mirror of
https://github.com/hanzoai/datastore.git
synced 2026-09-26 13:55:07 +00:00
Amos Bird
16eef62137
- Replace expensive `hex(sipHash128(...))` fills with cheap `toString(number)` variants and reduce row count (5M → 2M). The previous fill dominated the test runtime and tripped the "too slow" perf-test warning. - Broaden the PK / skip-index overlap to 5 String columns, each with its own `bloom_filter` skip index (plus a `minmax` on the DateTime64 PK column), so the benchmark exposes more of the speedup the cache enables.
70 lines
2.4 KiB
XML
70 lines
2.4 KiB
XML
<test>
|
|
<!-- Stresses the permuted columns cache: 5 String ORDER BY columns,
|
|
each with its own `bloom_filter` skip index, plus a `minmax` index
|
|
on a DateTime64 column that is also in the primary key.
|
|
|
|
Without the cache, every primary-key column is permuted twice
|
|
(once for the primary index, once for its skip index). With the
|
|
cache, each column is permuted only once. The more PK/skip-index
|
|
overlap, the bigger the win. -->
|
|
|
|
<create_query>
|
|
CREATE TABLE permute_cache_wide
|
|
(
|
|
c1 String,
|
|
c2 String,
|
|
c3 String,
|
|
c4 String,
|
|
c5 String,
|
|
ts DateTime64(6),
|
|
payload String,
|
|
value UInt64,
|
|
INDEX idx_c1 (c1) TYPE bloom_filter,
|
|
INDEX idx_c2 (c2) TYPE bloom_filter,
|
|
INDEX idx_c3 (c3) TYPE bloom_filter,
|
|
INDEX idx_c4 (c4) TYPE bloom_filter,
|
|
INDEX idx_c5 (c5) TYPE bloom_filter,
|
|
INDEX idx_ts (ts) TYPE minmax
|
|
)
|
|
ENGINE = MergeTree
|
|
ORDER BY (c1, c2, c3, c4, c5, ts)
|
|
SETTINGS min_rows_for_wide_part = 0, min_bytes_for_wide_part = 0
|
|
</create_query>
|
|
|
|
<!-- Pre-materialize the source data so the measured query only does
|
|
the part-writing work (which is what the cache affects). Use cheap
|
|
string generation (`toString(number)`) to keep the fill fast. -->
|
|
<create_query>
|
|
CREATE TABLE permute_cache_src
|
|
(
|
|
c1 String,
|
|
c2 String,
|
|
c3 String,
|
|
c4 String,
|
|
c5 String,
|
|
ts DateTime64(6),
|
|
payload String,
|
|
value UInt64
|
|
)
|
|
ENGINE = Memory
|
|
</create_query>
|
|
|
|
<fill_query>
|
|
INSERT INTO permute_cache_src SELECT
|
|
toString(number % 1000),
|
|
toString((number * 17) % 10000),
|
|
toString((number * 31) % 100000),
|
|
toString((number * 53) % 1000000),
|
|
toString(number),
|
|
toDateTime64('2024-01-01', 6) + number / 1000000,
|
|
toString(number),
|
|
number
|
|
FROM numbers(2000000)
|
|
</fill_query>
|
|
|
|
<query>INSERT INTO permute_cache_wide SELECT * FROM permute_cache_src</query>
|
|
|
|
<drop_query>DROP TABLE IF EXISTS permute_cache_wide</drop_query>
|
|
<drop_query>DROP TABLE IF EXISTS permute_cache_src</drop_query>
|
|
</test>
|