OTSSignsOrchestrator/OTSSignsOrchestrator.Server/Workers/SuspendPipeline.cs

using Microsoft.AspNetCore.SignalR;
using Microsoft.EntityFrameworkCore;
using Renci.SshNet;
using OTSSignsOrchestrator.Core.Services;
using OTSSignsOrchestrator.Server.Data;
using OTSSignsOrchestrator.Server.Data.Entities;
using OTSSignsOrchestrator.Server.Hubs;

namespace OTSSignsOrchestrator.Server.Workers;

/// <summary>
/// Subscription suspension pipeline — scales down Docker services and marks status.
/// Handles <c>JobType = "suspend"</c>.
///
/// Steps:
///   1. scale-down     — SSH docker service scale web=0, xmr=0
///   2. update-status  — Customer.Status = Suspended, Instance.HealthStatus = Degraded
///   3. audit-log      — Append-only AuditLog entry
///   4. broadcast      — InstanceStatusChanged via FleetHub
/// </summary>
public sealed class SuspendPipeline : IProvisioningPipeline
{
    public string HandlesJobType => "suspend";

    private const int TotalSteps = 4;

    private readonly IServiceProvider _services;
    private readonly ILogger<SuspendPipeline> _logger;

    public SuspendPipeline(
        IServiceProvider services,
        ILogger<SuspendPipeline> logger)
    {
        _services = services;
        _logger = logger;
    }

    public async Task ExecuteAsync(Job job, CancellationToken ct)
    {
        await using var scope = _services.CreateAsyncScope();
        var db = scope.ServiceProvider.GetRequiredService<OrchestratorDbContext>();
        var hub = scope.ServiceProvider.GetRequiredService<IHubContext<FleetHub, IFleetClient>>();
        var settings = scope.ServiceProvider.GetRequiredService<SettingsService>();

        var ctx = await BuildContextAsync(job, db, ct);
        var runner = new StepRunner(db, hub, _logger, job.Id, TotalSteps);

        var abbrev = ctx.Abbreviation;

        // ── Step 1: scale-down ──────────────────────────────────────────────
        await runner.RunAsync("scale-down", async () =>
        {
            var sshHost = await GetSwarmSshHostAsync(settings);
            using var sshClient = CreateSshClient(sshHost);
            sshClient.Connect();

            try
            {
                var cmd = $"docker service scale xibo-{abbrev}_web=0 xibo-{abbrev}_xmr=0";
                var result = RunSshCommand(sshClient, cmd);
                return $"Scaled down services for xibo-{abbrev}. Output: {result}";
            }
            finally
            {
                sshClient.Disconnect();
            }
        }, ct);

        // ── Step 2: update-status ───────────────────────────────────────────
        await runner.RunAsync("update-status", async () =>
        {
            var customer = await db.Customers.FirstOrDefaultAsync(c => c.Id == ctx.CustomerId, ct)
                ?? throw new InvalidOperationException($"Customer {ctx.CustomerId} not found.");
            var instance = await db.Instances.FirstOrDefaultAsync(i => i.Id == ctx.InstanceId, ct)
                ?? throw new InvalidOperationException($"Instance {ctx.InstanceId} not found.");

            customer.Status = CustomerStatus.Suspended;
            instance.HealthStatus = HealthStatus.Degraded;
            await db.SaveChangesAsync(ct);

            return $"Customer '{abbrev}' status → Suspended, instance health → Degraded.";
        }, ct);

        // ── Step 3: audit-log ───────────────────────────────────────────────
        await runner.RunAsync("audit-log", async () =>
        {
            db.AuditLogs.Add(new AuditLog
            {
                Id = Guid.NewGuid(),
                InstanceId = ctx.InstanceId,
                Actor = "system/stripe-webhook",
                Action = "suspend",
                Target = $"xibo-{abbrev}",
                Outcome = "success",
                Detail = $"Subscription suspended. Services scaled to 0. Job {job.Id}.",
                OccurredAt = DateTime.UtcNow,
            });
            await db.SaveChangesAsync(ct);

            return "AuditLog entry written for suspension.";
        }, ct);

        // ── Step 4: broadcast ───────────────────────────────────────────────
        await runner.RunAsync("broadcast", async () =>
        {
            await hub.Clients.All.SendInstanceStatusChanged(
                ctx.CustomerId.ToString(), CustomerStatus.Suspended.ToString());
            return "Broadcast InstanceStatusChanged → Suspended.";
        }, ct);

        _logger.LogInformation("SuspendPipeline completed for job {JobId} (abbrev={Abbrev})", job.Id, abbrev);
    }

    // ─────────────────────────────────────────────────────────────────────────
    // Helpers (shared pattern from Phase1Pipeline)
    // ─────────────────────────────────────────────────────────────────────────

    private static async Task<PipelineContext> BuildContextAsync(Job job, OrchestratorDbContext db, CancellationToken ct)
    {
        var customer = await db.Customers
            .Include(c => c.Instances)
            .FirstOrDefaultAsync(c => c.Id == job.CustomerId, ct)
            ?? throw new InvalidOperationException($"Customer {job.CustomerId} not found for job {job.Id}.");

        var instance = customer.Instances.FirstOrDefault()
            ?? throw new InvalidOperationException($"No instance found for customer {job.CustomerId}.");

        var abbrev = customer.Abbreviation.ToLowerInvariant();

        return new PipelineContext
        {
            JobId = job.Id,
            CustomerId = customer.Id,
            InstanceId = instance.Id,
            Abbreviation = abbrev,
            CompanyName = customer.CompanyName,
            AdminEmail = customer.AdminEmail,
            AdminFirstName = customer.AdminFirstName,
            InstanceUrl = instance.XiboUrl,
            DockerStackName = instance.DockerStackName,
            ParametersJson = job.Parameters,
        };
    }

    private static async Task<SshConnectionInfo> GetSwarmSshHostAsync(SettingsService settings)
    {
        var host = await settings.GetAsync("Ssh.SwarmHost")
            ?? throw new InvalidOperationException("SSH Swarm host not configured (Ssh.SwarmHost).");
        var portStr = await settings.GetAsync("Ssh.SwarmPort", "22");
        var user = await settings.GetAsync("Ssh.SwarmUser", "root");
        var keyPath = await settings.GetAsync("Ssh.SwarmKeyPath");
        var password = await settings.GetAsync("Ssh.SwarmPassword");

        if (!int.TryParse(portStr, out var port)) port = 22;

        return new SshConnectionInfo(host, port, user, keyPath, password);
    }

    private static SshClient CreateSshClient(SshConnectionInfo info)
    {
        var authMethods = new List<AuthenticationMethod>();

        if (!string.IsNullOrEmpty(info.KeyPath))
            authMethods.Add(new PrivateKeyAuthenticationMethod(info.Username, new PrivateKeyFile(info.KeyPath)));

        if (!string.IsNullOrEmpty(info.Password))
            authMethods.Add(new PasswordAuthenticationMethod(info.Username, info.Password));

        if (authMethods.Count == 0)
        {
            var defaultKeyPath = Path.Combine(
                Environment.GetFolderPath(Environment.SpecialFolder.UserProfile), ".ssh", "id_rsa");
            if (File.Exists(defaultKeyPath))
                authMethods.Add(new PrivateKeyAuthenticationMethod(info.Username, new PrivateKeyFile(defaultKeyPath)));
            else
                throw new InvalidOperationException(
                    $"No SSH authentication method available for {info.Host}:{info.Port}.");
        }

        var connInfo = new Renci.SshNet.ConnectionInfo(info.Host, info.Port, info.Username, authMethods.ToArray());
        return new SshClient(connInfo);
    }

    private static string RunSshCommand(SshClient client, string command)
    {
        using var cmd = client.RunCommand(command);
        if (cmd.ExitStatus != 0)
            throw new InvalidOperationException(
                $"SSH command failed (exit {cmd.ExitStatus}): {cmd.Error}");
        return cmd.Result;
    }

    internal sealed record SshConnectionInfo(
        string Host, int Port, string Username, string? KeyPath, string? Password);
}