feat: wire all health metrics and add instance counts to dashboard
Wired ISiteHealthCollector calls for script errors (ScriptExecutionActor), alarm eval errors (AlarmActor), dead letters (DeadLetterMonitorActor), and S&F buffer depth placeholder. Added instance count tracking (deployed/ enabled/disabled) to SiteHealthReport via DeploymentManagerActor. Updated Health Dashboard UI to show instance counts per site. All metrics flow through the existing health report pipeline via ClusterClient.
This commit is contained in:
@@ -5,6 +5,7 @@ using ScadaLink.Commons.Messages.DebugView;
|
||||
using ScadaLink.Commons.Messages.Deployment;
|
||||
using ScadaLink.Commons.Messages.Lifecycle;
|
||||
using ScadaLink.Commons.Types.Enums;
|
||||
using ScadaLink.HealthMonitoring;
|
||||
using ScadaLink.SiteRuntime.Persistence;
|
||||
using ScadaLink.SiteRuntime.Scripts;
|
||||
using ScadaLink.SiteRuntime.Streaming;
|
||||
@@ -30,7 +31,9 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
private readonly SiteRuntimeOptions _options;
|
||||
private readonly ILogger<DeploymentManagerActor> _logger;
|
||||
private readonly IActorRef? _dclManager;
|
||||
private readonly ISiteHealthCollector? _healthCollector;
|
||||
private readonly Dictionary<string, IActorRef> _instanceActors = new();
|
||||
private int _totalDeployedCount;
|
||||
|
||||
public ITimerScheduler Timers { get; set; } = null!;
|
||||
|
||||
@@ -41,7 +44,8 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
SiteStreamManager? streamManager,
|
||||
SiteRuntimeOptions options,
|
||||
ILogger<DeploymentManagerActor> logger,
|
||||
IActorRef? dclManager = null)
|
||||
IActorRef? dclManager = null,
|
||||
ISiteHealthCollector? healthCollector = null)
|
||||
{
|
||||
_storage = storage;
|
||||
_compilationService = compilationService;
|
||||
@@ -49,6 +53,7 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
_streamManager = streamManager;
|
||||
_options = options;
|
||||
_dclManager = dclManager;
|
||||
_healthCollector = healthCollector;
|
||||
_logger = logger;
|
||||
|
||||
// Lifecycle commands
|
||||
@@ -123,9 +128,11 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
}
|
||||
|
||||
var enabledConfigs = msg.Configs.Where(c => c.IsEnabled).ToList();
|
||||
_totalDeployedCount = msg.Configs.Count;
|
||||
_logger.LogInformation(
|
||||
"Loaded {Total} deployed configs ({Enabled} enabled) from SQLite",
|
||||
msg.Configs.Count, enabledConfigs.Count);
|
||||
UpdateInstanceCounts();
|
||||
|
||||
if (enabledConfigs.Count == 0)
|
||||
return;
|
||||
@@ -200,6 +207,8 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
|
||||
// Create the Instance Actor immediately (no existing actor to replace)
|
||||
CreateInstanceActor(instanceName, command.FlattenedConfigurationJson);
|
||||
_totalDeployedCount++;
|
||||
UpdateInstanceCounts();
|
||||
|
||||
// Persist to SQLite and clear static overrides asynchronously
|
||||
var sender = Sender;
|
||||
@@ -257,6 +266,8 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
_instanceActors.Remove(instanceName);
|
||||
}
|
||||
|
||||
UpdateInstanceCounts();
|
||||
|
||||
var sender = Sender;
|
||||
_storage.SetInstanceEnabledAsync(instanceName, false).ContinueWith(t =>
|
||||
{
|
||||
@@ -313,6 +324,7 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
{
|
||||
CreateInstanceActor(instanceName, result.Config.ConfigJson);
|
||||
}
|
||||
UpdateInstanceCounts();
|
||||
|
||||
result.OriginalSender.Tell(new InstanceLifecycleResponse(
|
||||
result.Command.CommandId, instanceName, true, null, DateTimeOffset.UtcNow));
|
||||
@@ -333,6 +345,8 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
Context.Stop(actor);
|
||||
_instanceActors.Remove(instanceName);
|
||||
}
|
||||
_totalDeployedCount = Math.Max(0, _totalDeployedCount - 1);
|
||||
UpdateInstanceCounts();
|
||||
|
||||
var sender = Sender;
|
||||
_storage.RemoveDeployedConfigAsync(instanceName).ContinueWith(t =>
|
||||
@@ -536,7 +550,8 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
_streamManager,
|
||||
_options,
|
||||
loggerFactory.CreateLogger<InstanceActor>(),
|
||||
_dclManager));
|
||||
_dclManager,
|
||||
_healthCollector));
|
||||
|
||||
var actorRef = Context.ActorOf(props, instanceName);
|
||||
_instanceActors[instanceName] = actorRef;
|
||||
@@ -549,6 +564,18 @@ public class DeploymentManagerActor : ReceiveActor, IWithTimers
|
||||
/// </summary>
|
||||
internal int InstanceActorCount => _instanceActors.Count;
|
||||
|
||||
/// <summary>
|
||||
/// Updates the health collector with current instance counts.
|
||||
/// Total deployed = _totalDeployedCount, enabled = running actors, disabled = difference.
|
||||
/// </summary>
|
||||
private void UpdateInstanceCounts()
|
||||
{
|
||||
_healthCollector?.SetInstanceCounts(
|
||||
deployed: _totalDeployedCount,
|
||||
enabled: _instanceActors.Count,
|
||||
disabled: _totalDeployedCount - _instanceActors.Count);
|
||||
}
|
||||
|
||||
// ── Internal messages ──
|
||||
|
||||
internal record StartupConfigsLoaded(List<DeployedInstance> Configs, string? Error);
|
||||
|
||||
Reference in New Issue
Block a user