One of the biggest challenges of thread-management begins when the threads share non-const data
Data race and critical section
In the context of threads using shared data, you often hear the expressions race condition and critical section. But, what's that?
- Data Race
- A data race is a state, in which at least two threads access a shared data at the same time, and at least one of the threads is a writer.
- Critical Section
- A critical section is a section of the code, which not more than one thread should access at any point in time.
In case the program has a race condition, the program behaviour is undefined. To say it differently, anything can happen.
A nice way to visualize a race condition is to let a few threads write to std::cout. std::cout is the shared object (output stream), that should be protected from simultaneous access by multiple threads.
The boss assigns three work packages (lines 11 - 17) to each of its six workers (lines 32 - 36). When a worker is done with its work package it screams out loudly to the boss (line 16). When the boss has gotten notifications from all workers, it sends them home (line 45).
What a mess!
The same mess the next day. The workers scream out loudly. Totally unsynchronized.
The first solution is a mutex. A mutex ensures, that each thread exclusively accesses the shared variable std::cout.
A side note: std::cout is thread safe
- The C++11 standard guarantees, that you must not protect the single characters, written to std::cout. Each character will atomically be written. Of course, it is possible, that more output statements like in the example will interleave. But that is only an optical issue. The program is well defined. The remark is valid for all input and output streams.
Mutex stands for mutual exclusion. It ensures, that only one thread can access a critical section.
The key difference to the first example are the lines 19 to 21. By invoking the methods coutMutex.lock() and coutMutex.unlock(), you define the exclusive section. This section can only be accessed by at most a single thread. The access to std::cout is synchronized and the mess becomes harmony.
Mutexes have a lot of issues, which I will discuss in the next post. (Proofreader Alexey Elymanov)
Go to Leanpub/cpplibrary "What every professional C++ programmer should know about the C++ standard library". Get your e-book. Support my blog.