Feature/multi query stores dashboard overview

[rferraton]

Multi-Query Store Overview Dashboard

Summary

Adds a new Query Store Overview dashboard that provides a consolidated, cross-database view of all Query Store-enabled databases on a SQL Server instance. This gives users a single pane of glass to compare workload metrics across databases before drilling down into individual database Query Store details.

Features

Overview Dashboard Layout

• Donut Chart — Shows Query Store state distribution across all databases (Read/Write, Read-Only, Off) with interactive tooltips and a popup listing databases by state
• Consolidated Time Slicer — Reuses the existing TimeRangeSlicerControl with data aggregated across all databases, allowing users to select a time range that filters all dashboard panels
• Wait Stats by Database — Stacked bar chart showing hourly wait totals per database (not per wait category), colored using a unified palette shared with the metric cards. Includes a dashed average reference line with smart unit formatting (ms/sec, s/sec, min/sec)
• Metric Bar Cards (Total & Average) — Two rows of 7 cards each (CPU, Duration, Executions, Reads, Writes, Physical Reads, Memory) showing horizontal stacked bars per database with a unified color map based on top-N databases by CPU

Drill-Down

• Right-click any database bar in any card → "Drill Down to DB Query Store" opens the single-database Query Store tab directly (no connection dialog)
• The current time range selection from the overview slicer is automatically passed to the drilled-down Query Store tab via SetInitialTimeRange()

Data Pipeline

• QueryStoreOverviewService fetches states, time slices, metrics, and wait stats across all user databases in parallel using SemaphoreSlim (DOP=8)
• FetchAllWaitStatsWithErrorsAsync captures per-database errors instead of silently swallowing them, surfacing failures to the user in a dialog window
• Active database list is cached on initial load to avoid re-querying on every time slicer change

UX

• Progress bar (indeterminate) at the top of the dashboard during data loading — always reserves its 3px height to prevent layout hopping
• supportsWaitStats flag (SQL 2017+ / Azure) is checked before attempting wait stats queries, with a descriptive fallback message when unsupported
• Null-safe SQL (ISNULL, reader.IsDBNull guards) and reader.FieldCount checks to handle databases returning varying result shapes

New Files

File	Purpose
QueryStoreOverviewControl.axaml	Dashboard layout (donut + slicer + wait stats + 2×7 bar card grids)
QueryStoreOverviewControl.axaml.cs	Canvas-based charting, parallel data orchestration, drill-down events

Modified Files

File	Change
QueryStoreOverviewService.cs	New service with FetchAllStatesAsync, FetchAllMetricsAsync, FetchAllTimeSlicesAsync, FetchAllWaitStatsAsync, FetchAllWaitStatsWithErrorsAsync
QueryStoreOverviewModels.cs	New models: DatabaseQueryStoreState, DatabaseMetrics, DatabaseTimeSlice, DatabaseWaitAmountTimeSlice, DatabaseWaitCategoryTimeSlice
QuerySessionControl.axaml.cs	"QS Overview" button handler, DrillDownRequested wiring, OpenQueryStoreForDatabaseAsync now accepts optional initialStartUtc/initialEndUtc
QueryStoreGridControl.axaml.cs	Added SetInitialTimeRange() for pre-setting the slicer range on drill-down

Technical Notes

• Top-N databases (topN=4 by default) get distinct palette colors; remaining databases are aggregated as "Others" in grey
• All charting is custom Canvas-based (no third-party chart library)
• Color palette: #2EAEF1 (blue), #F2994A (orange), #27AE60 (green), #9B51E0 (purple), #EB5757 (red), #F2C94C (yellow), #56CCF2 (light blue), #BB6BD9 (violet)

Which component(s) does this affect?

• Desktop App (PlanViewer.App)
• Core Library (PlanViewer.Core)
• CLI Tool (PlanViewer.Cli)
• SSMS Extension (PlanViewer.Ssms)
• Tests
• Documentation

How was this tested?

Describe the testing you've done. Include:

• Plan files tested : no impact of plan
• Platforms tested : windows

Checklist

• I have read the contributing guide
• My code builds with zero warnings (dotnet build -c Debug)
• All tests pass (dotnet test)
• I have not introduced any hardcoded credentials or server names

[rferraton]

Arf... sorry for the conflicts! i may have done a mistake when branching... but hard to know where and what to do with them!

[erikdarlingdata]

Code Review

Overview

Adds a cross-database Query Store overview dashboard (donut + slicer + wait-stats ribbon + 14 metric bar cards) with right-click drill-down into the per-DB Query Store tab carrying the time range. ~1,524 lines, 7 files. Compiles clean against dev after the merge.

What's good

• Parallelism is well-bounded — SemaphoreSlim(maxDop=8) + ConcurrentBag for fan-out across N user databases. Cancellation propagates through to OpenAsync/ExecuteReaderAsync.
• All SQL is parameterized (SqlParameter for @start, @end, @daysBack); database names come from sys.databases, not user input; per-DB connection strings are built via SqlConnectionStringBuilder. Clean on the security side.
• FetchAllWaitStatsWithErrorsAsync is the right pattern — captures per-DB errors with when (ex is not OperationCanceledException) instead of swallowing them.
• _supportsWaitStats gating + descriptive fallback for SQL 2016/older.
• Drill-down API on QueryStoreGridControl.SetInitialTimeRange is small and explicit.

Bugs / correctness

1. Dead code — FetchDatabaseWaitStatsAsync (QueryStoreOverviewService.cs:332-373) and DatabaseWaitCategoryTimeSlice (QueryStoreOverviewModels.cs:66-73) have no callers. Only FetchDatabaseWaitAmountAsync is used. Delete both.

2. Errors swallowed silently — QueryStoreOverviewService.cs:109 and :143 use bare catch { /* skip */ }, which will eat OperationCanceledException too and defeat cancellation. Apply the same when (ex is not OperationCanceledException) pattern used in FetchAllWaitStatsWithErrorsAsync.

3. Permission errors look like "Off" — FetchAllStatesAsync:37-40 maps any per-DB exception to QueryStoreState.Off, so a missing VIEW DATABASE STATE grant looks identical to QS being disabled in the donut. At minimum log/surface; ideally add an Unknown/Error state.

4. _cts leak — LoadAsync:95-96 and OnSlicerRangeChanged:424-425 cancel the old token source but never Dispose() it. Add _cts?.Dispose() between cancel and reassign, and dispose on control detach.

5. Race in DrawWaitStatsChart — line 477 var topDbs = _dbColorMap.Keys.ToList(); depends on DrawBarCards having populated _dbColorMap first. The normal call order does this, but SizeChanged (constructor:84-88) calls DrawWaitStatsChart independently — if a resize fires between LoadAsync Phase 1 and Phase 3, every wait bar gets bucketed into Others. Defer the wait chart, or compute topDbs from _metrics directly.

6. Misnamed field — WaitRatio on DatabaseWaitAmountTimeSlice (and DatabaseWaitCategoryTimeSlice) is wait time in hours (SUM(total_query_wait_time_ms) / 3,600,000), not a ratio. Rename to WaitHours / WaitAmountHours to match the SQL and WaitRatioFormatter semantics elsewhere.

Style / conventions

• Tab vs space inconsistency — QueryStoreOverviewService.cs:375-411 (FetchDatabaseWaitAmountAsync) is tab-indented while every other method is 4-space. Same in QueryStoreOverviewModels.cs:82 (WaitRatio on DatabaseWaitAmountTimeSlice). Will look broken in any editor without tab-width=4.
• Trailing blank lines in QueryStoreOverviewModels.cs:85-87.
• _dbColorMap is declared mid-file (line 630) instead of with the other private fields (lines 31-39).
• if (db != "Others") on line 742 — extract private const string OthersLabel = "Others" and reuse.
• Repeated Color.Parse("#E4E6EB") is already available as DynamicResource ForegroundBrush in XAML; resolve once or use FindResource.

Performance (minor)

• topDbs.Contains(item.Db) inside the hour loop (:511) and per-metric loop (:699) — List<string>.Contains is O(n). Use a HashSet<string> since the list is small but checked many times per redraw.
• SizeChanged redraws everything; rapid resizing will thrash. A small debounce would help — not blocking.

Tests

No tests added for ~1,500 lines of new code (PR checklist has Tests unchecked). QueryStoreOverviewService is a static class with discrete query methods — at minimum, table-driven tests against a real DB would catch column-order regressions in the SQL.

SQL notes

• WHERE database_id > 4 correctly excludes system DBs.
• READ UNCOMMITTED on QS DMVs is fine and conventional.
• bucket_hour = DATEADD(HOUR, DATEDIFF(HOUR, 0, rsi.start_time), 0) truncates to hour-of-epoch in server local time (because DATEDIFF(HOUR, 0, ...) is naïve). The result is then SpecifyKind(..., Utc) on the C# side, which mislabels it on a non-UTC server. Either compute UTC-truncation server-side or stop forcing Utc kind when the SQL hasn't promised it. Worth verifying against a non-UTC server before merge.

Verdict

Functionally solid, security-clean, well-bounded async. Blockers are small: the dead code, the silent catches eating cancellation, and the bucket-hour timezone question. The rest is polish (indentation, naming, dispose). Recommend addressing #1–6 before merge; style/perf items can be follow-ups.

[rferraton]

1. Dead code removed — Deleted FetchDatabaseWaitStatsAsync and DatabaseWaitCategoryTimeSlice (no callers).
2. Bare catch blocks fixed — All 4 parallel fetch methods now use when (ex is not OperationCanceledException) or when (!ct.IsCancellationRequested) so cancellation propagates correctly.
3. Permission errors surfaced — Added QueryStoreState.Error enum value and ErrorMessage property to DatabaseQueryStoreState. The donut now shows a red "Error" segment, and clicking it lists databases with their error messages.
4. _cts leak fixed — Added _cts?.Dispose() before every reassignment in LoadAsync() and OnSlicerRangeChanged(object?, TimeRangeChangedEventArgs), plus a DetachedFromVisualTree handler that cancels and disposes on control teardown.
5. SizeChanged race fixed — DrawWaitStatsChart() now returns early if _dbColorMap is empty, preventing all bars from being bucketed into "Others" when a resize fires before DrawBarCards() has run.
6. Misnamed field renamed — WaitRatio → WaitAmountHours on DatabaseWaitAmountTimeSlice and all references in the service and control.

[rferraton]

@erikdarlingdata : thanks for the merge from dev :-)