19 About Dynamic Snitching 19 About Client Requests in Cassandra 19 About Write Requests 20 About Multi-Data Center Write Requests 20 About Read Requests 21 Planning a Cassandra Cluster Deployment About Reads in Cassandra 57 About Data Consistency in Cassandra 58 Tunable Consistency for Client Requests 58 About Write Consistency 58 About Read Consistency 58 Choosing Client Consistency Levels 59 information is also used by Cassandra to avoid routing client requests to unreachable nodes whenever possible. (Cassandra can also avoid routing requests to nodes that are alive, but performing poorly, through

leader. When this feature is enabled, TiDB evenly sends the requests to all PD nodes when acquiring TSO. The PD follower then forwards the TSO requests to reduce the CPU pressure of PD leader. 31 Variable _wait_time �→ Newly added Sets the maximum waiting time for a batch saving operation when TiDB requests TSO from PD. The default value is 0, which means no additional waiting. tidb_tmp_table_max_size PD by enabling PD Follower Proxy and modifying the batch waiting time required when the PD client requests TSO in batches. This helps improve the overall scalability of the system. 43 – Support enabling

com/tidb/v8.4/system-variables#tidb_tso_ �→ client_rpc_mode-new-in-v840">Parallel mode for TSO requests In high-concurrency scenarios, you can use this feature to reduce the �→ wait time parallel batching modes for TSO requests, reducing TSO retrieval latency #54960 #8432 @MyonKeminta Before v8.4.0, when requesting TSO from PD, TiDB collects multiple TSO requests during a specific period and and processes them in batches serially to decrease the number of Remote Procedure Call (RPC) requests and reduce PD workload. In latency-sensitive scenarios, however, the performance of this serial batching

the capability of the PD cluster to �→ handle GetRegion and ScanRegions requests �→ in clusters with a large number of TiDB nodes and Regions, thereby �→ reducing the CPU pressure scheduling tasks. 37 If the cluster has many TiDB instances, and there is a high concurrency of requests for Region information, the CPU pressure on the PD leader increases further and might cause PD services feature is enabled, TiDB evenly distributes Region information requests to all PD servers, and PD followers can also handle Region requests, thereby reducing the CPU pressure on the PD leader. For more

com/tidb/v8.4/system-variables#tidb_tso_ �→ client_rpc_mode-new-in-v840">Parallel mode for TSO requests In high-concurrency scenarios, you can use this feature to reduce the �→ wait time parallel batching modes for TSO requests, reducing TSO retrieval latency #54960 #8432 @MyonKeminta Before v8.4.0, when requesting TSO from PD, TiDB collects multiple TSO requests during a specific period and and processes them in batches serially to decrease the number of Remote Procedure Call (RPC) requests and reduce PD workload. In latency-sensitive scenarios, however, the performance of this serial batching

when it is started with systemd #47442 @hawkingrei • TiKV • Fix the issue that retrying prewrite requests in the pessimistic transaction mode might cause the risk of data inconsistency in rare cases #11187 TiFlash. 3.4.5 Data processing With TiDB, you can simply enter SQL statements for query or write requests. For the tables with TiFlash replicas, TiDB uses the front-end optimizer to automatically choose connections in the connection pool. It is mainly used to reserve some connections to respond to sudden requests when the application is idle. You also need to configure it according to your application characteristics

batch processing strategy for KV (key-value) requests #55206 @zyguan TiDB fetches data by sending KV requests to TiKV. Batching and processing KV requests in bulk can significantly improve execution performance tikv- �→ client �→ . �→ batch �→ - �→ policy �→ Newly added Controls the batching strategy for requests from TiDB to TiKV. 51 Configuration file Configuration parame- ter Change type Description PD configuration #8450 @lhy1024 • Optimize the RU consumption behavior of large query read requests to reduce the impact on other requests #8457 @nolouch • Optimize the error message that returns when you misconfigure

com/tidb/v7.1/system- �→ variables#tidb_store_batch_size" target="_blank">batch aggregating �→ data requests (introduced in v6.6.0) This enhancement significantly reduces total RPCs in TiKV highly dispersed and the gRPC �→ thread pool has insufficient resources, batching coprocessor �→ requests can improve performance by more than 50%. 35 Load-based replica read In a read hotspot scenario, TiDB can redirect read requests for a �→ hotspot TiKV node to its replicas. This feature efficiently scatters �→ read hotspots

TiDB Server OOM when setting it too large. It means that the number of sessions that can execute requests concur- rently can be config- ured to a maxi- mum of 1048576. 50 Configuration file Configuration is deprecated. The new version of the Region replica selector is used by default when sending RPC requests to TiKV. • Starting from v8.2.0, the BR snapshot restore parameter --concurrency is dep- recated failed snapshot files in time #16976 @hbisheng • Fix the issue that highly concurrent Coprocessor requests might cause TiKV OOM #16653 @overvenus 58 • Fix the issue that changing the raftstore.period

• Add TiKV write rate limiter for background tasks to ensure that the latency of read and write requests is stable. 26 2.2.1 Compatibility changes Note: When upgrading from an earlier TiDB version performance �→ . �→ committer �→ - �→ concurrency �→ Modified Controls the concurrency number for requests related to commit operations in the commit phase of a single transaction. The default value is changed TiKV background tasks (TiKV Write Rate Limiter) To ensure the duration stability of read and write requests, TiKV Write Rate Limiter smoothes the write traffic of TiKV background tasks such as GC and Compaction