I have a bit of weird problem that is hard to explain. I have singleton class where in the constructor I have to run a task to initialize some components/resources.
I used 2 implementation of singleton from C# in Depth and in one case everything is working fine, in another case - not.
Code is available below with some comments.
The main problem that for some reason task is not started in one case, when I am using static field with initialier and static contructor (class Test2).
I made some other tests and looks like with the implementation 2 task does not start asynchronically, but starts synchronically after waiting time.
Implementation one. everything is working as expected
public sealed class Test1
{
private static Test1 instance = null;
private static readonly object padlock = new object();
private Test1()
{
using (AutoResetEvent startEvent = new AutoResetEvent(false))
{
new Task(() => TaskThread(startEvent)).Start();
if (!startEvent.WaitOne(1000))
{
throw new Exception("ERROR");
}
}
}
public int Result()
{
return 10;
}
private void TaskThread(AutoResetEvent startEvent)
{
//I am initializing some stuff here
startEvent.Set();
}
public static Test1 Instance
{
get
{
lock (padlock)
{
if (instance == null)
{
instance = new Test1();
}
return instance;
}
}
}
}
Implementation 2, task is never started, or started after waiting time of an event
public sealed class Test2
{
private static readonly Test2 instance = new Test2();
static Test2()
{
}
private Test2()
{
using (AutoResetEvent startEvent = new AutoResetEvent(false))
{
new Task(() => TaskThread(startEvent)).Start();
//here it fails to wait successfully and throws an
//exception. Time limit is not reached
if (!startEvent.WaitOne(1000))
{
throw new Exception("ERROR");
}
}
}
public int Result()
{
return 20;
}
private void TaskThread(AutoResetEvent startEvent)
{
//I am initializing some stuff here as well
//but in this implementation code is never reached
startEvent.Set();
}
public static Test2 Instance
{
get
{
return instance;
}
}
}
I am curious why is this happening and how to avoid this problems in future. Thanks a lot!
The root cause of such 'strange' behavior is pretty simple - CLR executes static constructor under a lock. That prevents created thread from entering TaskThread() method and setting startEvent to signaled state.
After you face with such a problem and puzzle for several hours why this is happening, you start to understand why many sources advise not to use doubtful constructs like static constructors, global variables, etc.
Related
According to Jon Skeet's article, the following pattern is bad as it is not thread safe.
// Bad code! Do not use!
public sealed class Singleton
{
private static Singleton instance = null;
private Singleton()
{
}
public static Singleton Instance
{
get
{
if (instance == null)
{
instance = new Singleton();
}
return instance;
}
}
}
I have not learnt threading yet so it is a bit abstract to me. Could you give me a simple code to simulate the threading problem (we get notified when the problem occurs)?
Well thats pretty simple, just let something access a property within your singleton in parallel, for example like this console app.
class Program
{
static void Main(string[] args)
{
var threads = Enumerable.Repeat(new Action(() => Console.WriteLine(Singleton.Instance.guid)), 10);
Parallel.ForEach(threads, t => t());
Console.Read();
}
}
(I've added a guid property to your class to test that)
public sealed class Singleton
{
public Guid guid = Guid.NewGuid();
private static Singleton instance = null;
private Singleton()
{
}
public static Singleton Instance
{
get
{
if (instance == null)
{
instance = new Singleton();
}
return instance;
}
}
}
The issue with this singleton implementation is that 2 threads can access the getter simultaneously and each one will create a new instance. So the first thread might end up with a different instance than the second one... which can lead to unexpected behavior.
This is just in reply to OP comment:
static void Main(string[] args)
{
int test = 5;
Task<Singleton>[] arr =
{
Task<Singleton>.Factory.StartNew(() => Singleton.Instance),
Task<Singleton>.Factory.StartNew(() => Singleton.Instance),
};
Task.WaitAll(arr);
foreach (var item in arr)
{
Singleton s = item.Result;
s.MyProperty = test++;
Console.WriteLine(s.MyProperty);
}
}
MyProperty is just an int property i added.
Is there a built-in ThreadLocal<T>-like construct for sharing an object within each unique thread but recreating it if the original value was disposed/destructed/teared down/nulled?
Here's my attempt at implementing such behaviour with ConcurrentDictionary (the ThreadLocalDisposable2 below), but I was hoping to just use ThreadLocal<T> (as in ThreadLocalDisposable1), however I can't get the Foo test to pass, .Values.Remove(this) doesn't do what I was hoping it would do and still causes ObjectDisposedException.
public class Class1
{
[Test]
public void Foo()
{
using (var foo = ThreadLocalDisposable1.Get())
foo.Foo();
using (var foo = ThreadLocalDisposable1.Get())
foo.Foo();
}
[Test]
public void Bar()
{
using (var bar = ThreadLocalDisposable2.Get())
bar.Foo();
using (var bar = ThreadLocalDisposable2.Get())
bar.Foo();
}
}
[1]
public class ThreadLocalDisposable1 : IDisposable
{
private Stream _foo;
private static ThreadLocal<ThreadLocalDisposable1> _thread;
static ThreadLocalDisposable1()
{
_thread = new ThreadLocal<ThreadLocalDisposable1>(() => new ThreadLocalDisposable1(), true);
}
private ThreadLocalDisposable1()
{
_foo = new MemoryStream();
}
public static ThreadLocalDisposable1 Get()
{
return _thread.Value;
}
public void Foo()
{
_foo.WriteByte(1);
}
public void Dispose()
{
//I do not think it means what I think it means
_thread.Values.Remove(this);
_foo.Dispose();
}
}
[2]
public class ThreadLocalDisposable2 : IDisposable
{
private Stream _foo;
private int _thread;
private static ConcurrentDictionary<int, ThreadLocalDisposable2> _threads;
static ThreadLocalDisposable2()
{
_threads = new ConcurrentDictionary<int, ThreadLocalDisposable2>();
}
private ThreadLocalDisposable2(int thread)
{
_thread = thread;
_foo = new MemoryStream();
}
public static ThreadLocalDisposable2 Get()
{
return _threads.GetOrAdd(Thread.CurrentThread.ManagedThreadId, i => new ThreadLocalDisposable2(i));
}
public void Foo()
{
_foo.WriteByte(1);
}
public void Dispose()
{
ThreadLocalDisposable2 thread;
_threads.TryRemove(_thread, out thread);
_foo.Dispose();
}
}
Edit:
Just to clarify what I mean, basically I want all of the behaviour of ThreadLocal but when I call Dispose (on the value, the ThreadLocalDisposable* with underlying Stream in this example, not the static ThreadLocal itself) take that disposed instance out of circulation, i.e. if called upon again -- create a new value as if it's a brand new thread requiring a brand new instance.
The ThreadLocalDisposable1, [1], is sample class of what I think should've worked, except the .Values.Remove(this) line doesn't "take it out of circulation" and forces a new instance to be created for that thread.
The ThreadLocalDisposable2, [2], with ConcurrentDictionary, is a way I implemented alternative to ThreadLocal with "take out of circulation" behaviour I'm after.
Edit:
This is not the a real use case I have, just a general example I can think of, but if you have for example a static ThreadLocal<SqlConnection>, or a socket, and it's forcefully closed (and disposed in final block) -- drop that connection instance and create a new one transparently if called again.
It seems like you're making this much harder than it has to be. Consider this:
public class MyClass: IDisposable
{
private Stream _foo;
public MyClass Get()
{
if (_foo == null)
{
_foo = new MemoryStream();
}
}
public void Foo()
{
_foo.WriteByte(1);
}
public void Dispose()
{
if (_foo != null)
{
_foo.Dispose();
_foo = null;
}
}
}
Now, you can create one of those:
ThreadLocal<MyClass> MyThing = new ThreadLocal<MyClass>();
And you can write:
using (MyThing.Value.Get())
{
// do stuff
}
That seems functionally equivalent to what you're trying to do with your ConcurrentDictionary stuff.
That said, it seems like this is something that would be better managed another way. I don't know your application so I can't say for sure, but it seems like a bad idea to have a stateful object like a Stream or SqlConnection as a global variable. Usually those things are job-specific rather than thread-specific, and as such should be passed as parameters when you start the job.
I've created a class that is not thread-safe and can lead to bad bugs if assumed to be thread-safe. While I work to make my class thread-safe, I'd like to make instances only usable by one thread. Currently my implementation is to check that the current thread is the same as the thread used to construct the instance at every exposure point.
public class NotThreadSafeClass
{
private readonly int _creatorThreadId;
public NotThreadSafeClass()
{
_creatorThreadId = Thread.CurrentThread.ManagedThreadId;
}
public string ExposedProp
{
get
{
AssertSameThread();
return "My Prop";
}
}
public void ExposedMethod()
{
AssertSameThread();
/* Do stuff */
}
private void AssertSameThread()
{
Throw.If(_creatorThreadId != Thread.CurrentThread.ManagedThreadId,
#"NotThreadSafeClass is not thread safe. Please don't use
the same instance of NotThreadSafeClass in multiple threads.");
}
}
Note: Throw.If is defined in http://www.codeducky.org/10-utilities-c-developers-should-know-part-one/
This pattern seems to work, but it's cumbersome and susceptible to bugs if a developer forgets to add this check to any new exposures. Is there a safer and/or more elegant way to ensure that an instance is only used by one thread?
I think that short of using an AOP framework, you will have to "intercept" all such access to your class' methods/properties in your own code, just like you're describing.
I'm thinking Ninject's Interception Extension, or PostSharp
Nothing is built into the language/framework for this.
Cheers
Edit: Moved ThreadLocal<T> to a private field inside the class declaration.
Unless I completely misunderstand, ThreadLocal<T> should meet your needs. An example:
class Foo {
private ThreadLocal<int> _internalState;
public Foo() {
_internalState = new ThreadLocal<int>();
}
public int IntValue {
get { return _internalState.Value; }
set { _internalState.Value = value; }
}
public override string ToString() {
return _internalState.ToString();
}
}
class Program {
public static void Main(string[] args) {
Demonstrate();
}
static void Demonstrate() {
var local = new Foo {IntValue = 5};
Console.WriteLine("Start thread value: {0}", local.IntValue);
new Thread(() => {
local.IntValue += 5;
Console.WriteLine("New thread value: {0}", local.IntValue);
}).Start();
local.IntValue += 10;
Console.WriteLine("Start thread value: {0}", local.IntValue);
}
}
Sample output:
Start thread value: 5
Start thread value: 15
New thread value: 5
I post my understanding of C# lock as follows, please help me validate whether or not I get it right.
public class TestLock
{
private object threadLock = new object();
...
public void PrintOne()
{
lock (threadLock)
{
// SectionOne
}
}
public void PrintTwo()
{
lock (threadLock)
{
// SectionTwo
}
}
...
}
Case I> Thread1 and Thread2 simultaneously try to call PrintOne.
Since PrintOne is guarded by the instance lock, at any time, only
one thread can exclusively enter the SectionOne.
Is this correct?
Case II> Thread1 and Thread2 simultaneously try to call PrintOne and PrintTwo
respectively (i.e. Thread1 calls PrintOne and Thread2 calls PrintTwo)
Since two print methods are guarded by the same instance lock, at any time,
only one thread can exclusively access either SectionOne or SectionTwo, but NOT both.
Is this correct?
1 and 2 are true only if all your threads use the same instance of the class. If they use different instances, then both cases are false
Sample
public class TestLock
{
private object threadLock = new object();
public void PrintOne()
{
lock (threadLock)
{
Console.WriteLine("One");
var f = File.OpenWrite(#"C:\temp\file.txt"); //same static resource
f.Close();
}
}
public void PrintTwo()
{
lock (threadLock)
{
Console.WriteLine("Two");
var f = File.OpenWrite(#"C:\temp\file.txt"); //same static resource
f.Close();
}
}
}
And testing code
static void Main(string[] args)
{
int caseNumber = 100;
var threads = new Thread[caseNumber];
for (int i = 0; i < caseNumber; i++)
{
var t = new Thread(() =>
{
//create new instance
var testLock = new TestLock();
//for this instance we safe
testLock.PrintOne();
testLock.PrintTwo();
});
t.Start();
//once created more than one thread, we are unsafe
}
}
One of the possible solutions is to add a static keyword to the locking object declaration and methods that use it.
private static object threadLock = new object();
UPDATE
Good point made by konrad.kruczynski
..."thread safety" is also assumed from
context. For example, I could take
your file opening code and also
generate exception with static lock -
just taking another application
domain. And therefore propose that OP
should use system-wide Mutex class or
sth like that. Therefore static case
is just inferred as the instance one.
Case I: Check ✓
Case II: Check ✓
Don't forget that locking is only one way of thread synchronization. For other userfull methods, read: Thread Synchronization
Straight from MSDN sample:
public class TestThreading
{
private System.Object lockThis = new System.Object();
public void Process()
{
lock (lockThis)
{
// Access thread-sensitive resources.
}
}
}
Yes and yes. Cases are correct.
Your understanding is 100% correct. So if, for instance, you wanted to allow entry into the two methods separately you would want to have two locks.
Yes, you're correct in both counts.
here are the basics (more or less)
1) use instance locks for instance data
public class InstanceOnlyClass{
private int callCount;
private object lockObject = new object();
public void CallMe()
{
lock(lockObject)
{
callCount++;
}
}
}
2) use static locks for static data
public class StaticOnlyClass{
private int createdObjects;
private static object staticLockObject = new object();
public StaticOnlyClass()
{
lock(staticLockObject)
{
createdObjects++;
}
}
}
3) if you are protecting static and instance data use separate static and instance locks
public class StaticAndInstanceClass{
private int createdObjects;
private static object staticLockObject = new object();
private int callCount;
private object lockObject = new object();
public StaticAndInstanceClass()
{
lock(staticLockObject)
{
createdObjects++;
}
}
public void CallMe()
{
lock(lockObject)
{
callCount++;
}
}
}
based on this your code is fine if you are accessing instance data but unsafe if you are modifying static data
For a "log information for support" type of function I'd like to enumerate and dump active thread information.
I'm well aware of the fact that race conditions can make this information semi-inaccurate, but I'd like to try to get the best possible result, even if it isn't 100% accurate.
I looked at Process.Threads, but it returns ProcessThread objects, I'd like to have a collection of Thread objects, so that I can log their name, and whether they're background threads or not.
Is there such a collection available, even if it is just a snapshot of the active threads when I call it?
ie.
Thread[] activeThreads = ??
Note, to be clear, I am not asking about Process.Threads, this collection gives me a lot, but not all of what I want. I want to know how much time specific named threads in our application is currently using (which means I will have to look at connecting the two types of objects later, but the names is more important than the CPU time to begin with.)
If you're willing to replace your application's Thread creations with another wrapper class, said wrapper class can track the active and inactive Threads for you. Here's a minimal workable shell of such a wrapper:
namespace ThreadTracker
{
using System.Collections.Generic;
using System.Collections.ObjectModel;
using System.Threading;
public class TrackedThread
{
private static readonly IList<Thread> threadList = new List<Thread>();
private readonly Thread thread;
private readonly ParameterizedThreadStart start1;
private readonly ThreadStart start2;
public TrackedThread(ParameterizedThreadStart start)
{
this.start1 = start;
this.thread = new Thread(this.StartThreadParameterized);
lock (threadList)
{
threadList.Add(this.thread);
}
}
public TrackedThread(ThreadStart start)
{
this.start2 = start;
this.thread = new Thread(this.StartThread);
lock (threadList)
{
threadList.Add(this.thread);
}
}
public TrackedThread(ParameterizedThreadStart start, int maxStackSize)
{
this.start1 = start;
this.thread = new Thread(this.StartThreadParameterized, maxStackSize);
lock (threadList)
{
threadList.Add(this.thread);
}
}
public TrackedThread(ThreadStart start, int maxStackSize)
{
this.start2 = start;
this.thread = new Thread(this.StartThread, maxStackSize);
lock (threadList)
{
threadList.Add(this.thread);
}
}
public static int Count
{
get
{
lock (threadList)
{
return threadList.Count;
}
}
}
public static IEnumerable<Thread> ThreadList
{
get
{
lock (threadList)
{
return new ReadOnlyCollection<Thread>(threadList);
}
}
}
// either: (a) expose the thread object itself via a property or,
// (b) expose the other Thread public methods you need to replicate.
// This example uses (a).
public Thread Thread
{
get
{
return this.thread;
}
}
private void StartThreadParameterized(object obj)
{
try
{
this.start1(obj);
}
finally
{
lock (threadList)
{
threadList.Remove(this.thread);
}
}
}
private void StartThread()
{
try
{
this.start2();
}
finally
{
lock (threadList)
{
threadList.Remove(this.thread);
}
}
}
}
}
and a quick test driver of it (note I do not iterate over the list of threads, merely get the count in the list):
namespace ThreadTracker
{
using System;
using System.Threading;
internal static class Program
{
private static void Main()
{
var thread1 = new TrackedThread(DoNothingForFiveSeconds);
var thread2 = new TrackedThread(DoNothingForTenSeconds);
var thread3 = new TrackedThread(DoNothingForSomeTime);
thread1.Thread.Start();
thread2.Thread.Start();
thread3.Thread.Start(15);
while (TrackedThread.Count > 0)
{
Console.WriteLine(TrackedThread.Count);
}
Console.ReadLine();
}
private static void DoNothingForFiveSeconds()
{
Thread.Sleep(5000);
}
private static void DoNothingForTenSeconds()
{
Thread.Sleep(10000);
}
private static void DoNothingForSomeTime(object seconds)
{
Thread.Sleep(1000 * (int)seconds);
}
}
}
Not sure if you can go such a route, but it will accomplish the goal if you're able to incorporate at an early stage of development.
Is it feasible for you to store thread information in a lookup as you create each thread in your application?
As each thread starts, you can get its ID using AppDomain.GetCurrentThreadId(). Later, you can use this to cross reference with the data returned from Process.Threads.