Each of us know what is lock and what its purpose is. It is used to block a critical section of code – the statements from the lock block will be executed only by one thread at the same time.
Because there was only one instance of Foo, locks were added to Read and Write operations. They could not add locks in the locations where Read and Write methods because there were different flows and combinations.
Can you spot the problem?
Even if lock block the access to a critical section, this is not the solution for our case. We need to make Read/Write operations atomic and to not allow a Read or Write operation to be executed in the same time.
To resolve this problem we have two simple solutions. The most simple fix, that would work for short time is using a Mutex. It is a synchronization primitive similar with a semaphore that has the size 1.
Other solutions could be used based on transaction or connection pattern (similar with the one that we use in ORM framework.
public class Foo
{
FooConnection con;
public void OpenConnection()
{
...
con.Open();
}
public void CloseConnection()
{
...
con.Close();
}
public void Write(...)
{
...
con.Write(...);
}
public object Read()
{
...
... con.Read();
...
}
}
...
Foo foo = ...;
foo.OpenConnection();
foo.Read();
foo.Read();
...
foo.Read();
foo.RWRWRWR..();
foo.CloseConnection();
Being in a multi-thread environment problems with multiple access to the same resources appeared, information started to be corrupted and many more. For example, multiple read and write started to execute at the same time and having only one connection (cursor) the information retrieved was corrupted.Because there was only one instance of Foo, locks were added to Read and Write operations. They could not add locks in the locations where Read and Write methods because there were different flows and combinations.
public void Write(...)
{
lock(padLock)
{
...
con.Write(...);
}
}
public object Read()
{
lock(padLock)
{
...
... con.Read();
...
}
}
The application started to work better, but they still observed that from time to time they have the same concurrence problems that cause the application to crash.Can you spot the problem?
Even if lock block the access to a critical section, this is not the solution for our case. We need to make Read/Write operations atomic and to not allow a Read or Write operation to be executed in the same time.
To resolve this problem we have two simple solutions. The most simple fix, that would work for short time is using a Mutex. It is a synchronization primitive similar with a semaphore that has the size 1.
public class Foo
{
FooConnection con;
Mutex mutex = new Mutex();
public void OpenConnection()
{
...
con.Open();
}
public void CloseConnection()
{
...
con.Close();
}
public void Write(...)
{
mutex.WaitOne();
...
con.Write(...);
mutex.ReleaseMutex();
}
public object Read()
{
mutex.WaitOne();
...
... con.Read();
...
mutex.ReleaseMutex();
}
}
Using a Mutex, we can ensure that only one thread will execute the code from Read or Write block. We will not have cases when on one thread execute the Read method and the other one executes the Write operation.Other solutions could be used based on transaction or connection pattern (similar with the one that we use in ORM framework.
There is only one way that the Mutex would work, and that is making the application run slower. Because Mutex is an interprocess communication synchronization object, being therefore heavier than the use of a simple lock. So this works as long as you want interprocess communication locking.
ReplyDeleteThe first method should've worked... but it is more likely that the problem is caused by the fact that one of the threads closed the resource too fast or opened the resource too late. First of all, you want to add locking around opening and closing, and second of all, you want to clear that handle when it's closed/not opened.