I want to implement an expiration time on a Lazy object. The expiration cooldown must start with the first retrieve of the value. If we get the value, and the expiration time is
I needed the same thing. But I would prefer an implementation without locked reads when there is no write.
public class ExpiringLazy<T>
{
private readonly Func<T> factory;
private readonly TimeSpan lifetime;
private readonly ReaderWriterLockSlim locking = new ReaderWriterLockSlim(LockRecursionPolicy.NoRecursion);
private T value;
private DateTime expiresOn = DateTime.MinValue;
public ExpiringLazy(Func<T> factory, TimeSpan lifetime)
{
this.factory = factory;
this.lifetime = lifetime;
}
public T Value
{
get
{
DateTime now = DateTime.UtcNow;
locking.EnterUpgradeableReadLock();
try
{
if (expiresOn < now)
{
locking.EnterWriteLock();
try
{
if (expiresOn < now)
{
value = factory();
expiresOn = DateTime.UtcNow.Add(lifetime);
}
}
finally
{
locking.ExitWriteLock();
}
}
return value;
}
finally
{
locking.ExitUpgradeableReadLock();
}
}
}
}
I don't think Lazy<T>
would have any influence here, it's more like a general approach, essentially being similar to the singleton pattern.
You'll need a simple wrapper class which will either return the real object or pass all calls to it.
I'd try something like this (out of memory, so might include bugs):
public class Timed<T> where T : new() {
DateTime init;
T obj;
public Timed() {
init = new DateTime(0);
}
public T get() {
if (DateTime.Now - init > max_lifetime) {
obj = new T();
init = DateTime.Now;
}
return obj;
}
}
To use, you'd then just use Timed<MyClass> obj = new Timed<MyClass>();
rather than MyClass obj = new MyClass();
. And actual calls would be obj.get().doSomething()
instead of obj.doSomething()
.
Edit:
Just to note, you won't have to combine an approach similar to mine above with Lazy<T>
because you're essentially forcing a delayed initialization already. You could of course define the maximum lifetime in the constructor for example.
I agree with the other commenters that you probably shouldn't touch Lazy at all. Lazy isn't very complicated if you ignore the multiple thread-safety options, so just implement it from scratch.
I quite like the idea by the way, although I don't know if I'd be comfortable using it as a general purpose caching strategy. It might be sufficient for some of the simpler scenarios.
Here's my stab at it. If you don't need it to be thread-safe you can just remove the locking stuff. I don't think it's possible to use the double-checking lock pattern here because of the chance that the cached value may be be invalidated inside the lock.
public class Temporary<T>
{
private readonly Func<T> factory;
private readonly TimeSpan lifetime;
private readonly object valueLock = new object();
private T value;
private bool hasValue;
private DateTime creationTime;
public Temporary(Func<T> factory, TimeSpan lifetime)
{
this.factory = factory;
this.lifetime = lifetime;
}
public T Value
{
get
{
DateTime now = DateTime.Now;
lock (this.valueLock)
{
if (this.hasValue)
{
if (this.creationTime.Add(this.lifetime) < now)
{
this.hasValue = false;
}
}
if (!this.hasValue)
{
this.value = this.factory();
this.hasValue = true;
// You can also use the existing "now" variable here.
// It depends on when you want the cache time to start
// counting from.
this.creationTime = Datetime.Now;
}
return this.value;
}
}
}
}