主要是为了未来的我:)

我用Rx写了一个包装器:

 public class WatcherWrapper : IDisposable
{
    private readonly FileSystemWatcher _fileWatcher;
    private readonly Subject<FileSystemEventArgs> _infoSubject;
    private Subject<FileSystemEventArgs> _eventSubject;

    public WatcherWrapper(string path, string nameFilter = "*.*", NotifyFilters? notifyFilters = null)
    {
        _fileWatcher = new FileSystemWatcher(path, nameFilter);

        if (notifyFilters != null)
        {
            _fileWatcher.NotifyFilter = notifyFilters.Value;
        }

        _infoSubject = new Subject<FileSystemEventArgs>();
        _eventSubject = new Subject<FileSystemEventArgs>();

        Observable.FromEventPattern<FileSystemEventArgs>(_fileWatcher, "Changed").Select(e => e.EventArgs)
            .Subscribe(_infoSubject.OnNext);
        Observable.FromEventPattern<FileSystemEventArgs>(_fileWatcher, "Created").Select(e => e.EventArgs)
            .Subscribe(_infoSubject.OnNext);
        Observable.FromEventPattern<FileSystemEventArgs>(_fileWatcher, "Deleted").Select(e => e.EventArgs)
            .Subscribe(_infoSubject.OnNext);
        Observable.FromEventPattern<FileSystemEventArgs>(_fileWatcher, "Renamed").Select(e => e.EventArgs)
            .Subscribe(_infoSubject.OnNext);

        // this takes care of double events and still works with changing the name of the same file after a while
        _infoSubject.Buffer(TimeSpan.FromMilliseconds(20))
            .Select(x => x.GroupBy(z => z.FullPath).Select(z => z.LastOrDefault()).Subscribe(
                infos =>
                {
                    if (infos != null)
                        foreach (var info in infos)
                        {
                            {
                                _eventSubject.OnNext(info);
                            }
                        }
                });

        _fileWatcher.EnableRaisingEvents = true;
    }

    public IObservable<FileSystemEventArgs> FileEvents => _eventSubject;


    public void Dispose()
    {
        _fileWatcher?.Dispose();
        _eventSubject.Dispose();
        _infoSubject.Dispose();
    }
}

用法:

var watcher = new WatcherWrapper(_path, "*.info");
// all more complicated and scenario specific filtering of events can be done here    
watcher.FileEvents.Where(x => x.ChangeType != WatcherChangeTypes.Deleted).Subscribe(x => //do stuff)

我的场景是，我有一个虚拟机，其中有一个Linux服务器。我正在Windows主机上开发文件。当我在主机上的一个文件夹中更改某些内容时，我希望所有的更改都被上传，并通过Ftp同步到虚拟服务器上。这是我如何消除重复更改事件当我写入一个文件(这标志着包含文件要修改的文件夹):

private Hashtable fileWriteTime = new Hashtable();

private void fsw_sync_Changed(object source, FileSystemEventArgs e)
{
    string path = e.FullPath.ToString();
    string currentLastWriteTime = File.GetLastWriteTime( e.FullPath ).ToString();

    // if there is no path info stored yet
    // or stored path has different time of write then the one now is inspected
    if ( !fileWriteTime.ContainsKey(path) ||
         fileWriteTime[path].ToString() != currentLastWriteTime
    )
    {
        //then we do the main thing
        log( "A CHANGE has occured with " + path );

        //lastly we update the last write time in the hashtable
        fileWriteTime[path] = currentLastWriteTime;
    }
}

我主要创建一个哈希表来存储文件写时间信息。然后，如果哈希表有被修改的文件路径，并且它的时间值与当前通知的文件更改相同，那么我就知道它是事件的副本，并忽略它。

我只想对最后一个事件做出反应，以防万一，也对linux文件更改，它似乎在第一次调用时文件是空的，然后在下一个调用时再次填充，不介意浪费一些时间，以防操作系统决定做一些文件/属性更改。

我在这里使用。net async来帮助我做线程。

    private static int _fileSystemWatcherCounts;
    private async void OnChanged(object sender, FileSystemEventArgs e)
    {
        // Filter several calls in short period of time
        Interlocked.Increment(ref _fileSystemWatcherCounts);
        await Task.Delay(100);
        if (Interlocked.Decrement(ref _fileSystemWatcherCounts) == 0)
            DoYourWork();
    }

我认为解决这个问题的最佳方案是使用响应式扩展 当你将事件转换为可观察对象时，你可以添加Throttling(..)(最初称为Debounce(..))

这里是示例代码

        var templatesWatcher = new FileSystemWatcher(settingsSnapshot.Value.TemplatesDirectory)
        {
            NotifyFilter = NotifyFilters.LastWrite,
            IncludeSubdirectories = true
        };

        templatesWatcher.EnableRaisingEvents = true;

        Observable.FromEventPattern<FileSystemEventHandler, FileSystemEventArgs>(
                addHandler => templatesWatcher.Changed += addHandler,
                removeHandler => templatesWatcher.Changed -= removeHandler)
            .Throttle(TimeSpan.FromSeconds(5))
            .Subscribe(args =>
            {
                _logger.LogInformation($"Template file {args.EventArgs.Name} has changed");
                //TODO do something
            });

以下是我的方法:

// Consider having a List<String> named _changedFiles

private void OnChanged(object source, FileSystemEventArgs e)
{
    lock (_changedFiles)
    {
        if (_changedFiles.Contains(e.FullPath))
        {
            return;
        }
        _changedFiles.Add(e.FullPath);
    }

    // do your stuff

    System.Timers.Timer timer = new Timer(1000) { AutoReset = false };
    timer.Elapsed += (timerElapsedSender, timerElapsedArgs) =>
    {
        lock (_changedFiles)
        {
            _changedFiles.Remove(e.FullPath);
        }
    };
   timer.Start();
}

这是我在一个项目中用来解决这个问题的解决方案，在该项目中，我将文件作为附件发送到邮件中。 它可以很容易地避免两次触发事件，即使是一个较小的定时器间隔，但在我的情况下，1000是可以的，因为我更喜欢错过一些变化，而不是每秒用> 1条消息淹没邮箱。 至少在同时更改多个文件的情况下，它可以正常工作。

Another solution I've thought of would be to replace the list with a dictionary mapping files to their respective MD5, so you wouldn't have to choose an arbitrary interval since you wouldn't have to delete the entry but update its value, and cancel your stuff if it hasn't changed. It has the downside of having a Dictionary growing in memory as files are monitored and eating more and more memory, but I've read somewhere that the amount of files monitored depends on the FSW's internal buffer, so maybe not that critical. Dunno how MD5 computing time would affect your code's performances either, careful =\

解决方案实际上取决于用例。您是否在注意不更改的新文件，或每隔一段时间更改一次的文件或经常更改的文件?在我的情况下，它的变化不太频繁，我不想错过任何这些变化。

但是我也不想在写入过程尚未完成写入的地方发生更改事件。

在我的情况下，我注意到6 (6 !!)onchange事件写一个125字符的txt文件。

我的解决方案是民意调查和改变事件的混合，民意调查经常被消极地看待。正常轮询比较慢，比如每10秒一次，以防FileSystemWatcher (FSW)“错过”一个事件。轮询立即响应FSW更改事件。

关键是在FSW。更改事件时，轮询速度加快，例如每100毫秒，并等待直到文件稳定。因此我们有了“两阶段轮询”:阶段1比较慢，但在FSW文件更改事件时立即响应。第二阶段是快速的，等待一个稳定的文件。

如果FSW检测到多个文件更改，每个事件都会加速轮询循环，并有效地启动一个新的短等待周期。只有在轮询循环检测到上次写入时文件没有进一步的变化之后，它才假定文件是稳定的，并且您的代码可以处理更改后的文件。

我选择了10秒和100毫秒的超时，但是您的用例可能需要不同的超时值。

这里是轮询，其中AppConfig。fiIO是要注意的FileInfo:

private readonly EventWaitHandle ewhTimeout = new AutoResetEvent(false);

private void TwoPhasedPolling()
{
    bool WaitForChange = true; //false: wait until stable
    DateTime LastWriteTime = DateTime.MinValue;
    while (true)
    {
        // wait for next poll (timeout), or FSW event
        bool GotOne = ewhTimeout.WaitOne(WaitForChange ? 10 * 1000 : 100);
        if (GotOne)
        {
            // WaitOne interrupted: end of Phase1: FSW detected file change
            WaitForChange = false;
        }
        else
        {
            // WaitOne timed out: Phase2: check file write time for change
            if (AppConfig.fiIO.LastWriteTime > LastWriteTime)
            {
                LastWriteTime = AppConfig.fiIO.LastWriteTime;
            }
            else
            {
                // End of Phase2: file has changed and is stable
                WaitForChange = true;
                // action on changed file
                ... your code here ...
            }}}}

private void fileSystemWatcher1_Changed(object sender, FileSystemEventArgs e)
{
    ewhTimeout.Set();
}

NB:是的，我也不喜欢}}}}，但它使列表更短，这样你就不必滚动了:-)