When can i cleanup objects stored in static variables in C#?
I have a static variable that is lazily initialized:
public class Sqm
{
private static Lazy<Sqm> _default = new Lazy<Sqm>();
public static Sqm Default { get { return _default.Value; } }
}
Note: That i've just changed Foo
to be a static
class. It doesn't change the question in any way if Foo
is static or not. But some people are convinced that there is no way that an instance of Sqm
could be constructed without first constructing an instance of Foo
. Even if i did create a Foo
object; even if i created 100 of them, it wouldn't help me solve the problem (of when to "cleanup" a static member).
Sample usage
Foo.Default.TimerStart("SaveQuestion");
//...snip...
Foo.Default.TimerStop("SaveQuestion");
Now, my Sqm
class implements a method that must be called when the object is no longer needed, and needs to clean itself up (save state to filing system, release locks, etc). This method must be called before the garbage collectors runs (i.e. before my object's finalizer is called):
public class Sqm
{
var values = new List<String>();
Boolean shutdown = false;
protected void Cleanup(ICollection stuff)
{
WebRequest http = new HttpWebRequest();
http.Open("POST", "https://stackoverflow.com/SubmitUsageTelemetry");
http.PostBody = stuff;
http.Send();
}
public void Shutdown()
{
if (!alreadyShutdown)
{
Cleanup(values);
alreadyShutdown = true;
}
}
}
When, and where, can i call my Shutdown()
method?
Note: I don't want the developer who uses the Sqm
class to have to worry about calling Shutdown
. That's not his job. In other language environments he would not have to.
The Lazy<T>
class does not seem to call Dispose
on the Value
it lazily owns. So i cannot hook the IDisposable
pattern - and use that as the time to call Shutdown
. I need to call Shutdown
myself.
But when?
It's a static
variable, it exists once for the life of the application/domain/appdomain/apartment.
Yes, the finalizer is the wrong time
Some people do understand, and some people don't, that trying to upload my data during a finalizer
is wrong.
///WRONG: Don't do this!
~Sqm
{
Shutdown(_values); //<-- BAD! _values might already have been finalized by the GC!
}
Why is it wrong? Because values
might not be there anymore. You don't control what objects are finalized in what order. It is entirely possible that values
was finalized before the containing Sqm
.
What about dispose?
The IDisposable
interface, and the Dispose()
method is a convention. There is nothing that dictates that if my object implements a Dispose()
method that it will ever be called. In fact, i could go ahead and implement it:
public class Sqm : IDisposable
{
var values = new List<String>();
Boolean alreadyDiposed = false;
protected void Cleanup(ICollection stuff)
{
WebRequest http = new HttpWebRequest();
http.Open("POST", "https://stackoverflow.com/SubmitUsageTelemetry");
http.PostBody = stuff;
http.Send();
}
public void Dispose()
{
if (!alreadyDiposed)
{
Cleanup(values);
alreadyDiposed = true;
}
}
}
To the person actually reading the question, you might notice that i didn't actually change anything. The only thing i did was to change the name of a method from Shutdown to Dispose. The Dispose pattern is simply a convention. i still have the problem: when can i call Dispose
?
Well you should call dispose from your finalizer
Calling Dispose
from my finalizer is just as incorrect as calling Shutdown
from my finalizer (they are identically wrong):
public class Sqm : IDisposable
{
var values = new List<String>();
Boolean alreadyDiposed = false;
protected void Cleanup(ICollection stuff)
{
WebRequest http = new HttpWebRequest();
http.Open("POST", "https://stackoverflow.com/SubmitUsageTelemetry");
http.PostBody = stuff;
http.Send();
}
public void Dispose()
{
if (!alreadyDiposed)
{
Cleanup(_values); // <--BUG: _values might already have been finalized by the GC!
alreadyDiposed = true;
}
}
~Sqm
{
Dispose();
}
}
Because, again, values
might not be there anymore. For completeness, we can return to the full original correct code:
public class Sqm : IDisposable
{
var values = new List<String>();
Boolean alreadyDiposed = false;
protected void Cleanup(ICollection stuff)
{
WebRequest http = new HttpWebRequest();
http.Open("POST", "https://stackoverflow.com/SubmitUsageTelemetry");
http.PostBody = stuff;
http.Send();
}
protected void Dispose(Boolean itIsSafeToAlsoAccessManagedResources)
{
if (!alreadyDiposed)
{
if (itIsSafeToAlsoAccessManagedResources)
Cleanup(values);
alreadyDiposed = true;
}
}
public void Dispose()
{
this.Dispose(true);
GC.SuppressFinalize(this);
}
~Sqm
{
Dispose(false); //false ==> it is not safe to access values
}
}
i've come full circle. i have an object which i need to "cleanup" before the application domain shuts down. Something inside my object needs to be notified when it can call Cleanup
.
Make the developer call it
No.
i'm migrating existing concepts from another language into C#. If a developer happens to use the global singleton instance:
Foo.Sqm.TimerStart();
then the Sqm
class is lazy initialized. In a (native) application, the reference to the object is held. During (native) application shutdown, the variable holding the interface pointer is set to null
, and the singleton object's destructor
is called, and it can clean itself up.
Nobody should ever have to call anything. Not Cleanup
, not Shutdown
, not Dispose
. Shutdown should happen automatically by the infrastructure.
What is the C# equivalent of i see myself going away, clean myself up?
It's complicated by the fact that if you let the garbage collector collect the object: it's too late. The internal state objects i want to persist are likely already finalized.
It would be easy if from ASP.net
If i could guarantee that my class were being used from ASP.net, i could ask the HostingEnvironment
to notify before the domain shuts down by registering my object with it:
System.Web.Hosting.HostingEnvironment.RegisterObject(this);
And implement the the Stop
method:
public class Sqm : IDisposable, IRegisteredObject
{
var values = new List<String>();
Boolean alreadyDiposed = false;
protected void Cleanup(ICollection stuff)
{
WebRequest http = new HttpWebRequest();
http.Open("POST", "https://stackoverflow.com/SubmitUsageTelemetry");
http.PostBody = stuff;
http.Send();
}
protected void Dispose(Boolean itIsSafeToAlsoAccessManagedResources)
{
if (!alreadyDiposed)
{
if (itIsSafeToAlsoAccessManagedResources)
Cleanup(values);
alreadyDiposed = true;
}
}
public void Dispose()
{
this.Dispose(true);
GC.SuppressFinalize(this);
}
Sqm
{
//Register ourself with the ASP.net hosting environment,
//so we can be notified with the application is shutting down
HostingEnvironment.RegisterObject(this); //asp.net will call Stop() when it's time to cleanup
}
~Sqm
{
Dispose(false); //false ==> it is not safe to access values
}
// IRegisteredObject
protected void Stop(Boolean immediate)
{
if (immediate)
{
//i took too long to shut down; the rug is being pulled out from under me.
//i had my chance. Oh well.
return;
}
Cleanup(); //or Dispose(), both good
}
}
Except my class doesn't know if i'll be called from ASP.net, or from WinForms, or from WPF, or a console application, or shell extension.
Edit: People seem to be confused by what the IDisposable
pattern exists for. Removed references to Dispose
in order to remove the confusion.
Edit 2: People seem to be demanding full, detailed, example code before they will answer the question. Personally i think the question already contains too much example code, as it doesn't serve to help ask the question.
And now that i've added sooo much code, the question has been lost. People refuse to answer a question until the question has been justified. Now that it's been justified, nobody will read it.
It's like diagnostics
It's like the System.Diagnostics.Trace
class. People call it when they want:
Trace.WriteLine("Column sort: {0} ms", sortTimeInMs);
and never have to think of it again.
And then desperation sets in
i was even desperate enough, that i considered hiding my object behind an COM IUnknown
interface, which is reference counted
public class Sqm : IUnknown
{
IUnknown _default = new Lazy<Sqm>();
}
And then hopefully i could trick the CLR into decrementing the reference count on my interface. When my reference count becomes zero, i know everything is shutting down.
The downside of that is that i cannot make it work.
You shouldn't need to call
Dispose
. If the class that implementsIDisposable
uses only managed resources, then those resources will be released as they naturally would when the program completes. If the class uses un-managed resources, then that class should extendCriticalFinalizerObject
and free those resources in its finalizer (as well as in itsDispose
method).In other words, proper use of the
IDisposable
interface doesn't require thatDispose
ever be called. It can be called to release managed or un-managed resources at a particular point in the program, but a leak that occurs because of not calling it should be considered a bug.Edit
In response to your edited question, I think you're looking for the finalizer:
Be aware, though, that there's no guarantee this method will be called. If you absolutely need it called, you should extend
System.Runtime.ConstrainedExecution.CriticalFinalizerObject
.I may still be confused by your question, though. The finalizer is definitely NOT the place to "save my internal values to a file."
There are two problems here:
You are insistent that the
List<string>
might have been finalized.List<string>
doesn't have a finalizer, and it won't have been garbage collected yet (because you have a reference to it). (Those are different operations.) YourSQL
finalizer will still see valid data. So actually, a finalizer might be okay - although by the time the finalizer runs some other resources that you require may have gone away - and the finalizer may not even end up being called. So I think this is simultaneously more feasible than you expect - and a worse idea in general.You are insistent that you don't want to make this deterministic by putting it in the control of the developer, whether that's using
IDisposable
or not. This is simply fighting against what .NET provides. The garbage collector is meant to be for memory resources; any non-memory resources which require deterministic cleanup (including flushing etc) should be cleaned up explicitly. You can use a finalizer as a last "best effort" clean-up, but it should not be used in the way you're trying to use it.There are some approaches that you could use to try to work around this, such as using a "canary" object with a reference to the "real" object: keep a strong reference to the object you're interested elsewhere, and have a finalizer just in the canary object, so that the only thing to be finalized is the canary object - which then triggers the appropriate flushing and removes the last strong reference, leaving the real object eligible for GC - but it's still fundamentally a bad idea, and with static variables in the mix it becomes even worse.
Likewise you can use the
AppDomain.DomainUnload
event - but again, I wouldn't. By the time the domain is being unloaded, I would be worried about the state of the rest of the objects - and it doesn't get called for the default domain.Basically, I think you should change your design. We don't really know the background of the API you're trying to design, but the way you're going at the moment just won't work. I would try to avoid the static variable, personally - at least for anything which is important in terms of timing. There could still be a single object behind the scenes for coordination, but exposing that in your API feels like a mistake to me. However much you protest about other languages and other platforms, if you're working in .NET you need to accept it is what it is. Fighting against the system is not going to help you in the long term.
The earlier you come to the conclusion that you need to change your API design, the more time you have to think about what that new API should look like.
In addition to Ken's answer, the answer to "How can I dispose my object?" is, you can't.
The concept you're looking for is a static deconstructor, or a deconstructor that would run when the static methods are being freed. This does not exist in managed code, and in most (all?) cases shouldn't be necessary. You're more than likely looking at your static methods being unloaded when the executable ends, and the OS will clean up everything at that point.
If you absolutely need to free up resources, and this object must be shared between all active instances, you could create a reference counter and dispose of the object when you're sure all references have been released. I would put heavy consideration into whether this is the correct approach for you first. New instances would need to verify whether your object is
null
, and instantiate it again if so.There's a
ProcessExit
event on the AppDomain you could try to hook, though I don't know much more about it, and it has a default time limit of 2 seconds.Something like this (if it's suitable for you);
i had asked this question four times, if four different ways. Phrasing each one slightly different; trying to solve the problem from a different direction. Eventually it was M.A. Hanin that pointed me at this question that solved the problem.
The issue is that there is no single way to know when the domain is shutting down. The best you can do is try to catch various kinds of events that cover you 100% (rounded to the nearest percentage) of the time.
If the code is in some domain besides the default, then use the
DomainUnload
event. Unfortunately the default AppDomain doesn't raise aDomainUnload
event. So then we catchProcessExit
:This has been tested inside a "web-site" an a "WinForms" application.
The more complete code, showing an implementation of
IDisposable
:.
Sample usage
From a library that does image processing:
From a helper class that can execute a query
For WinForms themed drawing
In a web-site:
In an extension method
The AppDomain Domain Unload Event seems like it would be a good fit for what you're looking for. Since static variables persist until the AppDomain is unloaded, this should give you a hook in right before the variable is destroyed.