ClickHouse/dbms/src/DataStreams/SquashingTransform.h

56 lines
1.6 KiB
C++
Raw Normal View History

#pragma once
#include <Core/Block.h>
namespace DB
{
2016-07-07 01:57:48 +00:00
/** Merging consecutive passed blocks to specified minimum size.
*
* (But if one of input blocks has already at least specified size,
* then don't merge it with neighbours, even if neighbours are small.)
*
* Used to prepare blocks to adequate size for INSERT queries,
* because such storages as Memory, StripeLog, Log, TinyLog...
* store or compress data in blocks exactly as passed to it,
* and blocks of small size are not efficient.
*
* Order of data is kept.
*/
class SquashingTransform
{
public:
/// Conditions on rows and bytes are OR-ed. If one of them is zero, then corresponding condition is ignored.
2019-08-03 11:02:40 +00:00
SquashingTransform(size_t min_block_size_rows_, size_t min_block_size_bytes_);
/// When not ready, you need to pass more blocks to add function.
struct Result
{
bool ready = false;
2018-09-08 19:23:48 +00:00
MutableColumns columns;
Result(bool ready_) : ready(ready_) {}
2019-08-03 11:02:40 +00:00
Result(MutableColumns && columns_) : ready(true), columns(std::move(columns_)) {}
};
/** Add next block and possibly returns squashed block.
* At end, you need to pass empty block. As the result for last (empty) block, you will get last Result with ready = true.
*/
2018-09-08 19:23:48 +00:00
Result add(MutableColumns && columns);
private:
size_t min_block_size_rows;
size_t min_block_size_bytes;
2018-09-08 19:23:48 +00:00
MutableColumns accumulated_columns;
2018-09-08 19:23:48 +00:00
void append(MutableColumns && columns);
2018-09-08 19:23:48 +00:00
bool isEnoughSize(const MutableColumns & columns);
bool isEnoughSize(size_t rows, size_t bytes) const;
};
}