Performance test

src/Processors/Formats/Impl/ParquetBlockInputFormat.cpp

@@ -389,7 +389,9 @@ ParquetBlockInputFormat::~ParquetBlockInputFormat()
 
 void ParquetBlockInputFormat::setQueryInfo(const SelectQueryInfo & query_info, ContextPtr context)
 {
-    if (format_settings.parquet.filter_push_down)
+    /// When analyzer is enabled, query_info.filter_asts is missing sets and maybe some type casts,
+    /// so don't use it. I'm not sure how to support analyzer here: https://github.com/ClickHouse/ClickHouse/issues/53536
+    if (format_settings.parquet.filter_push_down && !context->getSettingsRef().allow_experimental_analyzer)
         key_condition.emplace(query_info, context, getPort().getHeader().getNames(),
             std::make_shared<ExpressionActions>(std::make_shared<ActionsDAG>(
                 getPort().getHeader().getColumnsWithTypeAndName())));
@@ -428,7 +430,7 @@ void ParquetBlockInputFormat::initializeIfNeeded()
 
         if (key_condition.has_value() &&
             !key_condition->checkInHyperrectangle(
-                getHyperrectangleForRowGroup(*metadata, idx, getPort().getHeader(), format_settings),
+                getHyperrectangleForRowGroup(*metadata, row_group, getPort().getHeader(), format_settings),
                 getPort().getHeader().getDataTypes()).can_be_true)
             continue;
 

tests/performance/README.md

@@ -4,11 +4,11 @@ This directory contains `.xml`-files with performance tests for @akuzm tool.
 
 ### How to write performance test
 
-First of all you should check existing tests don't cover your case. If there are no such tests than you should write your own.
+First of all you should check existing tests don't cover your case. If there are no such tests then you should write your own.
 
 You can use `substitions`, `create`, `fill` and `drop` queries to prepare test. You can find examples in this folder.
 
-If your test continued more than 10 minutes, please, add tag `long` to have an opportunity to run all tests and skip long ones.
+If your test takes more than 10 minutes, please, add tag `long` to have an opportunity to run all tests and skip long ones.
 
 ### How to run performance test
 

tests/performance/parquet_filter.xml

@@ -0,0 +1,9 @@
+<test>
+    <create_query>create table if not exists t (key UInt64, value String) engine = File(Parquet) settings output_format_parquet_use_custom_encoder=1, output_format_parquet_row_group_size=100000</create_query>
+
+    <fill_query>insert into t select number, toString(number) from numbers(2000000) settings max_threads=16, max_insert_threads=16, max_insert_block_size=100000, max_block_size=100000</fill_query>
+
+    <query>select sum(cityHash64(*)) from t where key between 1050000 and 1150000 settings max_threads=1</query>
+
+    <drop_query>drop table if exists t</drop_query>
+</test>

tests/queries/0_stateless/02841_parquet_filter_pushdown.sql

@@ -8,6 +8,10 @@ set optimize_or_like_chain = 0;
 set max_block_size = 100000;
 set max_insert_threads = 1;
 
+-- Analyzer breaks the queries with IN and some queries with BETWEEN.
+-- TODO: Figure out why.
+set allow_experimental_analyzer=0;
+
 -- Try all the types.
 insert into function file('02841.parquet')
     -- Use negative numbers to test sign extension for signed types and lack of sign extension for