[Enhancement Request] Fine-Grained GTID Support for Improved Read-After-Write Performance

[gerayking]

Background

The current implementation of the read_after_write consistency feature in the system relies on waiting for the execution of the last global transaction identifier (GTID), indiscriminately applying this method across SQL operations regardless of their data dependencies. This broad-stroke approach leads to unnecessarily high latency and decreased throughput for read-after-write operations, particularly when these operations do not interact with the same table. The lack of differentiation significantly hinders performance, especially in use cases where operations could otherwise proceed in parallel without data consistency issues.

Proposal

Implement Table-Level Read-After-Write Support: Introduce the capability for the system to intelligently discern operations across different tables, allowing for parallel processing of read-after-write operations where there are no direct data dependencies. This refinement is anticipated to substantially lower wait times for operations not confined to the same table, enhancing responsiveness.

Provide Configuration Options for Global and Table Levels: Offer users the ability to adjust read-after-write settings specifically for global and table levels. This granularity in configuration would empower users to tailor performance optimization strategies more precisely to their application's operational characteristics and requirements.

Performance Analysis for Global and Table Level Settings: Undertake a comprehensive analysis to evaluate the performance implications of utilizing global versus table-level settings for read-after-write operations. The insights gained from this analysis would equip users with the knowledge to make informed decisions, optimizing their configurations for either broader or more targeted performance improvements based on their specific scenarios.