Nafas.Observability.Server 0.1.0

Nafas.Observability

An embedded, self-hosted observability dashboard for ASP.NET Core: logs, metrics, traces, and alerting, captured automatically from what your app already produces and served from inside your own process.

builder.Services.AddNafasServer();   // storage + real ingestion + alert evaluation
app.UseNafasDashboard("/nafas");     // mounts the dashboard UI at /nafas

No collector, no message broker, no external database, no Redis, no separate process to run or deploy. One NuGet package, two lines in Program.cs.

Table of contents

• Why Nafas
• Quick start
• What gets captured, and how
• Configuration
• Security
• Desktop and other non-ASP.NET Core apps
• Alerting
• Multi-instance deployments
• Requirements
• How it's built internally
• Repository layout
• Local development
• Status
• License

Why Nafas

Standing up observability for a small or medium ASP.NET Core service usually means an OTel Collector, a time-series database, and a separate dashboard to run and secure — real infrastructure for a problem that's often just "let me see recent errors and know when something breaks." Nafas is the alternative for that case: it lives inside your application's own process, stores data in a local SQLite file (or your existing SQL Server) by default, and needs nothing external to run.

It is not a replacement for a full OTel/Prometheus/Grafana stack at scale — it does not federate across services, it does not do long-term high-cardinality storage, and it does not export to other backends. It is for the case where a single self-hosted app (or a handful of them sharing a database) needs real logs, metrics, traces, and threshold-based alerting without introducing new infrastructure to operate.

Quick start

dotnet add package Nafas.Observability

var builder = WebApplication.CreateBuilder(args);

builder.Services.AddNafasServer(options =>
{
    // Every option below is optional -- these are the actual defaults.
    options.Provider = DatabaseProvider.Sqlite;
    options.RetentionDays = 30;
    // options.AlertWebhookUrl = "https://example.com/webhooks/nafas";
});

var app = builder.Build();

app.UseNafasDashboard("/nafas"); // -> https://your-app/nafas

app.Run();

That's it — no instrumentation code to add anywhere else. Every ILogger call, every Activity/span (including ASP.NET Core's own per-request span), and every Meter measurement (including ASP.NET Core 8+'s built-in request-duration metric) your app already produces is captured from this point on and shown on a real-time dashboard mounted directly inside your own app, in English or Persian, light or dark.

What gets captured, and how

Nafas hooks into the same extension points ASP.NET Core and Microsoft.Extensions.Logging/System.Diagnostics already expose — it does not require replacing ILogger<T>, wrapping HttpClient, or adding any [Trace]-style attributes to your code:

All of it is written through a bounded, non-blocking queue (System.Threading.Channels) so a burst of telemetry never applies backpressure to your request pipeline — under sustained overload, Nafas drops its own data rather than slow your app down. A dedicated background service drains and batches that queue into storage, and the dashboard's live views (the log stream, in particular) are pushed to the browser over Server-Sent Events at write time — they are not a polling query against the database.

Configuration

All of these are set inside AddNafasServer(options => { ... }).

UseNafasDashboard(path) takes the mount path for the dashboard UI itself (default "/nafas") — pick any path your app isn't already using.

Security

The dashboard shows logs, traces, and metrics — which routinely carry sensitive data (request bodies, stack traces, connection strings inside exception messages). By default, UseNafasDashboard only serves requests that come from the local machine — the same default posture as Hangfire's own dashboard. Every other request gets a 403.

builder.Services.AddNafasServer(options =>
{
    // Runs after your own auth middleware, if you have one registered
    // earlier in the pipeline -- httpContext.User is already populated.
    options.Authorize = httpContext => httpContext.User.IsInRole("Admin");

    // Or, deliberately, open to everyone (not recommended without your
    // own auth in front of it):
    // options.Authorize = _ => true;
});

If you don't set Authorize and the dashboard doesn't load from where you expect (a phone on the same network, a teammate's machine, behind a reverse proxy), this is why — set it explicitly once you know who should be allowed in.

Desktop and other non-ASP.NET Core apps

UseNafasDashboard needs an IApplicationBuilder to mount onto — a plain WinForms, WPF, or console/worker app has no ASP.NET Core pipeline of its own to provide one. Nafas.Observability.Server is a separate, optional package for exactly that case: it runs a second, independent HTTP listener (on its own port) that serves the same dashboard against the same running app's real data — no Kestrel, no OWIN, just a minimal HTTP/1.1 server directly on TcpListener (see Nafas.Observability.Server/Nafas.Observability.Server.csproj's own comment for why: the last version of Kestrel ever published as a plain NuGet package is 2.3.13 from 2019, with no HTTP/2 and no security patches since, and System.Net.HttpListener — the BCL's own listener — turned out not to be usable for the LAN case either: on Windows it's backed by HTTP.sys, which refuses to bind any prefix other than "localhost" unless the process runs elevated or a URL ACL was reserved beforehand, verified empirically). Being plain netstandard2.0 like the core package itself, this also works from classic .NET Framework 4.6.1+ apps, not just modern .NET.

dotnet add package Nafas.Observability.Server

services.AddNafasServer();               // same as always -- storage + ingestion
services.AddNafasHttpServer(o =>
{
    o.Enabled = true;                    // opt-in; does nothing while false
    o.Port = 5099;
    o.AccessMode = NafasHttpServerAccessMode.LocalhostOnly; // or .Lan
});

then open http://localhost:5099/nafas. NafasHttpServerOptions.Enabled is only read once at startup — to start/stop/reconfigure it later (e.g. a settings screen's "enable dashboard" checkbox), resolve NafasHttpServer from DI and call its StartAsync/StopAsync directly; see Nafas.Observability.Server.Sample/ for a worked WPF example.

AccessMode only controls which network interface is bound — LocalhostOnly (the default) binds loopback only, Lan binds every interface so other devices on the network can reach the port via this machine's own IP, with no administrator rights required either way (unlike System.Net.HttpListener, verified empirically on both). It does not by itself relax NafasServerOptions.Authorize (still local-requests-only by default, see Security): set Authorize explicitly too if Lan should actually let those requests through, otherwise they still get a 403.

One request per connection, not HTTP/1.1 keep-alive — this listener is a local/LAN admin tool, not a public high-throughput API, so giving up connection reuse (an extra TCP handshake per dashboard API call, imperceptible on localhost/LAN) removes an entire dimension of correctness surface instead. See Nafas.Observability.Server/Internal/NafasRawHttpConnection.cs's own comment.

Alerting

Threshold and absence rules are defined from the dashboard itself (Alerts page) — no code required to create one. A background evaluator polls every 30 seconds and fires only on state transitions (an incident opens when a rule first goes out of bounds, and resolves when it recovers) so a rule that stays broken for an hour doesn't re-notify every cycle.

Nafas does not send email, SMS, or Slack messages itself. When AlertWebhookUrl is set, every open or resolve event is POSTed there as JSON, and your own system takes it from there:

{
  "ruleId": 12,
  "ruleName": "High error rate",
  "ruleType": "threshold",
  "status": "open",
  "detail": "error_rate is 8.4, threshold is 5",
  "triggeredAtUtc": "2026-01-15T09:32:00Z",
  "resolvedAtUtc": null
}

A failed webhook delivery is logged and retried on the next transition — it never fails the request pipeline, and the incident still exists in the dashboard's own incident list regardless of whether delivery succeeded.

Multi-instance deployments

Nafas has no concept of a central collector — each instance that calls AddNafasServer() ingests and stores its own process's telemetry. Two common setups:

• Single instance — the default. Nothing to configure; ServiceName resolves to IHostEnvironment.ApplicationName.
• Multiple instances sharing one SQL Server database — point every instance's ConnectionString/ConnectionStringName at the same database and give each a distinct ServiceName. All of them write to the same tables and are distinguishable in every dashboard view by service name; any instance's /nafas mount shows data from all of them.

SQLite is a single-file, single-process store and is not meant to be shared across instances — use SQL Server for that case.

Requirements

• The package targets netstandard2.0 for the broadest possible reach on the consuming app's side: .NET Core 2.0+, .NET 5+, or classic .NET Framework 4.6.1+ can all reference it (ConnectionStringName's classic web.config support exists for exactly that last case).
• ASP.NET Core only for the core package — UseNafasDashboard requires IApplicationBuilder. For WinForms/WPF/console apps, see Nafas.Observability.Server instead.
• SQLite (zero setup) or SQL Server 2016+ for storage.
• memoryUsage (one of several resource metrics — logs, traces, CPU usage, and alerting are all unaffected either way) is read differently depending on the host, in this order:
  1. .NET Core 3.0+ / .NET 5+ (the common case) — GC.GetGCMemoryInfo(), which is container-aware: it reflects a cgroup or Docker memory limit when one is set, not just physical host RAM.
  2. Classic .NET Framework 4.6.1+, or .NET Core 2.x, on Windows — falls back to the Win32 GlobalMemoryStatusEx API (the same one .NET Framework apps have always used for this, no managed equivalent exists there). This reports whole-machine physical memory, not container/Job-Object-aware, so a process capped by a Job Object memory limit will under-report here.
  3. Any non-Windows host without GC.GetGCMemoryInfo() (practically: .NET Core 2.x on Linux/macOS) — reports 0. This path is explicitly gated to Windows only, so a Linux container never attempts a kernel32.dll call in the first place.

How it's built internally

For anyone evaluating this beyond the quick start:

• Ingestion (Ingestion/) is entirely in-process: an ILoggerProvider, an ActivityListener, and a MeterListener, each writing into a bounded per-signal Channel<T> (capacity 20,000, drop-on-full). A single hosted service drains all three, batches writes (up to 500 records or every second, whichever comes first), and persists them.
• Storage (Storage/) is a thin interface (INafasIngestionWriter/INafasQueryStore) with two concrete implementations, SQLite and SQL Server — no ORM, parameterized SQL, schema created and migrated at startup.
• Alert evaluation (NafasAlertEvaluationHostedService) is a polling BackgroundService reusing the same query layer the dashboard's KPI views use, tracking one open-incident-per-rule at a time.
• Real-time delivery to the browser is an in-process pub/sub (NafasLiveFeed) fed by the ingestion writer and alert evaluator, read by the dashboard's SSE endpoints — nothing here is a polling query.
• The dashboard UI (ClientApp/) is a Vue 3 + Vite SPA (PrimeVue, Pinia, Chart.js, vue-i18n for English/Persian), built ahead of time and embedded directly into the assembly as embedded resources — a consumer installing the package gets the whole UI with no separate static-file deployment step, and no Node.js runtime on the host machine.
• The whole api/* surface (NafasDashboardEndpoints.cs) is a small hand-rolled router rather than ASP.NET Core's IEndpointRouteBuilder, because endpoint routing has been FrameworkReference-only since ASP.NET Core 3.0 and is unavailable to a netstandard2.x class library.

Repository layout

• Nafas.Observability/ — the package itself.
• Nafas.Dashboard.TestHost/ — a minimal ASP.NET Core app used to manually verify the package end to end; not a usage example to copy patterns from.
• Nafas.Observability.Server/ — the optional standalone HTTP listener package for non-ASP.NET Core apps (no Kestrel, no OWIN); see Desktop and other non-ASP.NET Core apps.
• Nafas.Observability.Server.Sample/ — a WPF app demonstrating it, with a settings-screen-style checkbox to enable/disable the dashboard server, pick its port, and choose localhost-only vs. LAN access at runtime.

Local development

For normal use of the published package, no Node.js/npm step is needed at all — the dashboard UI is already built and committed under Nafas.Observability/wwwroot/. Just build the solution:

dotnet build nafas.observability.sln

To try the package locally end to end:

cd Nafas.Dashboard.TestHost
dotnet run

then open http://localhost:<port>/nafas.

To change the dashboard UI itself:

cd Nafas.Observability/ClientApp
npm install        # first time only
npm run build       # output goes to ClientApp/dist/

# then replace (not merge) the contents of ../wwwroot/ with dist/'s contents

then dotnet build again so the updated files are re-embedded. .github/workflows/publish.yml does this same rebuild-and-sync step from source on every publish, so a tagged release never ships a stale UI.

Status

Pre-1.0 (see the NuGet badges at the top of this file for the exact versions) — real ingestion, storage, dashboard, and alert evaluation all work and are exercised through Nafas.Dashboard.TestHost, but this has not yet run in a production deployment. The public API surface (NafasServerOptions, UseNafasDashboard) is expected to stay stable, but has not been through a real upgrade cycle yet.

Nafas.Observability.Server is newer still (its first release) — exercised through the smoke-tested scenarios described in its own comments (static files, the JSON API, SSE streaming, restart/stop, the local-only security default, and genuine LAN reachability without administrator rights), but, like the core package, not yet through a real desktop deployment.

License

FSL-1.1-ALv2 — free to use, modify and redistribute for any purpose other than building a competing product or service; each release converts to the Apache License, Version 2.0 two years after it ships. See LICENSE.md.