Merge pull request #11953 from ClickHouse/revive-mmap

Revive mmap IO + add a test

src/Storages/MergeTree/MergeTreeDataSelectExecutor.cpp

@@ -608,6 +608,7 @@ Pipes MergeTreeDataSelectExecutor::readFromParts(
     MergeTreeReaderSettings reader_settings =
     {
         .min_bytes_to_use_direct_io = settings.min_bytes_to_use_direct_io,
+        .min_bytes_to_use_mmap_io = settings.min_bytes_to_use_mmap_io,
         .max_read_buffer_size = settings.max_read_buffer_size,
         .save_marks_in_cache = true
     };

src/Storages/MergeTree/MergeTreeIOSettings.h

@@ -10,7 +10,7 @@ struct MergeTreeReaderSettings
     size_t min_bytes_to_use_direct_io = 0;
     size_t min_bytes_to_use_mmap_io = 0;
     size_t max_read_buffer_size = DBMS_DEFAULT_BUFFER_SIZE;
-    /// If save_marks_in_cache is false, then, if marks are not in cache, 
+    /// If save_marks_in_cache is false, then, if marks are not in cache,
     ///  we will load them but won't save in the cache, to avoid evicting other data.
     bool save_marks_in_cache = false;
 };
@@ -33,4 +33,5 @@ struct MergeTreeWriterSettings
 
     size_t estimated_size = 0;
 };
+
 }

src/Storages/MergeTree/MergeTreeReaderCompact.cpp

@@ -57,7 +57,7 @@ MergeTreeReaderCompact::MergeTreeReaderCompact(
                     buffer_size,
                     0,
                     settings.min_bytes_to_use_direct_io,
-                    0);
+                    settings.min_bytes_to_use_mmap_io);
             },
             uncompressed_cache);
 
@@ -71,7 +71,8 @@ MergeTreeReaderCompact::MergeTreeReaderCompact(
     {
         auto buffer =
             std::make_unique<CompressedReadBufferFromFile>(
-                data_part->volume->getDisk()->readFile(full_data_path, buffer_size, 0, settings.min_bytes_to_use_direct_io, 0));
+                data_part->volume->getDisk()->readFile(
+                    full_data_path, buffer_size, 0, settings.min_bytes_to_use_direct_io, settings.min_bytes_to_use_mmap_io));
 
         if (profile_callback_)
             buffer->setProfileCallback(profile_callback_, clock_type_);

tests/queries/0_stateless/01343_min_bytes_to_use_mmap_io.reference

@@ -0,0 +1,2 @@
+Hello, world
+1

tests/queries/0_stateless/01343_min_bytes_to_use_mmap_io.sql

@@ -0,0 +1,11 @@
+DROP TABLE IF EXISTS test;
+CREATE TABLE test (x String) ENGINE = MergeTree ORDER BY tuple();
+INSERT INTO test VALUES ('Hello, world');
+
+SET min_bytes_to_use_mmap_io = 1;
+SELECT * FROM test;
+
+SYSTEM FLUSH LOGS;
+SELECT PE.Values FROM system.query_log ARRAY JOIN ProfileEvents AS PE WHERE event_date >= yesterday() AND event_time >= now() - 300 AND query LIKE 'SELECT * FROM test%' AND PE.Names = 'CreatedReadBufferMMap' AND type = 2 ORDER BY event_time DESC LIMIT 1;
+
+DROP TABLE test;