MoShiro/backend/src/MoShiro.Backend.Application/Updates/BackendUpdates.cs

using System.Globalization;
using Fleet.Core;
using Microsoft.Extensions.Logging;
using MoShiro.Protocol;

namespace MoShiro.Backend.Application;

/// <summary>让外面的服务管理器（Quadlet 的 <c>Pull=always</c> 加 <c>Restart=always</c>）换上新镜像：Backend 只负责有序退出。</summary>
public interface IBackendRestart
{
    void Request();
}

/// <summary>更新之前备份数据库：新版启动时会迁移数据库，旧版不认比自己新的库，回滚要靠这份备份。</summary>
public interface IDatabaseBackup
{
    /// <summary>在 <paramref name="directory"/> 里写一份一致的数据库副本，返回文件路径。</summary>
    ValueTask<string> BackupAsync(string directory, string name, CancellationToken ct);
}

public sealed record BackendUpdateView(string Channel, bool Enabled, string State, string? CurrentTag, string? CurrentDigest, ReleaseOffer? Available, string? Error);

/// <summary>
/// Backend 的发布通知和确认，和 Agent 是同一种更新：面板登记 CI 签名的清单，管理员确认之后 Backend 有序退出，
/// Quadlet 单元每次启动都先拉最新镜像（<c>Pull=always</c>），所以退出再被拉起就是更新。Backend 自己不下载、不替换任何东西。
/// 只有设了环境变量 <c>MOSHIRO_UPDATE=restart</c>（Quadlet 模板里写好了）才启用；别的部署方式退出之后不会重新拉取。
/// 只提示不自动重启：Backend 指挥所有节点，重启会让它们全部断一两分钟。
/// </summary>
public sealed partial class BackendUpdates
{
    /// <summary>更新前的数据库备份保留几份。</summary>
    private const int BackupsKept = 3;

    private readonly ReleaseService _releases;
    private readonly BackendOptions _options;
    private readonly IBackendRestart _restart;
    private readonly IDatabaseBackup _backup;
    private readonly ILogger<BackendUpdates> _log;
    private readonly TimeProvider _time;
    private readonly object _gate = new();
    private long _applyingSeq;

    public BackendUpdates(ReleaseService releases, BackendOptions options, IBackendRestart restart, IDatabaseBackup backup, ILogger<BackendUpdates> log, TimeProvider time)
    {
        _releases = releases;
        _options = options;
        _restart = restart;
        _backup = backup;
        _log = log;
        _time = time;
    }

    public static string ChannelOf(BackendOptions options)
    {
        if (options.UpdateChannel is "dev" or "release")
        {
            return options.UpdateChannel;
        }

        if (options.UpdateChannel != "auto")
        {
            throw new InvalidOperationException("UpdateChannel must be auto, dev or release");
        }

        string imageNamespace = options.ImageNamespace.Trim('/');
        if (imageNamespace == "moshiro-dev" || imageNamespace.EndsWith("/moshiro-dev", StringComparison.Ordinal))
        {
            return "dev";
        }

        if (imageNamespace == "moshiro" || imageNamespace.EndsWith("/moshiro", StringComparison.Ordinal))
        {
            return "release";
        }

        return IsDevTag(BuildInfo.Tag) ? "dev" : "release";
    }

    /// <summary>开发构建的 tag：新格式 v0.4.1-dev.12-abc1234（含 -dev.），以及旧的 dev-&lt;sha&gt;、main-&lt;sha&gt;。</summary>
    public static bool IsDevTag(string? tag) =>
        tag is not null
        && (tag.Contains("-dev.", StringComparison.Ordinal) || tag.StartsWith("dev-", StringComparison.Ordinal) || tag.StartsWith("main-", StringComparison.Ordinal));

    /// <summary>环境变量 <c>MOSHIRO_UPDATE=restart</c> 才启用。</summary>
    public bool Enabled => string.Equals(_options.UpdateMode, "restart", StringComparison.Ordinal);

    public async ValueTask<BackendUpdateView> GetAsync(CancellationToken ct)
    {
        string channel = ChannelOf(_options);
        ReleaseOffer? offer = await _releases.GetBackendAsync(channel, ct);
        bool installed = offer is not null && IsInstalled(offer);
        long applying;
        lock (_gate)
        {
            applying = _applyingSeq;
        }

        string state = !Enabled ? "unavailable"
            : installed || offer is null ? "up_to_date"
            : applying == offer.Seq ? "applying"
            : "available";
        return new BackendUpdateView(channel, Enabled, state, BuildInfo.Tag, null, installed ? null : offer, null);
    }

    public async ValueTask<Result<bool>> ConfirmAsync(long seq, CancellationToken ct)
    {
        // 与 webhook 共用发布锁：确认的是此刻面板上登记的那一份清单，确认之后又来了新 tag 也不会改变要更新到哪里。
        await _releases.BackendGate.WaitAsync(ct);
        try
        {
            BackendUpdateView view = await GetAsync(ct);
            if (!view.Enabled || view.State != "available" || view.Available?.Seq != seq)
            {
                return Result<bool>.Fail(MoShiroErrors.PreconditionFailed, view.Enabled
                    ? "the offered release changed or is already installed; refresh before confirming"
                    : "updating needs MOSHIRO_UPDATE=restart (the Quadlet unit sets it)");
            }

            string name = $"pre-update-{_time.GetUtcNow():yyyyMMddTHHmmssZ}-{seq}.db";
            string directory = Path.Combine(_options.DataDirectory, "backups");
            try
            {
                string path = await _backup.BackupAsync(directory, name, ct);
                BackedUp(path);
                Prune(directory);
            }
            catch (Exception ex) when (ex is IOException or InvalidOperationException or UnauthorizedAccessException)
            {
                // 备份不了就不更新：旧版不认新库，没有备份就没有回头路
                BackupFailed(ex);
                return Result<bool>.Fail(MoShiroErrors.Internal, "the database could not be backed up; the update was not started");
            }

            lock (_gate)
            {
                _applyingSeq = seq;
            }

            Confirmed(view.Available.Tag);
            _restart.Request();
            return Result<bool>.Ok(true);
        }
        finally
        {
            _releases.BackendGate.Release();
        }
    }

    [LoggerMessage(Level = LogLevel.Information, Message = "database backed up before the update to {Path}")]
    private partial void BackedUp(string path);

    [LoggerMessage(Level = LogLevel.Error, Message = "the database backup failed; the update was not started")]
    private partial void BackupFailed(Exception ex);

    [LoggerMessage(Level = LogLevel.Information, Message = "update to {Tag} confirmed; exiting so the supervisor pulls the latest image")]
    private partial void Confirmed(string tag);

    private static bool IsInstalled(ReleaseOffer offer)
    {
        if (BuildInfo.Tag.Length > 0 && offer.Tag == BuildInfo.Tag)
        {
            return true;
        }

        // 清单的 seq 是发布时间（秒）：不比正在运行的这个构建新，就没有更新
        return DateTimeOffset.TryParse(BuildInfo.BuiltAt, CultureInfo.InvariantCulture, DateTimeStyles.AssumeUniversal, out DateTimeOffset built) && offer.Seq <= built.ToUnixTimeSeconds();
    }

    private static void Prune(string directory)
    {
        foreach (FileInfo old in new DirectoryInfo(directory).GetFiles("pre-update-*.db").OrderByDescending(f => f.Name, StringComparer.Ordinal).Skip(BackupsKept))
        {
            old.Delete();
        }
    }
}