Performance

• Hot region consumes too much memory #5692
• Imporve the preformance of hot cache #3725
• statistics: sometimes history hot region also consumes a lot of memory #5713
• speed up to create operator in big cluster

Statistic

• Show CPU usage for each Region  #4042
• Hot Read/Write Rate Bytes statistic is unstable and jitter #2497
• Entry number should not be suddenly increased a lot #4481
• The statistics of read keys does not match the RocksDB keys flows #3158
• Reduce the impact of GC in hot region scheduler #3084
• statistics: collect write region info from store heartbeat #5717

Why no hot-scheduler operator?

• Imbalanced read results in instability of sysbench select_random_ranges/select_random_points performance (-25%~8%) #4529
• Operator timeout due to add learner too slow tikv#13461
• unbalanced bytes and keys rate between auto scaling node with others #3098
• hot scheduler is not fast enough for insert workload #8944
• PD can't schedule well with the simple sysbench workload #4041
• scheduler: hot region scheduler doesn't work in v2 policy #6645

Why has hot-scheduler operator?

• Robustness of hotspot scheduling under high load #3919
• scheduler: unexpected increase in hot scheduler v2 #5559
• Reduce hot read region scheduling during rolling upgrade TiKV #3298
• Hot-region-scheduler should be more stable if the loads are not stable  #6297
• scheduler: cache history loads in hot region scheduler #6314

Reduce schedule cost

• Guaranteed priority of Hot Region Operator #5465
• Consider priority to schedule hot regions of smaller size with others same #3852
• Reduce the block cache miss when balancing the hot region #2999
• Consider priority to fix the Hot Region  #3842
• consider moving hot regions first if we delete a TiKV #662

Dimension selection

• scheduler: introduce read cpu dimension #5718
• QPS drop more than 30% after upgrade from 5.2.3 to master with sysbench select_random_point. #4612
• enhance: dimension of hot scheduler  #4513
• Adjustable Priority Multidimensional Hot Scheduler Tasklist #3869
• Improve balance region scheduler to prefer select regions with write traffic #1432

Dimension conflict

• Hot Region: Resolve dimensional conflict problem and enhance stability  #5446
• Verify better hot scheduler default parameter values #4332
• Improvement of hot scheduler when meets dimension conflict #4949
• Improve the search space and dimension conflicts resulting in hot scheduler unschedulable #4628

Expectation and variance

• UCP: support store weight for hot region scheduler  #2168
• scheduler: better expectation and variance in hot region scheduler #5545
• scheduler: remove exp from hot scheduler #3830

split

• scheduler: move hot region should be splitted if it's size is too large .  #6281
• schedulers: region should splitted if region is too hot. #6618

Load base split

• Provide feedback functionality for TiKV #5267
• hot scheduler can create split operator

Metrics, logs and pd-ctl

• pd-ctl: hot write store query is zero when run bench #5643
• hot region grafana should be more clearier  #4350
• hot metrics still stop last value when a store meets pending #4993
• No update logs when set balance-hot-region-scheduler attributes #4607
• Record the reason why hot region operators are created #3049
• statistics: clean store info in hot cache when store is deleted #5733
• add cpu usage in hot region statistics and TIDB_HOT_REGIONS_HISTORY #8817

Config

• Reduce the number of hot region scheduler configurations #3943
• statistics, scheduler: unreasonable two sets of hotspot thresholds #5155
• statistics: replace const StoreHeartBeatReportInterval with tikv config #5732
• When upgrading from version 5.x to version 6.4, the hot-scheduler config is not as expected #5698
• The config is not persisted after the balance-hot-region-scheduler is started #5701
• statistics: smaller query thresholds #5897

Refactor

• Refactor hot region related modules #4477
• scheduler: introduce base class of hot region scheduler #5708
• statistics: reduce unnecessary copy when collect hot metrics #5709

Tools

• Add analysis for heartbeat interval #3222
• Add distribution of region bytes, keys, and QPS to the telemetry #3030
• Modeling the hot region scheduler #2501

Related Problem

Cold peer

• PD can not identify hot region if region report interval greater than 10s period #4351
• scheduler: balance hot stores formed a lot of cold peers  #5602

Collect TiDB Info

• cached table and copr cache