scadalink-design/src/ScadaLink.AuditLog/Site/Telemetry/CachedCallTelemetryForwarder.cs

using Microsoft.Extensions.Logging;
using ScadaLink.Commons.Entities.Audit;
using ScadaLink.Commons.Interfaces;
using ScadaLink.Commons.Interfaces.Services;
using ScadaLink.Commons.Messages.Integration;
using ScadaLink.Commons.Types;
using ScadaLink.Commons.Types.Enums;

namespace ScadaLink.AuditLog.Site.Telemetry;

/// <summary>
/// Site-side dual emitter for cached-call lifecycle telemetry (Audit Log #23 /
/// M3). Sister to <see cref="SiteAuditTelemetryActor"/>: where the M2 actor
/// drains audit-only events, this forwarder takes a combined
/// <see cref="CachedCallTelemetry"/> packet and fans it out to the two
/// site-local stores in a single call:
/// <list type="bullet">
///   <item><description>The <see cref="AuditEvent"/> row is written via
///   <see cref="IAuditWriter"/> (the site <c>FallbackAuditWriter</c> +
///   <c>SqliteAuditWriter</c> chain established in M2).</description></item>
///   <item><description>The operational <see cref="SiteCallOperational"/> half
///   updates the site-local <c>OperationTracking</c> SQLite store via
///   <see cref="IOperationTrackingStore"/>, with the per-lifecycle method
///   (<c>Enqueue</c> / <c>Attempt</c> / <c>Terminal</c>) selected from the
///   audit row's <see cref="AuditKind"/>.</description></item>
/// </list>
/// </summary>
/// <remarks>
/// <para>
/// <b>Best-effort contract (alog.md §7):</b> a thrown writer OR a thrown
/// tracking store must never propagate to the calling script. Both emission
/// halves are wrapped in independent try/catch blocks so a SQLite outage on
/// one side cannot starve the other — the failure is logged and the call
/// returns normally.
/// </para>
/// <para>
/// <b>Wire push deferred to M6.</b> M3 keeps this forwarder synchronous
/// against the local stores: there is no site→central gRPC channel yet, so
/// the <see cref="ISiteStreamAuditClient.IngestCachedTelemetryAsync"/> RPC
/// is registered on the interface (Bundle E1) but the production binding
/// remains <c>NoOpSiteStreamAuditClient</c>. Once M6 wires a real client the
/// drain pattern from <c>SiteAuditTelemetryActor</c> can be reused — the
/// <c>AuditEvent</c> rows already live in SQLite tagged
/// <see cref="AuditForwardState.Pending"/>, so a single drain loop sweeps
/// both M2 and M3 emissions.
/// </para>
/// </remarks>
public sealed class CachedCallTelemetryForwarder : ICachedCallTelemetryForwarder
{
    private readonly IAuditWriter _auditWriter;
    private readonly IOperationTrackingStore _trackingStore;
    private readonly ILogger<CachedCallTelemetryForwarder> _logger;

    public CachedCallTelemetryForwarder(
        IAuditWriter auditWriter,
        IOperationTrackingStore trackingStore,
        ILogger<CachedCallTelemetryForwarder> logger)
    {
        _auditWriter = auditWriter ?? throw new ArgumentNullException(nameof(auditWriter));
        _trackingStore = trackingStore ?? throw new ArgumentNullException(nameof(trackingStore));
        _logger = logger ?? throw new ArgumentNullException(nameof(logger));
    }

    /// <summary>
    /// Fan out one combined-telemetry packet to the audit writer and the
    /// tracking store. Returns once both halves have been attempted (success
    /// OR logged failure). NEVER throws — exceptions are caught per-half and
    /// logged at warning level so the calling script's outbound action is not
    /// disturbed.
    /// </summary>
    public async Task ForwardAsync(CachedCallTelemetry telemetry, CancellationToken ct = default)
    {
        ArgumentNullException.ThrowIfNull(telemetry);

        // Independent try/catch — a thrown audit writer must not prevent the
        // tracking-store update from running (and vice-versa). Both halves
        // are best-effort.
        await TryEmitAuditAsync(telemetry, ct).ConfigureAwait(false);
        await TryEmitTrackingAsync(telemetry, ct).ConfigureAwait(false);
    }

    private async Task TryEmitAuditAsync(CachedCallTelemetry telemetry, CancellationToken ct)
    {
        try
        {
            await _auditWriter.WriteAsync(telemetry.Audit, ct).ConfigureAwait(false);
        }
        catch (Exception ex)
        {
            // alog.md §7 best-effort contract — log and swallow. The audit
            // pipeline's own retry/recovery (RingBufferFallback in the
            // FallbackAuditWriter) handles transient writer failures upstream;
            // a throw bubbling up here means the writer's own swallow contract
            // failed, which is itself best-effort-handled.
            _logger.LogWarning(ex,
                "CachedCallTelemetryForwarder: audit emission threw for EventId {EventId} (Kind {Kind}, Status {Status})",
                telemetry.Audit.EventId, telemetry.Audit.Kind, telemetry.Audit.Status);
        }
    }

    private async Task TryEmitTrackingAsync(CachedCallTelemetry telemetry, CancellationToken ct)
    {
        try
        {
            switch (telemetry.Audit.Kind)
            {
                case AuditKind.CachedSubmit:
                    // Enqueue — insert-if-not-exists with the operational
                    // channel as the kind discriminator. RetryCount is fixed
                    // at 0 by the tracking store's INSERT contract.
                    await _trackingStore.RecordEnqueueAsync(
                        telemetry.Operational.TrackedOperationId,
                        telemetry.Operational.Channel,
                        telemetry.Operational.Target,
                        telemetry.Audit.SourceInstanceId,
                        telemetry.Audit.SourceScript,
                        ct).ConfigureAwait(false);
                    break;

                case AuditKind.ApiCallCached:
                case AuditKind.DbWriteCached:
                    // Attempt — advance retry counter + last-error/HTTP-status.
                    // Terminal rows are guarded by the store's WHERE clause.
                    await _trackingStore.RecordAttemptAsync(
                        telemetry.Operational.TrackedOperationId,
                        telemetry.Operational.Status,
                        telemetry.Operational.RetryCount,
                        telemetry.Operational.LastError,
                        telemetry.Operational.HttpStatus,
                        ct).ConfigureAwait(false);
                    break;

                case AuditKind.CachedResolve:
                    // Terminal — first-write-wins on the resolve flip.
                    await _trackingStore.RecordTerminalAsync(
                        telemetry.Operational.TrackedOperationId,
                        telemetry.Operational.Status,
                        telemetry.Operational.LastError,
                        telemetry.Operational.HttpStatus,
                        ct).ConfigureAwait(false);
                    break;

                default:
                    // Defensive — only the four cached-lifecycle kinds are
                    // expected on this path. Anything else is logged so a
                    // mis-routed packet is visible but never crashes the
                    // forwarder.
                    _logger.LogWarning(
                        "CachedCallTelemetryForwarder: unexpected audit kind {Kind} on tracking emission for EventId {EventId}",
                        telemetry.Audit.Kind, telemetry.Audit.EventId);
                    break;
            }
        }
        catch (Exception ex)
        {
            _logger.LogWarning(ex,
                "CachedCallTelemetryForwarder: tracking-store emission threw for TrackedOperationId {Id} (Status {Status})",
                telemetry.Operational.TrackedOperationId, telemetry.Operational.Status);
        }
    }
}