[Daddyoftwins1]

This PR improves file-cache search performance for the tested large-cache workloads while preserving the compared search results and maintaining compatibility with the current upstream architecture.

The changes update the search implementation for current connection handling and sharding, integrate optional indexes into Nextcloud's native missing-index lifecycle, and add upstream-native regression tests.

Performance

The patched implementation was compared with unchanged Nextcloud 36 development code at the same commit. The controlled synthetic benchmark used 100,000 target files and approximately 504,000 total cache rows.

Selected warm-cache median results:

Search	Unchanged	Patched	Result
Common name	76.2 ms	5.9 ms	12.9× faster
Recent selective name	64.9 ms	8.8 ms	7.4× faster
No matches	64.3 ms	24.8 ms	2.6× faster
Small-cache control	3.1 ms	5.6 ms	2.5 ms slower

All search results compared within the benchmark fixture matched. Adding indexes alone did not produce a consistent material improvement in this experiment; the search implementation changes were necessary for the measured gains.

These are controlled synthetic benchmarks. Performance will vary with cache size, workload, and environment. The small-cache slowdown is a known tradeoff.

Validation

• The compatibility regression suite completed with 126 tests, 626 assertions, and one Normalizer-related environment skip.
• Formatting, syntax, and autoloader checks passed.
• I personally completed live testing of the change on my running Nextcloud system.

Please review the search implementation, optional-index lifecycle, connection and sharding compatibility, and regression coverage.

AI assistance

I was assisted by Codex AI with implementation, automated testing, benchmarking, and code review.

Evidence

Code and benchmark evidence bundle