dbms: correctly determine mark to end of previous block [#MTRSADMIN-1093]

dbms/src/Storages/MergeTree/MergeTreePartChecker.cpp

@@ -11,6 +11,7 @@
 #include <DB/IO/CompressedReadBuffer.h>
 #include <DB/IO/HashingReadBuffer.h>
 #include <DB/Columns/ColumnsNumber.h>
+#include <statdaemons/ext/scope_guard.hpp>
 
 
 namespace DB
@@ -115,6 +116,12 @@ struct Stream
 		readIntBinary(mrk_mark.offset_in_compressed_file, mrk_hashing_buf);
 		readIntBinary(mrk_mark.offset_in_decompressed_block, mrk_hashing_buf);
 
+		/// На всякий случай, сохраним смещение в файле и размер предыдущего блока.
+		SCOPE_EXIT(
+			prev_offset_in_compressed_file = mrk_mark.offset_in_compressed_file;
+			prev_buffer_size = uncompressed_hashing_buf.buffer().size();
+		);
+
 		bool has_alternative_mark = false;
 		MarkInCompressedFile alternative_data_mark;
 		MarkInCompressedFile data_mark;
@@ -138,6 +145,18 @@ struct Stream
 			if (uncompressed_hashing_buf.eof())
 				return;
 		}
+		else if (uncompressed_hashing_buf.offset() == 0)
+		{
+			/// Восстановим засечку на конец предыдущего блока по сохраненным данным
+			has_alternative_mark = true;
+			alternative_data_mark.offset_in_compressed_file = prev_offset_in_compressed_file;
+			alternative_data_mark.offset_in_decompressed_block = prev_buffer_size;
+
+			if (mrk_mark == alternative_data_mark)
+				return;
+		}
+
+		std::cout << "mrk_mark " << mrk_mark.offset_in_compressed_file << ' ' << mrk_mark.offset_in_decompressed_block << std::endl;
 
 		data_mark.offset_in_compressed_file = compressed_hashing_buf.count() - uncompressing_buf.getSizeCompressed();
 		data_mark.offset_in_decompressed_block = uncompressed_hashing_buf.offset();
@@ -161,6 +180,10 @@ struct Stream
 		checksums.files[name + ".mrk"] = MergeTreeData::DataPart::Checksums::Checksum(
 			mrk_hashing_buf.count(), mrk_hashing_buf.getHash());
 	}
+
+private:
+	size_t prev_offset_in_compressed_file{};
+	size_t prev_buffer_size{};
 };
 
 /// Возвращает количество строк. Добавляет в checksums чексуммы всех файлов столбца.