Lower block sizes and look what will happen #9248

This commit is contained in:
Alexey Milovidov 2020-07-12 04:05:09 +03:00
parent 74afd5d6d1
commit bc8e7e04df

View File

@ -21,14 +21,14 @@
/** Which blocks by default read the data (by number of rows).
* Smaller values give better cache locality, less consumption of RAM, but more overhead to process the query.
*/
#define DEFAULT_BLOCK_SIZE 65536
#define DEFAULT_BLOCK_SIZE 65505 /// 65536 minus 16 + 15 bytes padding that we usually have in arrays
/** Which blocks should be formed for insertion into the table, if we control the formation of blocks.
* (Sometimes the blocks are inserted exactly such blocks that have been read / transmitted from the outside, and this parameter does not affect their size.)
* More than DEFAULT_BLOCK_SIZE, because in some tables a block of data on the disk is created for each block (quite a big thing),
* and if the parts were small, then it would be costly then to combine them.
*/
#define DEFAULT_INSERT_BLOCK_SIZE 1048576
#define DEFAULT_INSERT_BLOCK_SIZE 1048545 /// 1048576 minus 16 + 15 bytes padding that we usually have in arrays
/** The same, but for merge operations. Less DEFAULT_BLOCK_SIZE for saving RAM (since all the columns are read).
* Significantly less, since there are 10-way mergers.