, so the queries measure `arrayDistinct`
|
|
rather than the cost of producing its input. A new element costs a hash computation, a
|
|
probe and an append; a repeated one only a hash computation and a probe. The mostly-distinct
|
|
inputs therefore stress insertion, the duplicate-heavy one stresses lookups of existing keys. -->
|
|
<settings>
|
|
<max_threads>1</max_threads>
|
|
</settings>
|
|
|
|
<create_query>CREATE TABLE array_distinct_numbers (distinct_values Array(UInt64), duplicate_values Array(UInt64)) ENGINE = Memory</create_query>
|
|
<create_query>CREATE TABLE array_distinct_strings (distinct_values Array(String)) ENGINE = Memory</create_query>
|
|
|
|
<!-- 30 elements per row: all distinct, or 5 distinct values repeated. -->
|
|
<fill_query>
|
|
INSERT INTO array_distinct_numbers
|
|
SELECT arrayMap(i -> (number + i) % 40, range(30)), arrayMap(i -> (number + i) % 5, range(30))
|
|
FROM numbers(1000000)
|
|
</fill_query>
|
|
<fill_query>
|
|
INSERT INTO array_distinct_strings
|
|
SELECT arrayMap(i -> toString((number + i) % 40), range(30))
|
|
FROM numbers(500000)
|
|
</fill_query>
|
|
|
|
<query>SELECT sum(length(arrayDistinct(distinct_values))) FROM array_distinct_numbers FORMAT Null</query>
|
|
<query>SELECT sum(length(arrayDistinct(duplicate_values))) FROM array_distinct_numbers FORMAT Null</query>
|
|
<query>SELECT sum(length(arrayDistinct(distinct_values))) FROM array_distinct_strings FORMAT Null</query>
|
|
|
|
<drop_query>DROP TABLE IF EXISTS array_distinct_numbers</drop_query>
|
|
<drop_query>DROP TABLE IF EXISTS array_distinct_strings</drop_query>
|
|
</test>
|