mirror of
https://github.com/hanzoai/datastore.git
synced 2026-09-26 13:55:07 +00:00
Alexey Milovidov
1c0d641c64
Address review feedback (@rschu1ze): add an .xml performance test that demonstrates the speedup of the bulk single-copy path for trivially serializable vectors. The test exercises writeVectorBinary / readVectorBinary through the analysisOfVariance aggregate state, whose state holds three large contiguous vectors (std::vector<Float64> and std::vector<UInt64>). Writing the AggregateFunction column to MergeTree serializes the state; reading it back deserializes it.
42 lines
1.6 KiB
XML
42 lines
1.6 KiB
XML
<test>
|
|
<!--
|
|
Exercises binary (de)serialization of std::vector of trivially-serializable
|
|
elements via writeVectorBinary / readVectorBinary.
|
|
|
|
The analysisOfVariance aggregate state holds three contiguous vectors
|
|
(std::vector<Float64> twice and std::vector<UInt64>), one element per group.
|
|
With many groups these vectors are large, so the bulk single-copy path added
|
|
in this change replaces an element-by-element (de)serialization loop.
|
|
|
|
Writing the AggregateFunction column to MergeTree serializes the state
|
|
(writeVectorBinary); reading it back deserializes it (readVectorBinary).
|
|
-->
|
|
<create_query>
|
|
CREATE TABLE aov_states
|
|
(
|
|
key UInt32,
|
|
state AggregateFunction(analysisOfVariance, Float64, UInt32)
|
|
)
|
|
ENGINE = MergeTree ORDER BY key
|
|
</create_query>
|
|
|
|
<!-- 200 states, each holding ~100000 groups => ~480 MB of vector data total.
|
|
The fill itself serializes the states when writing the parts. -->
|
|
<fill_query>
|
|
INSERT INTO aov_states
|
|
SELECT
|
|
intDiv(number, 100000) % 200 AS key,
|
|
analysisOfVarianceState(number * 1.5, (number % 100000)::UInt32)
|
|
FROM numbers_mt(20000000)
|
|
GROUP BY key
|
|
</fill_query>
|
|
|
|
<!-- Deserialization only (readVectorBinary). -->
|
|
<query>SELECT state FROM aov_states FORMAT Null</query>
|
|
|
|
<!-- Deserialization followed by merging the states. -->
|
|
<query>SELECT analysisOfVarianceMerge(state) FROM aov_states GROUP BY key FORMAT Null</query>
|
|
|
|
<drop_query>DROP TABLE IF EXISTS aov_states</drop_query>
|
|
</test>
|