Seek block when starts a ScanKey

be/src/common/config.h

@@ -431,10 +431,6 @@ namespace config {
     // and the tablet will be marked as bad, so that FE will try to repair it.
     CONF_Bool(auto_recover_index_loading_failure, "false");
 
-    // This configuration is used to recover compaction under the corner case.
-    // If this configuration is set to true, block will seek position.
-    CONF_Bool(block_seek_position, "false");
-
     // max external scan cache batch count, means cache max_memory_cache_batch_count * batch_size row
     // default is 10, batch_size's defualt value is 1024 means 10 * 1024 rows will be cached
     CONF_Int32(max_memory_sink_batch_count, "20");

be/src/olap/rowset/segment_reader.cpp

@@ -289,6 +289,15 @@ OLAPStatus SegmentReader::seek_to_block(
     *next_block_id = _next_block_id;
     *eof = _eof;
 
+    // Must seek block when starts a ScanKey.
+    // In Doris, one block has 1024 rows.
+    // 1. If the previous ScanKey scan rows multiple blocks,
+    //    and also the final block has 1024 rows just right.
+    // 2. The current ScanKey scan rows with number less than one block.
+    // Under the two conditions, if not seek block, the position
+    // of prefix shortkey columns is wrong.
+    _need_to_seek_block = true;
+
     return OLAP_SUCCESS;
 }
 
@@ -832,7 +841,7 @@ OLAPStatus SegmentReader::_create_reader(size_t* buffer_size) {
 
 OLAPStatus SegmentReader::_seek_to_block_directly(
         int64_t block_id, const std::vector<uint32_t>& cids) {
-    if (!config::block_seek_position && _at_block_start && block_id == _current_block_id) {
+    if (!_need_to_seek_block && block_id == _current_block_id) {
         // no need to execute seek
         return OLAP_SUCCESS;
     }
@@ -861,7 +870,7 @@ OLAPStatus SegmentReader::_seek_to_block_directly(
         }
     }
     _current_block_id = block_id;
-    _at_block_start = true;
+    _need_to_seek_block = false;
     return OLAP_SUCCESS;
 }
 
@@ -933,7 +942,7 @@ OLAPStatus SegmentReader::_load_to_vectorized_row_batch(
     if (size == _num_rows_in_block) {
         _current_block_id++;
     } else {
-        _at_block_start = false;
+        _need_to_seek_block = true;
     }
 
     _stats->blocks_load++;

be/src/olap/rowset/segment_reader.h

@@ -306,9 +306,9 @@ class SegmentReader {
 
     bool _eof;                             // eof标志
 
-    // If this field is false, client must to call seek_to_block before
+    // If this field is true, client must to call seek_to_block before
     // calling get_block.
-    bool _at_block_start = false;
+    bool _need_to_seek_block = true;
 
     int64_t _end_block;                           // 本次读取的结束块
     int64_t _current_block_id = 0;                       // 当前读取到的块

be/test/olap/delete_handler_test.cpp

@@ -54,6 +54,9 @@ void set_up() {
     std::vector<StorePath> paths;
     paths.emplace_back(config::storage_root_path, -1);
 
+    config::min_file_descriptor_number = 65536;
+    config::max_file_descriptor_number = 131072;
+
     doris::EngineOptions options;
     options.store_paths = paths;
     doris::StorageEngine::open(options, &k_engine);