Region performance tuning

This article introduces how to adjust Region Size and other methods Region Performance tuning and how to do it in large Region Next use bucket Concurrent query optimization .

summary

TiKV Automatically process the underlying data Fragmentation , All data are in accordance with key The scope of is divided into several Region. When a Region After the size of exceeds a certain limit ,TiKV Will split it into multiple Region.

In the scenario of a large amount of data , There may be Region Too many , This will bring more resource overhead and lead to Performance fallback The problem of . Under a fixed amount of data ,Region The bigger it is , be Region The smaller the number . from v6.1.0 Start ,TiDB Support setting custom Region size .Region The default size is about 96 MiB, Increasing it can reduce Region Number .

Turn on  Hibernate Region  or  Region Merge  It can also reduce too much Region Performance overhead .

Use  region-split-size  adjustment Region size

Warning

Customize Region The size is in TiDB v6.1.0 Experimental characteristics introduced , It is not recommended to configure . The risks of using this feature include ：

• More prone to performance jitter .
• Query performance fallback , In particular, the performance of large-scale data query will fall back .
• Slow scheduling .

Region The size of can pass through  coprocessor.region-split-size  Set it up . The recommended Region The size is 96 MiB、128 MiB、256 MiB.region-split-size  The bigger it is , Performance will be more prone to jitter . It is not recommended that Region The size setting exceeds 1 GiB, It is strongly recommended not to exceed 10 GiB. If you use TiFlash, be Region The size cannot exceed 256 MiB. If you use Dumpling Tools , be Region The size cannot exceed 1 GiB.Region After you turn it up , Use Dumpling Tool time , Need to reduce concurrency , otherwise TiDB There will be OOM The risk of .

Use bucket Increase concurrency

Warning

Currently, this function is an experimental feature , It is not recommended for use in a production environment .

Region After you turn it up , To increase query concurrency , Should be set up  coprocessor.enable-region-bucket  by  true. This configuration will change each Region Divide into smaller intervals bucket, And take this smaller interval as the concurrent query unit , To improve the concurrency of scanning data .bucket The size of the  coprocessor.region-bucket-size  To control , The default value is  96MiB.