Tracking: General (non-vector) performance improvement PRs

[mykaul]

Overview

This issue tracks open PRs that improve general driver performance — query planning, metadata parsing, memory layout, and row decoding — excluding vector-specific optimizations already tracked in #746.

All PRs listed here are by @mykaul.

Directly performance-related PRs

HIGH priority

PR	Title	Area	Measured Impact
#651	Query planning policy optimization	Query planning	2.3x–6.4x faster policy decisions
#745	Metadata schema parsing optimizations	Schema parsing	1.98x faster full pipeline, 5.7x memory reduction per row
#740	namedtuple factory caching	Row decoding	135x faster class creation, 3.7x end-to-end

MEDIUM priority

PR	Title	Area	Measured Impact
#659	slots on connection/protocol frame classes	Memory	~5-7x memory reduction per frame instance
#630	column_encryption_policy fast-path check	Row decoding	Eliminates per-cell branch when encryption is unused

LOW-MEDIUM priority

PR	Title	Area	Measured Impact
#629	Remove TRIGGERS query for ScyllaDB	Schema refresh	1 fewer CQL round-trip per schema refresh

Test infrastructure

PR	Title	Area
#653	Query planning benchmarks	Benchmarks for #651

Proposed merge order

The PRs have minimal interdependencies. Suggested order based on risk and impact:

Order	PR	Rationale
1	#653	Benchmark infrastructure - no functional changes, enables measuring #651
2	#651	Query planning - high impact, self-contained
3	#629	Remove TRIGGERS - small, low-risk
4	#630	Encryption fast-path - small, low-risk
5	#740	namedtuple caching - high impact, touches row factory
6	#745	Metadata parsing - high impact, touches metadata.py extensively
7	#659	slots on frames - medium impact, touches connection/protocol classes

Relationship to #746

Issue #746 tracks vector search (VectorType) performance PRs. Some PRs in #746 benefit general workloads too (e.g., #734 remove copies on read path, #742 ParseDesc caching, #741 deserializer caching, #743 UTF8 decode). Those are tracked in #746 and not duplicated here.

Together, #746 and this issue cover all open performance-related PRs in the repository.