Recs rebuild WS1a: unified Recommendations reader + de-dupe (data layer)

[erikdarlingdata]

Scope

WS1a of the recommendations-engine rebuild: the data layer only for the unified Recommendations surface. No XAML / no WPF control / no tab wiring (that is WS1b, a later PR), so nothing renders yet — this is foundational, like prior PRs. Dashboard-only.

Builds a unified reader + view-model + de-dupe over the two producers for a server:

• Engine findings (config.analysis_findings) read via the existing SqlServerFindingStore.GetRecentFindingsAsync (no new raw SQL for the findings read).
• Legacy config recs (config.critical_issues) read via the existing DatabaseService.GetCriticalIssuesAsync.

Implements decisions D1 (single surface / hybrid producers), C3 (concrete cross-store de-dupe), D3 (advise/copy-paste + Apply tiering — Apply carried via the persisted action, surfaced in WS1b).

Key design note (verified against the code)

GetRecentFindingsAsync returns remediation_action_json (the built RemediationAction, D2) but NOT the ephemeral DrillDown. So on read the engine Setting and copy-paste SQL are derived from the persisted RemediationAction (FactKey=="RCSI" → Rcsi; DbConfigTargets → AutoShrink/AutoClose/PageVerify), not from a drill-down that is absent. The copy-paste ALTER DATABASE statements are rebuilt from those persisted targets with the same QUOTENAME-equivalent bracketing + SET-clause literals the executor uses.

File:line changelog

New files (4; +1067):

• Dashboard/Services/Recommendations/RecommendationItem.cs:1 — the unified VM (plain DTO, no WPF dep) + the CanonicalSeverity (Info/Warning/Critical), RecommendationSource (Engine/Legacy), RecommendationSetting (None/AutoShrink/AutoClose/QueryStore/Rcsi/PageVerify) enums. Fields: CanonicalSeverity, RawSeverity, Database, Title, ProblemArea, AdviceText, CopyPasteSql, Remediation (RemediationAction?), Source, StoryPathHash, Setting.
• Dashboard/Services/Recommendations/RecommendationDeduper.cs:1 — the PURE static core: FromEngineSeverity (>=1.5 Critical / >=0.75 Warning / else Info, :30), FromLegacySeverity (text → band, :46), Merge (the de-dupe over two mapped lists, :79), PreferEngineFor (collision precedence, :135), NormalizeDatabase (trim/lower, :170), CompareForDisplay (severity desc total order, :181).
• Dashboard/Services/Recommendations/RecommendationsReader.cs:1 — GetRecommendationsAsync(serverId, serverName, hoursBack, limit) (:59) reads both stores, maps, de-dupes, sorts. MapEngineFinding (:83), MapLegacyIssue (:111), SettingFromAction (:145), SettingFromDbConfig (:177), SettingFromLegacy (ALTER-keyword parse for the two config problem-areas only, :194), BuildCopyPasteFromAction (:225).
• Dashboard.Tests/RecommendationDeduperTests.cs:1 — 41 xUnit tests (no DB).

No existing files modified (git diff --stat vs HEAD is empty).

De-dupe (C3) summary

• Key = (normalized database [trim/lower], Setting). Setting=None rows never de-dupe (memory pressure, server-level MAXDOP/CTFP, dumps, CPU/wait findings) — all pass through.
• Engine Setting from the persisted action; legacy Setting only for problem_area ∈ {Database Configuration, Query Store Configuration}, derived from the investigate_query ALTER keyword (AUTO_SHRINK/AUTO_CLOSE/QUERY_STORE) — the free-text message is never parsed.
• Collision precedence: AutoShrink/AutoClose → Engine (Apply-capable); QueryStore → Legacy (engine has no QS Apply today). Rcsi/PageVerify are engine-only and never collide.

Test counts (REAL — dotnet test --no-build -c Debug)

• Build: dotnet build PerformanceMonitor.sln -c Debug → 0 errors (1 pre-existing warning in RemediationTests.cs, not from this PR).
• Dashboard.Tests: Passed 279, Failed 0, Skipped 0 (of which 41 are the new RecommendationDeduperTests).
• Lite.Tests: Passed 360, Failed 0, Skipped 0.

New-test coverage: AUTO_SHRINK in both stores same DB → one row Source=Engine; AUTO_CLOSE same; QUERY_STORE in both (synthesized engine QS row) → one row Source=Legacy; same setting different DBs → both kept; memory-pressure legacy (None) + engine CPU finding → both kept; severity boundary cases (0.74→Info, 0.75→Warning, 1.49→Warning, 1.5→Critical, plus the text mapping); plus the setting/copy-paste mapper helpers.

Needs integration verification (live DB — NOT unit-tested here)

• The actual two-store read end-to-end: RecommendationsReader.GetRecommendationsAsync against a real config.analysis_findings + config.critical_issues (the unit tests exercise the pure de-dupe + the mappers with synthesized inputs, never the DB).
• That a real engine DB_CONFIG / RCSI finding round-trips its remediation_action_json such that SettingFromAction + BuildCopyPasteFromAction produce the expected Setting and ALTER text from the persisted action on read.
• That a real legacy AUTO_SHRINK/AUTO_CLOSE/QUERY_STORE row carries the expected problem_area + investigate_query keyword for the de-dupe to fire (and a real per-DB AUTO_SHRINK present in BOTH stores collapses to the single Engine row).
• Sort/severity-band behavior over real mixed-producer data.

🤖 Generated with Claude Code