mance and stability • Reduce the write latency of TiKV by separating I/O operations from Raftstore thread pool (disabled by default) 2.2.1 Compatibility changes Note: When upgrading from an earlier TiDB Raft I/O tasks, which is the size of the StoreWriter thread pool. When you modify the size of this thread pool, refer to Performance tuning for TiKV thread pools. 34 Configuration file Configuration item write latency by separating I/O operations from Raftstore thread pool (disabled by default). For more information about tuning, see Tune TiKV Thread Pool Performance #10540 • PD – Add more types of write

is1000. TiKV snap- �→ generator �→ -pool- �→ size Newly added The size of the snap- �→ generator �→ thread pool. The default value is 2. 35 Configuration file Configuration Change type Description TiKV �→ enable_elastic_threadpool �→ Newly added Determines whether to enable or disable the elastic thread pool function. Enabling this configu- ration item can significantly improve TiFlash CPU utilization Newly added The allowable number of threads that process Raft, which is the size of the Raftstore thread pool. The default value is 4. TiDB Data Migration (DM) collation_compatible �→ Newly added The

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1127 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1134 11.2.4 Tune TiKV Memory file into multiple files #11595 • Move the snapshot garbage collection from Raftstore to background thread to prevent snapshot GC from blocking Raftstore message loops #11966 • Support dynamic setting of run -U root --db tpcc --host 127.0.0.1 --port �→ 4000 --time 1800s --warehouses 5000 --threads {{thread}} command to perform stress tests on TiDB via HAProxy. For each concurrency, the test takes 30 minutes

allow-concurrent-memtable-write: true server.grpc-concurrency: 6 readpool.unified.min-thread-count: 5 readpool.unified.max-thread-count: 20 readpool.storage.normal-concurrency: 10 pessimistic-txn.pipelined: true store-pool-size: 3 readpool.storage.normal-concurrency: 10 readpool.unified.max-thread-count: 20 readpool.unified.min-thread-count: 5 rocksdb.max-background-jobs: 8 server.grpc-concurrency: 6 storage.sc run -U root --db tpcc --host 127.0.0.1 --port �→ 4000 --time 300s --warehouses 5000 --threads {{thread}} command to perform stress tests on TiDB via HAProxy. 5. Extract the tpmC data of New Order from

allow-concurrent-memtable-write: true server.grpc-concurrency: 6 readpool.unified.min-thread-count: 5 50 readpool.unified.max-thread-count: 20 readpool.storage.normal-concurrency: 10 pessimistic-txn.pipelined: store-pool-size: 3 readpool.storage.normal-concurrency: 10 readpool.unified.max-thread-count: 20 readpool.unified.min-thread-count: 5 rocksdb.max-background-jobs: 8 server.grpc-concurrency: 6 storage.sc run -U root --db tpcc --host 127.0.0.1 --port �→ 4000 --time 300s --warehouses 5000 --threads {{thread}} command to perform stress tests on TiDB via HAProxy. 5. Extract the tpmC data of New Order from

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1392 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1400 11.2.4 Tune TiKV Memory the connection every time. • Regularly check whether the connection is available using a separate thread. • Send a test query regularly to keep alive connections. Different connection pool implementations is high, increase your maximum number of connec- tions to reduce the latency waiting time. Once a thread is blocked by latency, other threads can take over and continue processing. • If you have multiple

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1338 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1346 11.2.4 Tune TiKV Memory performance by caching external storage objects #13798 @YuJuncen • Run the CheckLeader in a dedicated thread to reduce TiCDC replication latency #13774 @overvenus • Support pull model for Checkpoints #13824 the connection every time. • Regularly check whether the connection is available using a separate thread. • Send a test query regularly to keep alive connections. Different connection pool implementations

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1378 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1386 11.2.4 Tune TiKV Memory the connection every time. • Regularly check whether the connection is available using a separate thread. • Send a test query regularly to keep alive connections. Different connection pool implementations is high, increase your maximum number of connec- tions to reduce the latency waiting time. Once a thread is blocked by latency, other threads can take over and continue processing. • If you have multiple

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1861 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1870 11.2.4 Tune TiKV Memory the connection every time. • Regularly check whether the connection is available using a separate thread. • Send a test query regularly to keep alive connections. Different connection pool implementations is high, increase your maximum number of connec- tions to reduce the latency waiting time. Once a thread is blocked by latency, other threads can take over and continue processing. • If you have multiple

· · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1374 11.2.3 Tune TiKV Thread Pool Performance · · · · · · · · · · · · · · · · · · · · · · · · · · · · 1382 11.2.4 Tune TiKV Memory RPCs in TiKV batch-get �→ operations. In situations where data is highly dispersed and the gRPC �→ thread pool has insufficient resources, batching coprocessor �→ requests can improve performance by more the connection every time. • Regularly check whether the connection is available using a separate thread. • Send a test query regularly to keep alive connections. Different connection pool implementations