Process sometimes hangs while waiting for Exit

前端 未结 4 1217
天命终不由人
天命终不由人 2021-01-11 10:23

What may be the reason of my process hanging while waiting for exit?

This code has to start powershell script which inside performs many action e.g start recompiling

4条回答
  •  有刺的猬
    2021-01-11 10:42

    Let's start with a recap of the accepted answer in a related post.

    The problem is that if you redirect StandardOutput and/or StandardError the internal buffer can become full. Whatever order you use, there can be a problem:

    • If you wait for the process to exit before reading StandardOutput the process can block trying to write to it, so the process never ends.
    • If you read from StandardOutput using ReadToEnd then your process can block if the process never closes StandardOutput (for example if it never terminates, or if it is blocked writing to StandardError).

    Even the accepted answer, however, struggles with the order of execution in certain cases.

    EDIT: See answers below for how avoid an ObjectDisposedException if the timeout occurs.

    It's in these kind of situations, where you want to orchestrate several events, that Rx really shines.

    Note the .NET implementation of Rx is available as the System.Reactive NuGet package.

    Let's dive in to see how Rx facilitates working with events.

    // Subscribe to OutputData
    Observable.FromEventPattern(process, nameof(Process.OutputDataReceived))
        .Subscribe(
            eventPattern => output.AppendLine(eventPattern.EventArgs.Data),
            exception => error.AppendLine(exception.Message)
        ).DisposeWith(disposables);
    

    FromEventPattern allows us to map distinct occurrences of an event to a unified stream (aka observable). This allows us to handle the events in a pipeline (with LINQ-like semantics). The Subscribe overload used here is provided with an Action> and an Action. Whenever the observed event is raised, its sender and args will be wrapped by EventPattern and pushed through the Action>. When an exception is raised in the pipeline, Action is used.

    One of the drawbacks of the Event pattern, clearly illustrated in this use case (and by all the workarounds in the referenced post), is that it's not apparent when / where to unsubscribe the event handlers.

    With Rx we get back an IDisposable when we make a subscription. When we dispose of it, we effectively end the subscription. With the addition of the DisposeWith extension method (borrowed from RxUI), we can add multiple IDisposables to a CompositeDisposable (named disposables in the code samples). When we're all done, we can end all subscriptions with one call to disposables.Dispose().

    To be sure, there's nothing we can do with Rx, that we wouldn't be able to do with vanilla .NET. The resulting code is just a lot easier to reason about, once you've adapted to the functional way of thinking.

    public static void ExecuteScriptRx(string path, int processTimeOutMilliseconds, out string logs, out bool success, params string[] args)
    {
        StringBuilder output = new StringBuilder();
        StringBuilder error = new StringBuilder();
    
        using (var process = new Process())
        using (var disposables = new CompositeDisposable())
        {
            process.StartInfo = new ProcessStartInfo
            {
                WindowStyle = ProcessWindowStyle.Hidden,
                FileName = "powershell.exe",
                RedirectStandardOutput = true,
                RedirectStandardError = true,
                UseShellExecute = false,
                Arguments = $"-ExecutionPolicy Bypass -File \"{path}\"",
                WorkingDirectory = Path.GetDirectoryName(path)
            };
    
            if (args.Length > 0)
            {
                var arguments = string.Join(" ", args.Select(x => $"\"{x}\""));
                process.StartInfo.Arguments += $" {arguments}";
            }
    
            output.AppendLine($"args:'{process.StartInfo.Arguments}'");
    
            // Raise the Process.Exited event when the process terminates.
            process.EnableRaisingEvents = true;
    
            // Subscribe to OutputData
            Observable.FromEventPattern(process, nameof(Process.OutputDataReceived))
                .Subscribe(
                    eventPattern => output.AppendLine(eventPattern.EventArgs.Data),
                    exception => error.AppendLine(exception.Message)
                ).DisposeWith(disposables);
    
            // Subscribe to ErrorData
            Observable.FromEventPattern(process, nameof(Process.ErrorDataReceived))
                .Subscribe(
                    eventPattern => error.AppendLine(eventPattern.EventArgs.Data),
                    exception => error.AppendLine(exception.Message)
                ).DisposeWith(disposables);
    
            var processExited =
                // Observable will tick when the process has gracefully exited.
                Observable.FromEventPattern(process, nameof(Process.Exited))
                    // First two lines to tick true when the process has gracefully exited and false when it has timed out.
                    .Select(_ => true)
                    .Timeout(TimeSpan.FromMilliseconds(processTimeOutMilliseconds), Observable.Return(false))
                    // Force termination when the process timed out
                    .Do(exitedSuccessfully => { if (!exitedSuccessfully) { try { process.Kill(); } catch {} } } );
    
            // Subscribe to the Process.Exited event.
            processExited
                .Subscribe()
                .DisposeWith(disposables);
    
            // Start process(ing)
            process.Start();
    
            process.BeginOutputReadLine();
            process.BeginErrorReadLine();
    
            // Wait for the process to terminate (gracefully or forced)
            processExited.Take(1).Wait();
    
            logs = output + Environment.NewLine + error;
            success = process.ExitCode == 0;
        }
    }
    

    We already discussed the first part, where we map our events to observables, so we can jump straight to the meaty part. Here we assign our observable to the processExited variable, because we want to use it more than once.

    First, when we activate it, by calling Subscribe. And later on when we want to 'await' its first value.

    var processExited =
        // Observable will tick when the process has gracefully exited.
        Observable.FromEventPattern(process, nameof(Process.Exited))
            // First two lines to tick true when the process has gracefully exited and false when it has timed out.
            .Select(_ => true)
            .Timeout(TimeSpan.FromMilliseconds(processTimeOutMilliseconds), Observable.Return(false))
            // Force termination when the process timed out
            .Do(exitedSuccessfully => { if (!exitedSuccessfully) { try { process.Kill(); } catch {} } } );
    
    // Subscribe to the Process.Exited event.
    processExited
        .Subscribe()
        .DisposeWith(disposables);
    
    // Start process(ing)
    ...
    
    // Wait for the process to terminate (gracefully or forced)
    processExited.Take(1).Wait();
    

    One of the problems with OP is that it assumes process.WaitForExit(processTimeOutMiliseconds) will terminate the process when it times out. From MSDN:

    Instructs the Process component to wait the specified number of milliseconds for the associated process to exit.

    Instead, when it times out, it merely returns control to the current thread (i.e. it stops blocking). You need to manually force termination when the process times out. To know when time out has occurred, we can map the Process.Exited event to a processExited observable for processing. This way we can prepare the input for the Do operator.

    The code is pretty self-explanatory. If exitedSuccessfully the process will have terminated gracefully. If not exitedSuccessfully, termination will need to be forced. Note that process.Kill() is executed asynchronously, ref remarks. However, calling process.WaitForExit() right after will open up the possibility for deadlocks again. So even in the case of forced termination, it's better to let all disposables be cleaned up when the using scope ends, as the output can be considered interrupted / corrupted anyway.

    The try catch construct is reserved for the exceptional case (no pun intended) where you've aligned processTimeOutMilliseconds with the actual time needed by the process to complete. In other words, a race condition occurs between the Process.Exited event and the timer. The possibility of this happening is again magnified by the asynchronous nature of process.Kill(). I've encountered it once during testing.


    For completeness, the DisposeWith extension method.

    /// 
    /// Extension methods associated with the IDisposable interface.
    /// 
    public static class DisposableExtensions
    {
        /// 
        /// Ensures the provided disposable is disposed with the specified .
        /// 
        public static T DisposeWith(this T item, CompositeDisposable compositeDisposable)
            where T : IDisposable
        {
            if (compositeDisposable == null)
            {
                throw new ArgumentNullException(nameof(compositeDisposable));
            }
    
            compositeDisposable.Add(item);
            return item;
        }
    }
    

提交回复
热议问题