Refactored Prepared-Statement Cache Design (Lock-Free Hot Path) - Part 2

[rahim-kanji]

Summary

Concurrency and Memory Management

• Lock-Free Ref Counting: Replaced global mutex-protected integer reference counts with std::atomic<uint32_t> within PgSQL_STMT_Global_info. This eliminates lock contention during statement referencing.
• Modern Ownership: Adopted std::shared_ptr<const PgSQL_STMT_Global_info> for global and local storage, providing automatic, thread-safe memory and lifecycle management.
• Memory Optimization: Removed redundant auxiliary maps:
  • global_id_to_stmt_names (from local statement managers)
  • map_stmt_id_to_info (from global statement manager)
    This significantly reduces overall memory overhead.
• Optimized Purging: Statement removal logic was simplified for efficiently identifying and cleaning up unused statements.

Hot Path Performance (BIND, DESCRIBE, EXECUTE)

The refactor targets operations that occur frequently to minimize latency:

• Bypassed Global Lookups: Local session maps now store the shared_ptr directly, removing the need to acquire the global lock and search the global map during hot path operations.
• Direct Refcount Manipulation: Reference count modification functions now operate directly on the passed statement object, eliminating the overhead of searching the global map to find the object pointer based on statement ID.

Safety and Protocol Logic (PARSE)

• Efficient Statement Reuse (Local Fast Path): Implemented a local fast path check for the unnamed statement (""). This allows immediate reuse of an identical query (same hash) upon re-parse, bypassing global processing and locks.

Cleanup

Class Rename:

• PgSQL_STMT_Manager_v14 -> PgSQL_STMT_Manager
• PgSQL_STMTs_local_v14 -> PgSQL_STMTs_Local

Removed unused variables from PostgreSQL module:

• pgsql-client_session_track_gtid
• pgsql-enable_client_deprecate_eof
• pgsql-enable_server_deprecate_eof
• pgsql-enable_load_data_local_infile
• pgsql-log_mysql_warnings_enabled
• pgsql-default_session_track_gtids
• pgsql-handle_warnings

Introduced Xoshiro128++ RNG

Switched from rand() to the Xoshiro128++ RNG to improve performance and randomness.

Prepared Statement Cache Refactoring - Benchmark Results

Benchmark setup: pgbench via ProxySQL, 16 clients, 60s duration, query mode: prepared.
Metrics: TPS (transactions per second) and average latency (ms). Each value is averaged over two runs.

pgsql-thread (ProxySQL)	Threads (pgbench)	TPS Before (Pre-Refactor)	TPS After (Post-Refactor)	Latency Before (ms)	Latency After (ms)
1	1	3944	5078	4.06	3.16
1	2	3737	4751	4.28	3.37
1	4	3435	4454	4.66	3.59
1	8	3381	4327	4.73	3.70
1	16	3286	4312	4.87	3.71
2	1	6263	7377	2.56	2.17
2	2	5784	6622	2.77	2.42
2	4	5180	6557	3.10	2.44
2	8	4913	6199	3.26	2.58
2	16	4902	6219	3.26	2.57

---

Observations

1. Throughput (TPS) improvements:

   • pgsql-thread=1: +28–31%
   • pgsql-thread=2: +14–27%
   • Gains are largest at higher thread counts, showing better concurrency handling.

2. Latency improvements:

   • pgsql-thread=1: -21–24%
   • pgsql-thread=2: -13–21%
   • Lower latency indicates more efficient reuse of prepared statements.

---

Merged optimizations from PR #5196:

This PR includes several performance improvements and protocol handling enhancements:

• Replaced std::string with char[] in critical path.
• Introduced inline functions for ASCII whitespace detection and uint32 to string conversion.

Buffered response handling (9eb934e)

• Buffer query responses until the Extended Query frame completes.
• Early sending of results occurs only if the resultset threshold is reached, reducing partial responses and improving consistency.

PostgreSQL pipeline sync backport (9fa3d75)

• Backported PQsendPipelineSync from PostgreSQL 17.

Closes #5211

[renecannao]

retest this please

[sonarqubecloud]

Quality Gate passed

Issues
2 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[rahim-kanji]

retest this please