Reduce resource-selector benchmark regressions via selector parse caching and canonical fast-path

[Copilot]

This PR addresses the benchmark regressions introduced around resource selector query construction and config lookup paths (ResourceSelectorConfigs/*, ResourceSelectorQueryBuild/tags) and trims repeated selector-processing overhead in hot loops.

• Resource selector clause hot-path optimization

  • Added a cache for parsed PEG expressions used by SetResourceSelectorClause to avoid repeatedly parsing identical selector PEG strings.
  • Cached payload stores both parsed query field and flattened fields used for agent_id / deleted_at detection.

• Selector requirements parse deduplication

  • Added a cache for parsed label/tag/field selector requirements to avoid repeated labels.Parse(...).Requirements() for the same selector strings.
  • Applied to TagSelector, LabelSelector, and FieldSelector handling in SetResourceSelectorClause.

• Canonicalization fast-path

  • Added an early return in ResourceSelector.Canonical() when no wildcard is present, skipping normalization work on common non-wildcard selectors.
  • Added focused coverage for wildcard detection and no-wildcard canonical stability.

• Safety hardening for cached values

  • Replaced unsafe cache type assertions with checked assertions + cache eviction on mismatch to prevent panic from malformed cache entries.

func getSelectorRequirements(selector string) ([]labels.Requirement, error) {
	if value, ok := resourceSelectorLabelRequirementsCache.Get(selector); ok {
		if cached, ok := value.(parsedSelectorRequirements); ok {
			return cached.requirements, nil
		}
		resourceSelectorLabelRequirementsCache.Delete(selector)
	}

	parsedSelector, err := labels.Parse(selector)
	if err != nil {
		return nil, err
	}
	requirements, _ := parsedSelector.Requirements()
	resourceSelectorLabelRequirementsCache.SetDefault(selector, parsedSelectorRequirements{requirements: requirements})
	return requirements, nil
}

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[github-actions]

Benchstat (Other)

Base: 6e01f646baf3a630e0b92d7a9602a52202b1c131
Head: d37d0bd98cfe925d4ee08b5a4475c5c58d728dce

✅ 5 improvement(s)

Benchmark	Base	Head	Change	p-value
ResourceSelectorQueryBuild/name_and_type-4	63.30µ	21.41µ	-66.18%	0.002
ResourceSelectorQueryBuild/name-4	43.82µ	16.42µ	-62.52%	0.002
ResourceSelectorConfigs/name-4	208.3µ	174.7µ	-16.13%	0.009
ResourceSelectorConfigs/name_and_type-4	227.2µ	191.6µ	-15.64%	0.002
ResourceSelectorQueryBuild/tags-4	16.90µ	15.80µ	-6.51%	0.004

Full benchstat output

goos: linux
goarch: amd64
pkg: github.com/flanksource/duty/bench
cpu: AMD EPYC 7763 64-Core Processor                
                                                       │ bench-base.txt │           bench-head.txt            │
                                                       │     sec/op     │    sec/op     vs base               │
InsertionForRowsWithAliases/external_users.aliases-4       591.3µ ± 19%   600.2µ ±  7%        ~ (p=0.310 n=6)
InsertionForRowsWithAliases/config_items.external_id-4     1.136m ±  9%   1.128m ± 18%        ~ (p=0.937 n=6)
ResourceSelectorConfigs/name-4                             208.3µ ±  2%   174.7µ ± 19%  -16.13% (p=0.009 n=6)
ResourceSelectorConfigs/name_and_type-4                    227.2µ ±  5%   191.6µ ± 15%  -15.64% (p=0.002 n=6)
ResourceSelectorConfigs/tags-4                             27.89m ±  6%   29.25m ±  6%        ~ (p=0.132 n=6)
ResourceSelectorQueryBuild/name-4                          43.82µ ±  4%   16.42µ ±  1%  -62.52% (p=0.002 n=6)
ResourceSelectorQueryBuild/name_and_type-4                 63.30µ ±  3%   21.41µ ±  1%  -66.18% (p=0.002 n=6)
ResourceSelectorQueryBuild/tags-4                          16.90µ ±  3%   15.80µ ±  6%   -6.51% (p=0.004 n=6)
geomean                                                    283.4µ         209.3µ        -26.13%

[github-actions]

Benchstat (RLS)

Base: 6e01f646baf3a630e0b92d7a9602a52202b1c131
Head: d37d0bd98cfe925d4ee08b5a4475c5c58d728dce

📊 1 minor regression(s) (all within 5% threshold)

Benchmark	Base	Head	Change	p-value
RLS/Sample-15000/catalog_changes/With_RLS-4	129.8m	131.0m	+0.89%	0.002

✅ 4 improvement(s)

Benchmark	Base	Head	Change	p-value
RLS/Sample-15000/change_types/With_RLS-4	5.382m	5.229m	-2.84%	0.002
RLS/Sample-15000/analysis_types/With_RLS-4	3.944m	3.906m	-0.95%	0.041
RLS/Sample-15000/config_types/With_RLS-4	125.5m	124.4m	-0.81%	0.015
RLS/Sample-15000/configs/With_RLS-4	123.6m	122.6m	-0.80%	0.002

Full benchstat output

goos: linux
goarch: amd64
pkg: github.com/flanksource/duty/bench
cpu: AMD EPYC 7763 64-Core Processor                
                                               │ bench-base.txt │          bench-head.txt           │
                                               │     sec/op     │   sec/op     vs base              │
RLS/Sample-15000/catalog_changes/Without_RLS-4      5.311m ± 1%   5.316m ± 3%       ~ (p=0.589 n=6)
RLS/Sample-15000/catalog_changes/With_RLS-4         129.8m ± 1%   131.0m ± 1%  +0.89% (p=0.002 n=6)
RLS/Sample-15000/config_changes/Without_RLS-4       5.329m ± 1%   5.273m ± 2%       ~ (p=0.093 n=6)
RLS/Sample-15000/config_changes/With_RLS-4          128.3m ± 1%   127.7m ± 1%       ~ (p=0.310 n=6)
RLS/Sample-15000/config_detail/Without_RLS-4        3.966m ± 2%   4.002m ± 1%       ~ (p=0.180 n=6)
RLS/Sample-15000/config_detail/With_RLS-4           124.1m ± 0%   124.3m ± 1%       ~ (p=0.240 n=6)
RLS/Sample-15000/config_names/Without_RLS-4         12.61m ± 1%   12.67m ± 1%       ~ (p=0.485 n=6)
RLS/Sample-15000/config_names/With_RLS-4            124.6m ± 2%   125.1m ± 5%       ~ (p=0.394 n=6)
RLS/Sample-15000/config_summary/Without_RLS-4       60.23m ± 3%   59.77m ± 1%       ~ (p=0.132 n=6)
RLS/Sample-15000/config_summary/With_RLS-4          741.4m ± 1%   738.6m ± 1%       ~ (p=0.240 n=6)
RLS/Sample-15000/configs/Without_RLS-4              7.110m ± 4%   7.080m ± 1%       ~ (p=0.132 n=6)
RLS/Sample-15000/configs/With_RLS-4                 123.6m ± 2%   122.6m ± 0%  -0.80% (p=0.002 n=6)
RLS/Sample-15000/analysis_types/Without_RLS-4       3.910m ± 2%   3.912m ± 1%       ~ (p=0.937 n=6)
RLS/Sample-15000/analysis_types/With_RLS-4          3.944m ± 3%   3.906m ± 2%  -0.95% (p=0.041 n=6)
RLS/Sample-15000/analyzer_types/Without_RLS-4       3.736m ± 1%   3.748m ± 1%       ~ (p=0.180 n=6)
RLS/Sample-15000/analyzer_types/With_RLS-4          3.770m ± 3%   3.793m ± 2%       ~ (p=0.394 n=6)
RLS/Sample-15000/change_types/Without_RLS-4         5.262m ± 1%   5.298m ± 2%       ~ (p=0.310 n=6)
RLS/Sample-15000/change_types/With_RLS-4            5.382m ± 0%   5.229m ± 2%  -2.84% (p=0.002 n=6)
RLS/Sample-15000/config_classes/Without_RLS-4       3.302m ± 1%   3.309m ± 1%       ~ (p=0.310 n=6)
RLS/Sample-15000/config_classes/With_RLS-4          123.2m ± 1%   124.2m ± 0%       ~ (p=0.065 n=6)
RLS/Sample-15000/config_types/Without_RLS-4         3.981m ± 0%   3.947m ± 2%       ~ (p=0.240 n=6)
RLS/Sample-15000/config_types/With_RLS-4            125.5m ± 1%   124.4m ± 0%  -0.81% (p=0.015 n=6)
geomean                                             19.24m        19.21m       -0.17%