Prognosis 3.0.0-alpha

Prognosis

Unfiltered AI slop. Use at your own risk

A dependency-aware service health modeling library for .NET. Models the health of multiple services as a directed graph where each service's effective status is computed from its own intrinsic health and the weighted health of its dependencies.

graph TD
    Store["🛒 Online Store"] -->|"Required"| Checkout["Checkout"]
    Store -->|"Important"| Search["Product Search"]
    Store -->|"Optional"| Reviews["Reviews"]
    Checkout -->|"Required"| Payment["Payment Gateway"]
    Checkout -->|"Required"| Inventory["Inventory"]
    Payment -->|"Important"| Fraud["Fraud Detection"]
    Search -->|"Required"| Index["Search Index"]

    style Store fill:#22c55e,color:#fff
    style Checkout fill:#22c55e,color:#fff
    style Payment fill:#22c55e,color:#fff
    style Inventory fill:#22c55e,color:#fff
    style Fraud fill:#22c55e,color:#fff
    style Search fill:#22c55e,color:#fff
    style Index fill:#22c55e,color:#fff
    style Reviews fill:#22c55e,color:#fff

How it works: each service reports its own health and declares dependencies with an importance level. Prognosis walks the graph and computes the effective status — a Required dependency failing makes the parent unhealthy, an Important one degrades it, and an Optional one is ignored. If Fraud Detection goes down, Payment Gateway becomes degraded, which degrades Checkout, which degrades the whole store. If Payment Gateway itself goes down, Checkout becomes unhealthy — and since it's Required, the store is unhealthy too. If Reviews go down? Nothing happens.

Packages

Key concepts

Health statuses

Ordered worst-is-highest so comparisons naturally surface the most severe status.

Dependency importance

These rules describe the default strategy (HealthAggregator.Aggregate). Propagation behavior can be changed per-service by supplying a custom aggregation strategy.

Usage patterns

1. Implement IServiceHealth on a class you own

Expose a ServiceHealth property — no forwarding boilerplate:

class CacheService : IServiceHealth
{
    public ServiceHealth Health { get; }

    public CacheService()
    {
        Health = new DelegatingServiceHealth("Cache",
            () => IsConnected
                ? HealthStatus.Healthy
                : new HealthEvaluation(HealthStatus.Unhealthy, "Redis timeout"));
    }

    public bool IsConnected { get; set; } = true;
}

For services with fine-grained health attributes, use a CompositeServiceHealth backed by sub-nodes:

class DatabaseService : IServiceHealth
{
    public ServiceHealth Health { get; }

    public bool IsConnected { get; set; } = true;
    public double AverageLatencyMs { get; set; } = 50;
    public double PoolUtilization { get; set; } = 0.3;

    public DatabaseService()
    {
        var connection = new DelegatingServiceHealth("Database.Connection",
            () => IsConnected
                ? HealthStatus.Healthy
                : new HealthEvaluation(HealthStatus.Unhealthy, "Connection lost"));

        var latency = new DelegatingServiceHealth("Database.Latency",
            () => AverageLatencyMs switch
            {
                > 500 => new HealthEvaluation(HealthStatus.Degraded,
                    $"Avg latency {AverageLatencyMs:F0}ms exceeds 500ms threshold"),
                _ => HealthStatus.Healthy,
            });

        var connectionPool = new DelegatingServiceHealth("Database.ConnectionPool",
            () => PoolUtilization switch
            {
                >= 1.0 => new HealthEvaluation(HealthStatus.Unhealthy,
                    "Connection pool exhausted"),
                >= 0.9 => new HealthEvaluation(HealthStatus.Degraded,
                    $"Connection pool at {PoolUtilization:P0} utilization"),
                _ => HealthStatus.Healthy,
            });

        Health = new CompositeServiceHealth("Database")
            .DependsOn(connection, ServiceImportance.Required)
            .DependsOn(latency, ServiceImportance.Important)
            .DependsOn(connectionPool, ServiceImportance.Required);
    }
}

The sub-nodes show up automatically in EvaluateAll, CreateReport, and the JSON output. Reason strings chain from the leaf all the way up:

Database.Latency: Degraded — Avg latency 600ms exceeds 500ms threshold
Database: Degraded (3 dependencies) — Database.Latency: Avg latency 600ms exceeds 500ms threshold
AuthService: Degraded (2 dependencies) — Database: Database.Latency: ...

2. Wrap a service you can't modify

Use DelegatingServiceHealth with a health-check delegate:

var emailHealth = new DelegatingServiceHealth("EmailProvider",
    () => client.IsConnected
        ? HealthStatus.Healthy
        : new HealthEvaluation(HealthStatus.Unhealthy, "SMTP connection refused"));

3. Compose the graph

Wire services together with DependsOn or ServiceDependency lists:

var authService = new DelegatingServiceHealth("AuthService")
    .DependsOn(database.Health, ServiceImportance.Required)
    .DependsOn(cache.Health, ServiceImportance.Important);

// List style
var app = new CompositeServiceHealth("Application",
[
    new ServiceDependency(authService, ServiceImportance.Required),
    new ServiceDependency(notifications, ServiceImportance.Important),
]);

// Fluent style
var app = new CompositeServiceHealth("Application")
    .DependsOn(authService, ServiceImportance.Required)
    .DependsOn(notifications, ServiceImportance.Important);

Aggregation strategies

By default, health propagation follows the rules in Dependency importance via HealthAggregator.Aggregate. You can swap the strategy per-service to change how dependency statuses combine.

Built-in strategies

Using a built-in strategy

// Without redundancy (default) — if either service goes down, the parent is unhealthy.
var app = new CompositeServiceHealth("Application")
    .DependsOn(primaryDb, ServiceImportance.Required)
    .DependsOn(replicaDb, ServiceImportance.Required);

// With redundancy — if one goes down but the other is healthy, the parent is degraded.
var app = new CompositeServiceHealth("Application",
    aggregator: HealthAggregator.AggregateWithRedundancy)
    .DependsOn(primaryDb, ServiceImportance.Required)
    .DependsOn(replicaDb, ServiceImportance.Required);

Custom strategies

The AggregationStrategy delegate takes intrinsic health + dependencies and returns an evaluation. Pass any matching function:

var app = new CompositeServiceHealth("Application",
    aggregator: (intrinsic, deps) =>
    {
        // Your custom logic here
        return new HealthEvaluation(HealthStatus.Healthy);
    });

Strategies can also be applied to DelegatingServiceHealth and ServiceHealthTracker:

// DelegatingServiceHealth with a strategy
var svc = new DelegatingServiceHealth("Service", healthCheck,
    aggregator: HealthAggregator.AggregateWithRedundancy);

// ServiceHealthTracker (when embedding in your own class)
var tracker = new ServiceHealthTracker(intrinsicCheck,
    aggregator: HealthAggregator.AggregateWithRedundancy);

Graph operations

// Evaluate a single service (walks its dependencies)
HealthEvaluation eval = app.Evaluate();

// Snapshot the entire graph (depth-first post-order)
IReadOnlyList<ServiceSnapshot> snapshots = HealthAggregator.EvaluateAll(app);

// Package as a serialization-ready report with timestamp
HealthReport report = HealthAggregator.CreateReport(app);

// Detect circular dependencies
IReadOnlyList<IReadOnlyList<string>> cycles = HealthAggregator.DetectCycles(app);

// Diff two reports to find individual service changes
IReadOnlyList<ServiceStatusChange> changes = HealthAggregator.Diff(before, after);

Observable health monitoring

Every ServiceHealth node supports push-based notifications. Subscribe to individual services or monitor the full graph:

// Individual service notifications
database.Health.StatusChanged.Subscribe(observer);

// Graph-level polling with HealthMonitor
await using var monitor = new HealthMonitor([app], TimeSpan.FromSeconds(5));
monitor.ReportChanged.Subscribe(reportObserver);

// Manual poll (useful for testing or getting initial state)
monitor.Poll();

IObservable<T> is a BCL type — no System.Reactive dependency required. Add System.Reactive only when you want operators like DistinctUntilChanged() or Throttle().

Dependency injection

The Prognosis.DependencyInjection package provides a fluent builder for configuring the health graph within a hosted application:

builder.Services.AddPrognosis(health =>
{
    // Discover all IServiceHealth implementations and wire [DependsOn<T>] attributes.
    health.ScanForServices(typeof(Program).Assembly);

    // Wrap a third-party service with a health delegate.
    // Name defaults to typeof(T).Name when omitted.
    health.AddDelegate<ThirdPartyEmailClient>("EmailProvider",
        client => client.IsConnected
            ? HealthStatus.Healthy
            : new HealthEvaluation(HealthStatus.Unhealthy, "SMTP refused"));

    // Define composite aggregation nodes.
    // Use nameof for concrete types, constants for virtual nodes.
    health.AddComposite(ServiceNames.Application, app =>
    {
        app.DependsOn<AuthService>(ServiceImportance.Required);
        app.DependsOn(ServiceNames.NotificationSystem, ServiceImportance.Important);
    });

    // Composites with custom aggregation strategies.
    health.AddComposite(ServiceNames.DatabaseCluster, db =>
    {
        db.DependsOn(nameof(PrimaryDb), ServiceImportance.Required);
        db.DependsOn(nameof(ReplicaDb), ServiceImportance.Required);
    }, aggregator: HealthAggregator.AggregateWithRedundancy);

    health.AddRoots(ServiceNames.Application);
    health.UseMonitor(TimeSpan.FromSeconds(30));
});

// Central constants for virtual/composite service names.
static class ServiceNames
{
    public const string Application = nameof(Application);
    public const string NotificationSystem = nameof(NotificationSystem);
    public const string DatabaseCluster = nameof(DatabaseCluster);
}

Declare dependency edges on classes you own with attributes:

[DependsOn<DatabaseService>(ServiceImportance.Required)]
[DependsOn<CacheService>(ServiceImportance.Important)]
class AuthService : IServiceHealth
{
    public ServiceHealth Health { get; } = new DelegatingServiceHealth("AuthService");
}

Inject HealthGraph to access the materialized graph at runtime:

var graph = serviceProvider.GetRequiredService<HealthGraph>();
var report = graph.CreateReport();

// Type-safe lookup — uses typeof(AuthService).Name as key.
if (graph.TryGetService<AuthService>(out var auth))
    Console.WriteLine($"AuthService has {auth.Dependencies.Count} deps");

// String-based lookup still available.
var dbService = graph["Database"];

Reactive extensions

The Prognosis.Reactive package provides Rx-based alternatives to polling:

var roots = new ServiceHealth[] { app };

// Timer-driven polling — emits HealthReport on change.
roots.PollHealthReport(TimeSpan.FromSeconds(30))
    .Subscribe(report => Console.WriteLine(report.OverallStatus));

// Push-triggered — reacts to leaf StatusChanged events, throttled.
roots.ObserveHealthReport(TimeSpan.FromMilliseconds(500))
    .Subscribe(report => Console.WriteLine(report.OverallStatus));

// Diff-based change stream — composable with any report source.
roots.PollHealthReport(TimeSpan.FromSeconds(30))
    .SelectServiceChanges()
    .Subscribe(change =>
        Console.WriteLine($"{change.Name}: {change.Previous} → {change.Current}"));

Sharing streams across subscribers

The Rx helpers produce cold observables — each subscription runs its own pipeline. To share a single evaluation across multiple subscribers:

// Auto-stop when last subscriber unsubscribes.
var shared = roots.CreateSharedReportStream(TimeSpan.FromSeconds(30));

// Replay latest report to late subscribers.
var shared = roots.CreateSharedReportStream(TimeSpan.FromSeconds(30),
    ShareStrategy.ReplayLatest);

// Push-triggered variant.
var shared = roots.CreateSharedObserveStream(TimeSpan.FromMilliseconds(500));

Or use standard Rx multicast operators directly: Publish().RefCount() or Replay(1).RefCount().

Serialization

Both enums use [JsonStringEnumConverter] so they serialize as "Healthy" / "Degraded" / etc. The HealthReport and ServiceSnapshot records are designed as wire-friendly DTOs:

{
  "Timestamp": "2026-02-13T18:30:00+00:00",
  "OverallStatus": "Healthy",
  "Services": [
    { "Name": "Database.Connection", "Status": "Healthy", "DependencyCount": 0 },
    { "Name": "Database.Latency", "Status": "Healthy", "DependencyCount": 0 },
    { "Name": "Database.ConnectionPool", "Status": "Healthy", "DependencyCount": 0 },
    { "Name": "Database", "Status": "Healthy", "DependencyCount": 3 },
    { "Name": "AuthService", "Status": "Healthy", "DependencyCount": 2 }
  ]
}

Project structure

Core (Prognosis)

Reactive extensions (Prognosis.Reactive)

Dependency injection (Prognosis.DependencyInjection)

Requirements

• .NET Standard 2.0 or .NET Standard 2.1 compatible runtime (.NET Framework 4.6.1+, .NET Core 2.0+, .NET 5+)
• System.Text.Json (bundled as a dependency)
• Microsoft.Bcl.AsyncInterfaces (netstandard2.0 only, bundled as a dependency)