AG tab refreshes cards in place and virtualizes the list (#4238)

[erikdarlingdata]

Closes #4238.

Why

The Darling viewer's Availability Groups tab tore down and rebuilt every card and every per-database DataGrid on each 30 s refresh, on the UI thread. Nothing was virtualized. Render assigned a brand new List<AgTopologyCard> to AgCards.ItemsSource every time, made of brand new card instances, whether or not anything had changed. The card list was a plain ItemsControl inside a ScrollViewer, so every card realized whether or not it was on screen. #4238 measured a production topology of 42 AGs, 84 replicas and 398 database rows. There, each refresh realized 10,353 elements and stalled the UI thread for 2,548 ms.

This PR's pin tests were written first and run against the code before the fix, to confirm the cause. 4 of the 5 new tests failed. One of them checks that two refreshes with the same rows keep the same card instances. So the old code replaced every card object on every refresh, whether or not its data changed.

What changes

PerformanceMonitor.Common/AgTopology.cs is shared by the Darling viewer and Lite. A search of the whole repo found that the web dashboard does not use these types.

• AgTopologyCard, AgTopologyReplica, and AgTopologyDatabase now implement INotifyPropertyChanged, through a small shared AgTopologyObservable base. It stays plain BCL, no WPF dependency. Each type now has an UpdateFrom method. It copies every field from a newly read instance with the same identity, then raises one change notification for all properties.
• AgTopology.Reconcile<TItem, TKey> is new. It syncs an ObservableCollection<TItem> to a freshly read list, in the fresh list's order, and reuses existing instances (through UpdateFrom) for every identity that survives. It adds new instances and removes stale ones. It is reused at three levels: a tab's top-level card list, and, inside AgTopologyCard.UpdateFrom, a card's own nested Replicas and Databases collections. That gives one reconciliation rule instead of three near-identical ones.
• AgTopology.CardKey is new: a card's identity is its reporting server plus its normalized AG name, the same pair BuildCards already groups by.
• AgTopology.ComputeDigest is new: a fingerprint of every field a card renders, including both grains' collection times. A refresh whose digest matches the last render skips the update.
• BuildCards behavior is unchanged. Only its internal construction of a card changed, because Replicas/Databases are now get-only ObservableCollections populated with Add, not an object-initializer assignment.

Darling viewer, AvailabilityGroupsTab.xaml and .xaml.cs:

• AgCards now virtualizes. Its ScrollViewer moved inside the ItemsControl's own ControlTemplate, wrapping CanContentScroll="True" around an ItemsPresenter, with a VirtualizingStackPanel (IsVirtualizing="True", VirtualizationMode="Recycling", ScrollUnit="Pixel"). A ScrollViewer wrapped around a plain ItemsControl from outside gives the panel unbounded height and never actually virtualizes. The fix had to move the ScrollViewer inside the template, not just add virtualization properties to the old structure. The named CardScroller element is gone. Its visibility toggle now targets AgCards directly.
• The tab owns one long-lived ObservableCollection<AgTopologyCard>, bound to AgCards.ItemsSource exactly once, in the constructor. LoadAsync computes a digest of the freshly read rows and returns early when it matches the last render. Render reconciles that collection in place through AgTopology.Reconcile, instead of replacing ItemsSource. Render is now internal, mirroring the existing BuildSummary precedent, so tests can drive it directly without a live store.

Lite, Lite/Controls/AvailabilityGroupsTab.xaml and .xaml.cs: Lite shows Availability Groups the same way, with the same defect: AgCards.ItemsSource = cards wholesale, a plain ItemsControl in a ScrollViewer. It gets the same fix, the same shared AgTopology.Reconcile / ComputeDigest / UpdateFrom mechanism, and the same XAML virtualization change.

Tests: Darling/Darling.Tests/AvailabilityGroupsTabRefreshTests.cs (new) and Lite.Tests/AvailabilityGroupsTabRefreshTests.cs (new). Details below.

Test plan

• The pin tests ran once against the code before the fix, to prove the defect. Darling.Tests -class "*AvailabilityGroupsTabRefreshTests*" : 4 of 5 failed. The same-rows-same-instance, changed-value-updates-in-place, add/remove-one-card, and XAML virtualizing-panel pins all failed against the old ItemsSource = code. Only the timing pin passed. The note below explains why.
• dotnet build Darling/Darling.Tests/Darling.Tests.csproj: 0 Warning(s), 0 Error(s).
• After the fix, the changed classes ran with the viewer's other STA test class, because they share the dispatcher: Darling.Tests -class "*AvailabilityGroupsTabRefreshTests*" -class "*AgTopologyCardsTests*" -class "*RawWindowFloorViewerPortTests*". 34 total, 0 failed.
• dotnet build Lite.Tests/Lite.Tests.csproj: 0 Warning(s), 0 Error(s).
• Lite.Tests -class "*AvailabilityGroupsTabRefreshTests*" -class "*AgTopologyTests*" -class "*AgAvailabilityGroupNewestSnapshotEqualityTests*". 30 total, 0 failed.
• This PR changes Lite, so the full Lite suite ran once: Lite.Tests.exe, no filter. 5,392 total, 0 failed, 0 errors, 314.2 s.
• The timing pin uses the WPF Availability Groups tab rebuilds every card on each 30 s refresh: 10 k realized elements (42 AGs, no virtualization) stall the UI thread ~2.5 s per tick #4238 production shape: 42 AGs, 84 replicas and 398 database rows. A refresh with unchanged rows on an STA thread took about 4.9 ms, against a 100 ms budget.
• No full Darling suite run. This PR changes only viewer XAML, viewer code-behind, and a PerformanceMonitor.Common model file that no live or store test reads. Only the pure banding/projection tests read it, and those run in the classes above. CI runs the rest.
• GitHub Actions CI on f06b9b2d: build, Darling PostgreSQL tests and Lite tests all pass.

A note on the timing number: the pin runs headless, with no Measure/Arrange/UpdateLayout. It never makes WPF realize the visuals, so it passed even on the code before the fix. Without a layout pass, neither version pays the cost of creating containers, which caused the 2,548 ms that #4238 measured in the field. The 100 ms budget is still a real, permanent regression guard on the reconcile/digest code path. The object-identity pins are what actually prove the rebuild-on-every-refresh defect is gone: they fail before this fix and pass after it.

CHANGELOG entry

SECTION: Fixed
ENTRY:

• Availability Groups tab no longer freezes the UI every 30 seconds ([AG tab refreshes cards in place and virtualizes the list (#4238) #4308]) - The Darling viewer's and Lite's Availability Groups tabs rebuilt every card and every per-database grid from scratch on every refresh. In the Darling viewer, that blocked the UI thread for over 2.5 seconds on a 42-AG fleet, even when nothing had changed. Refreshes now update existing cards in place, skip entirely when the data has not changed, and only render the cards that are actually on screen.
  REF:
  [AG tab refreshes cards in place and virtualizes the list (#4238) #4308]: AG tab refreshes cards in place and virtualizes the list (#4238) #4308

🤖 Generated with Claude Code

https://claude.ai/code/session_01FVjn4PBJN71NQXdFo6ZxNQ