- Add the
Scatter Rangescheduler to balance Regions with the specified key range
- Optimize the scheduling of Merge Region to prevent the newly split Region from being merged
- Add Learner related metrics
- Fix the issue that the scheduler is mistakenly deleted after restart
- Fix the error that occurs when parsing the configuration file
- Fix the issue that the etcd leader and the PD leader are not synchronized
- Fix the issue that Learner still appears after it is closed
- Fix the issue that Regions fail to load because the packet size is too large
- Support using pd-ctl to scatter specified Regions for manually adjusting hotspot Regions in some cases
- Improve configuration check rules to prevent unreasonable scheduling configuration
- Optimize the scheduling strategy when a TiKV node has insufficient space so as to prevent the disk from being fully occupied
- Optimize hot-region scheduler execution efficiency and add more metrics
- Optimize Region health check logic to avoid generating redundant schedule operators
- Support adding the learner node
- Optimize the Balance Region Scheduler to reduce scheduling overhead
- Adjust the default value of
schedule-limitconfiguration - Fix the compatibility issue when adding a new scheduler
- Fix the issue of allocating IDs frequently
- Support splitting Region manually to handle the hot spot in a single Region
- Optimize metrics
- Fix the issue that the label property is not displayed when
pdctlrunsconfig show all
- Support Region Merge, to merge empty Regions or small Regions after deleting data
- Ignore the nodes that have a lot of pending peers during adding replicas, to improve the speed of restoring replicas or making nodes offline
- Optimize the scheduling speed of leader balance in scenarios of unbalanced resources within different labels
- Add more statistics about abnormal Regions
- Fix the frequent scheduling issue caused by a large number of empty Regions