OTSSignsOrchestrator/OTSSignsOrchestrator/Health/Checks/NfsAccessHealthCheck.cs

using OTSSignsOrchestrator.Data.Entities;
using OTSSignsOrchestrator.Services;

namespace OTSSignsOrchestrator.Health.Checks;

/// <summary>
/// Verifies NFS paths for the instance are accessible by running <c>ls</c> via SSH.
/// </summary>
public sealed class NfsAccessHealthCheck : IHealthCheck
{
    private readonly IServiceProvider _services;
    private readonly ILogger<NfsAccessHealthCheck> _logger;

    public string CheckName => "NfsAccess";
    public bool AutoRemediate => false;

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

    public async Task<HealthCheckResult> RunAsync(Instance instance, CancellationToken ct)
    {
        var nfsPath = instance.NfsPath;
        if (string.IsNullOrEmpty(nfsPath))
            return new HealthCheckResult(HealthStatus.Critical, "No NFS path configured");

        try
        {
            var settings = _services.GetRequiredService<SettingsService>();
            var nfsServer = await settings.GetAsync(SettingsService.NfsServer);
            var nfsExport = await settings.GetAsync(SettingsService.NfsExport);

            if (string.IsNullOrEmpty(nfsServer) || string.IsNullOrEmpty(nfsExport))
                return new HealthCheckResult(HealthStatus.Critical, "NFS server/export not configured");

            await using var shell = _services.GetRequiredService<SwarmShellService>();

            // Mount temporarily and check the path is listable.
            // In local (privileged container) mode, sudo is a no-op as root.
            var mountPoint = $"/tmp/healthcheck-nfs-{Guid.NewGuid():N}";
            await shell.RunCommandAsync($"sudo mkdir -p {mountPoint}");
            try
            {
                await shell.RunCommandAsync($"sudo mount -t nfs4 {nfsServer}:{nfsExport} {mountPoint}");
                await shell.RunCommandAsync($"ls {mountPoint}/{nfsPath} 2>&1");

                return new HealthCheckResult(HealthStatus.Healthy,
                    $"NFS path accessible: {nfsPath}");
            }
            finally
            {
                await shell.RunCommandAllowFailureAsync($"sudo umount {mountPoint}");
                await shell.RunCommandAllowFailureAsync($"sudo rmdir {mountPoint}");
            }
        }
        catch (Exception ex)
        {
            return new HealthCheckResult(HealthStatus.Critical,
                $"NFS access check failed for {nfsPath}: {ex.Message}");
        }
    }
}