ScadaBridge/tests/ZB.MOM.WW.ScadaBridge.Host.Tests/StandbyHealthCheckLogNoiseTests.cs

using Microsoft.Extensions.Configuration;
using Serilog;
using Serilog.Events;

namespace ZB.MOM.WW.ScadaBridge.Host.Tests;

/// <summary>
/// Deferred-work register row 71 — a standby node logged the correct-by-design
/// <c>active-node</c> health-check verdict at ERROR on every probe.
/// </summary>
/// <remarks>
/// <para>
/// <c>DefaultHealthCheckService</c> picks the log level from the verdict (Debug/Warning/Error
/// for Healthy/Degraded/Unhealthy). Both active-node checks answer Unhealthy on the standby
/// because 503 is the contract <c>/health/active</c> publishes, so a healthy standby logged
/// ERR every 10 s on a site node and every 5 s on a central node behind Traefik's probe.
/// </para>
/// <para>
/// The constraint on the fix is that a GENUINE failure must still log at Error, which is why
/// the filter is keyed on the check NAME rather than on the logger category.
/// </para>
/// </remarks>
public class StandbyHealthCheckLogNoiseTests
{
    private const string HealthCheckCategory =
        "Microsoft.Extensions.Diagnostics.HealthChecks.DefaultHealthCheckService";

    private const string Template =
        "Health check {HealthCheckName} with status {HealthStatus} completed after "
        + "{ElapsedMilliseconds}ms with message '{HealthCheckDescription}'";

    private static ILogger BuildLogger(InMemorySink sink)
    {
        var config = new ConfigurationBuilder()
            .AddInMemoryCollection(new Dictionary<string, string?>
            {
                ["ScadaBridge:Logging:MinimumLevel"] = "Debug",
            })
            .Build();

        return LoggerConfigurationFactory
            .Build(config, "Site", "site-a", "site-a-b")
            .WriteTo.Sink(sink)
            .CreateLogger();
    }

    /// <summary>Emits one event shaped exactly like the framework's per-check line.</summary>
    private static void EmitHealthCheckLine(
        ILogger logger, string category, LogEventLevel level, string checkName, string status) =>
        logger
            .ForContext("SourceContext", category)
            .Write(level, Template, checkName, status, 1.0, "some message");

    [Fact]
    public void StandbyActiveNodeVerdict_IsNotLogged()
    {
        // The row-71 regression: "Health check active-node with status Unhealthy … 'not the
        // site pair primary (standby)'" at Error, once per probe, forever.
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        EmitHealthCheckLine(
            logger, HealthCheckCategory, LogEventLevel.Error, "active-node", "Unhealthy");

        Assert.Empty(sink.LogEvents);
    }

    [Theory]
    [InlineData("localdb")]
    [InlineData("akka-cluster")]
    [InlineData("database")]
    [InlineData("required-singletons")]
    public void GenuineCheckFailure_StillLogsAtError(string checkName)
    {
        // The constraint. The filter is name-scoped, so every OTHER check on the same logger
        // category keeps its Error line — a blanket category override would have silenced
        // these too, which is exactly the trap this test exists to catch.
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        EmitHealthCheckLine(
            logger, HealthCheckCategory, LogEventLevel.Error, checkName, "Unhealthy");

        var evt = Assert.Single(sink.LogEvents);
        Assert.Equal(LogEventLevel.Error, evt.Level);
    }

    [Fact]
    public void ActiveNodeHealthyVerdict_IsUnaffected()
    {
        // The active node reports Healthy, which the framework logs at Debug. Nothing about
        // the filter should touch it — it only ever matches Error.
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        EmitHealthCheckLine(
            logger, HealthCheckCategory, LogEventLevel.Debug, "active-node", "Healthy");

        Assert.Single(sink.LogEvents);
    }

    [Fact]
    public void ActiveNodeDegradedVerdict_IsUnaffected()
    {
        // Degraded logs at Warning. If a future check shape ever reports Degraded for
        // active-node, that should still be visible — the filter is Error-only.
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        EmitHealthCheckLine(
            logger, HealthCheckCategory, LogEventLevel.Warning, "active-node", "Degraded");

        Assert.Single(sink.LogEvents);
    }

    [Fact]
    public void AnErrorMentioningActiveNode_FromAnotherCategory_IsNotSuppressed()
    {
        // The predicate requires BOTH the health-check category and the check name, so
        // application code logging about the active node keeps its Error.
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        EmitHealthCheckLine(
            logger,
            "ZB.MOM.WW.ScadaBridge.Host.Actors.AkkaHostedService",
            LogEventLevel.Error,
            "active-node",
            "Unhealthy");

        Assert.Single(sink.LogEvents);
    }

    [Fact]
    public void OrdinaryErrorsWithNoHealthCheckProperties_AreNotSuppressed()
    {
        var sink = new InMemorySink();
        var logger = BuildLogger(sink);

        logger.Error("something genuinely broke");

        Assert.Single(sink.LogEvents);
    }

    [Fact]
    public void Predicate_RejectsANullEvent()
    {
        // Defensive: Serilog never hands the filter a null, but the predicate is internal
        // and reachable, and returning "suppress" for null would be the wrong default.
        Assert.False(LoggerConfigurationFactory.IsExpectedStandbyActiveNodeProbe(null!));
    }
}