dohertj2/ScadaBridge
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

perf(ui): shared KPI cache, live-cache-backed alarm summary, coalesced debug renders

Browse Source
This commit is contained in:
Joseph Doherty
2026-08-14 20:53:38 -04:00
parent ee193cd2bb
commit 8c0b36b2fa
19 changed files with 1554 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/ZB.MOM.WW.ScadaBridge.CentralUI/Components/Pages/Monitoring/Health.razor
+54 -19
View File
@@ -15,8 +15,7 @@
@implements IDisposable
@inject ICentralHealthAggregator HealthAggregator
@inject ISiteRepository SiteRepository
@inject CommunicationService CommunicationService
@inject IAuditLogQueryService AuditLogQueryService
@inject IKpiSnapshotCache KpiSnapshots
@inject IKpiHistoryQueryService KpiHistory
@inject Microsoft.Extensions.Options.IOptions<HealthMonitoringOptions> HealthOptions
@@ -640,7 +639,9 @@
// Non-fatal — fall back to showing siteId only
}
await RefreshNow();
// First load rides whatever the shared KPI cache already has fresh — a second
// operator opening the dashboard inside the window costs no extra query.
await RefreshAllAsync(forceRefresh: false);
// Site Health Trends (M6) load on their own path — never from the
// timer tick below — so a trend-query fault can't disturb the live tile
@@ -655,7 +656,7 @@
{
try
{
await RefreshNow();
await RefreshAllAsync(forceRefresh: false);
StateHasChanged();
}
finally
@@ -769,20 +770,38 @@
? "Central Cluster"
: $"{GetSiteName(siteKey)} ({siteKey})";
private async Task RefreshNow()
// Per-tile budget for one refresh tick. The underlying KPI Asks carry the
// 30s CommunicationOptions.QueryTimeout, so without this a single hung
// singleton stalled the WHOLE tick (and, with a 10s timer, stacked ticks)
// rather than degrading its own tile. Cancelling here abandons only this
// circuit's wait — the shared KPI flight continues and its result lands in
// the process-level cache for the next tick.
private static readonly TimeSpan TileTimeout = TimeSpan.FromSeconds(5);
// The operator's explicit "Refresh Now" bypasses the shared cache's freshness
// window; the timer tick and the first load ride whatever is already fresh
// (which is the whole point — N dashboards share one query per window).
private Task RefreshNow() => RefreshAllAsync(forceRefresh: true);
// Every tile loads concurrently, each with its own short budget, so one slow
// KPI degrades one tile instead of the tick. Every loader swallows its own
// faults, so Task.WhenAll here can never observe an exception.
private async Task RefreshAllAsync(bool forceRefresh)
{
_siteStates = HealthAggregator.GetAllSiteStates();
await LoadOutboxKpis();
await Task.WhenAll(LoadSiteCallKpis(), LoadSiteCallNodeKpis());
await LoadAuditKpis();
await Task.WhenAll(
LoadOutboxKpis(forceRefresh),
LoadSiteCallKpis(forceRefresh),
LoadSiteCallNodeKpis(forceRefresh),
LoadAuditKpis(forceRefresh));
}
private async Task LoadOutboxKpis()
private async Task LoadOutboxKpis(bool forceRefresh)
{
try
{
var response = await CommunicationService.GetNotificationKpisAsync(
new NotificationKpiRequest(Guid.NewGuid().ToString("N")));
using var timeout = new CancellationTokenSource(TileTimeout);
var response = await KpiSnapshots.GetNotificationKpisAsync(forceRefresh, timeout.Token);
if (response.Success)
{
_outboxKpi = response;
@@ -795,6 +814,11 @@
_outboxKpiError = response.ErrorMessage ?? "KPI query failed.";
}
}
catch (OperationCanceledException)
{
_outboxKpiAvailable = false;
_outboxKpiError = "KPI query timed out.";
}
catch (Exception ex)
{
_outboxKpiAvailable = false;
@@ -807,12 +831,12 @@
// killing the dashboard. Mirrors LoadOutboxKpis's error handling shape — a
// response with Success == false (repository fault) and an Ask that threw
// (transport fault) both collapse to "unavailable".
private async Task LoadSiteCallKpis()
private async Task LoadSiteCallKpis(bool forceRefresh)
{
try
{
var response = await CommunicationService.GetSiteCallKpisAsync(
new SiteCallKpiRequest(Guid.NewGuid().ToString("N")));
using var timeout = new CancellationTokenSource(TileTimeout);
var response = await KpiSnapshots.GetSiteCallKpisAsync(forceRefresh, timeout.Token);
if (response.Success)
{
_siteCallKpi = response;
@@ -825,6 +849,11 @@
_siteCallKpiError = response.ErrorMessage ?? "KPI query failed.";
}
}
catch (OperationCanceledException)
{
_siteCallKpiAvailable = false;
_siteCallKpiError = "KPI query timed out.";
}
catch (Exception ex)
{
_siteCallKpiAvailable = false;
@@ -834,12 +863,12 @@
// Per-node site-call KPI loader (M5.2). Best-effort; a fault silently
// suppresses the per-node sub-table rather than degrading the dashboard.
private async Task LoadSiteCallNodeKpis()
private async Task LoadSiteCallNodeKpis(bool forceRefresh)
{
try
{
var response = await CommunicationService.GetPerNodeSiteCallKpisAsync(
new PerNodeSiteCallKpiRequest(Guid.NewGuid().ToString("N")));
using var timeout = new CancellationTokenSource(TileTimeout);
var response = await KpiSnapshots.GetPerNodeSiteCallKpisAsync(forceRefresh, timeout.Token);
if (response.Success)
{
_siteCallNodeKpis = response.Nodes;
@@ -864,14 +893,20 @@
// Audit KPI loader: wraps the service call so a transient DB outage degrades
// the three tiles to em dashes with an inline error rather than killing the
// dashboard. Mirrors LoadOutboxKpis's error handling shape.
private async Task LoadAuditKpis()
private async Task LoadAuditKpis(bool forceRefresh)
{
try
{
_auditKpi = await AuditLogQueryService.GetKpiSnapshotAsync();
using var timeout = new CancellationTokenSource(TileTimeout);
_auditKpi = await KpiSnapshots.GetAuditKpiSnapshotAsync(forceRefresh, timeout.Token);
_auditKpiAvailable = true;
_auditKpiError = null;
}
catch (OperationCanceledException)
{
_auditKpiAvailable = false;
_auditKpiError = "KPI query timed out.";
}
catch (Exception ex)
{
_auditKpiAvailable = false;