ClickHouse/tests/performance/uniq.xml
2021-05-21 09:29:56 +03:00

62 lines
2.4 KiB
XML

<test>
<preconditions>
<table_exists>hits_100m_single</table_exists>
<ram_size>30000000000</ram_size>
</preconditions>
<settings>
<max_memory_usage>30000000000</max_memory_usage>
<!--
Because of random distribution of data between threads, the number
of unique keys per thread might differ. This means that sometimes
we switch to two-level aggregation, and sometimes we don't, based on
the "bytes" threshold. Two-level aggregation turns out to be twice
as fast, because it merges aggregation states in multiple threads.
Lower the threshold here, to avoid jitter. It is unclear whether it
would be beneficial to lower the default as well.
-->
<group_by_two_level_threshold_bytes>10000000</group_by_two_level_threshold_bytes>
</settings>
<substitutions>
<substitution>
<name>key</name>
<values>
<!--
1 as a key doesn't make much sense - the queries are either
too long in case of uniqExact or too short in case of limited
uniq, so the result is drowned by the noise.
<value>1</value>
-->
<value>SearchEngineID</value>
<value>RegionID</value>
<value>SearchPhrase</value>
<value>ClientIP</value>
</values>
</substitution>
<substitution>
<name>func</name>
<values>
<value>sum</value>
<value>uniq</value>
<value>uniqExact</value>
<value>uniqHLL12</value>
<value>uniqCombined(12)</value>
<value>uniqCombined(13)</value>
<value>uniqCombined(14)</value>
<value>uniqCombined(15)</value>
<value>uniqCombined(16)</value>
<value>uniqCombined(17)</value>
<value>uniqCombined(18)</value>
<value>uniqUpTo(3)</value>
<value>uniqUpTo(5)</value>
<value>uniqUpTo(10)</value>
<value>uniqUpTo(25)</value>
<value>uniqUpTo(100)</value>
</values>
</substitution>
</substitutions>
<query>SELECT {key} AS k, {func}(UserID) FROM hits_100m_single GROUP BY k FORMAT Null</query>
</test>