std::timed_mutex::try_lock_for fails immediately - c++

Question

I'm using a std::timed_mutex for the first time and it's not behaving the way I expect. It appears to fail immediately instead of waiting for the mutex. I'm providing the lock timeout in milliseconds (as shown here http://www.cplusplus.com/reference/mutex/timed_mutex/try_lock_for/). But the call to try_lock_for() fails right away.
Here's the class that handles locking and unlocking the mutex:
const unsigned int DEFAULT_MUTEX_WAIT_TIME_MS = 5 * 60 * 1000;
class ScopedTimedMutexLock
{
public:
ScopedTimedMutexLock(std::timed_mutex* sourceMutex, unsigned int numWaitMilliseconds=DEFAULT_MUTEX_WAIT_TIME_MS)
m_mutex(sourceMutex)
{
if( !m_mutex->try_lock_for( std::chrono::milliseconds(numWaitMilliseconds) ) )
{
std::string message = "Timeout attempting to acquire mutex lock for ";
message += Conversion::toString(numWaitMilliseconds);
message += "ms";
throw MutexException(message);
}
}
~ScopedTimedMutexLock()
{
m_mutex->unlock();
}
private:
std::timed_mutex* m_mutex;
};
And this is where it's being used:
void CommandService::Process( RequestType& request )
{
unsigned long callTime =
std::chrono::duration_cast< std::chrono::milliseconds >(
std::chrono::system_clock::now().time_since_epoch()
).count();
try
{
ScopedTimedMutexLock lock( m_classMutex, request.getLockWaitTimeMs(DEFAULT_MUTEX_WAIT_TIME_MS) );
// ... command processing code goes here
}
catch( MutexException& mutexException )
{
unsigned long catchTime =
std::chrono::duration_cast< std::chrono::milliseconds >(
std::chrono::system_clock::now().time_since_epoch()
).count();
cout << "The following error occured while attempting to process command"
<< "\n call time: " << callTime
<< "\n catch time: " << catchTime;
cout << mutexException.description();
}
}
Here's the console output:
The following error occured while attempting to process command
call time: 1131268914
catch time: 1131268914
Timeout attempting to acquire mutex lock for 300000ms
Any idea where this is going wrong? Is the conversion to std::chrono::milliseconds correct? How do I make try_lock_for() wait for the lock?
ADDITIONAL INFO: The call to try_lock_for() didn't always fail immediately. Many times the call acquired the lock and everything worked as expected. The failures I was seeing were intermittent. See my answer below for details about why this was failing.

Answer 1

The root cause of the problem is mentioned in the description for try_lock_for() at http://en.cppreference.com/w/cpp/thread/timed_mutex/try_lock_for. Near the end of the description it says:
As with try_lock(), this function is allowed to fail spuriously and
return false even if the mutex was not locked by any other thread at
some point during timeout_duration.
I naively assumed there were only two possible outcomes: (1) the function acquires the lock within the time period, or (2) the function fails after the wait time has elapsed. But there is another possibility, (3) the function fails after a relatively short time for no specified reason. TL;DR, my bad.
I solved the problem by rewriting the ScopedTimedMutexLock constructor to loop on try_lock() until the lock is acquired or the wait time limit is exceeded.
ScopedTimedMutexLock(std::timed_mutex* sourceMutex, unsigned int numWaitMilliseconds=DEFAULT_MUTEX_WAIT_TIME_MS)
m_mutex(sourceMutex)
{
const unsigned SLEEP_TIME_MS = 5;
bool isLocked = false;
unsigned long startMS = now();
while( now() - startMS < numWaitMilliseconds && !isLocked )
{
isLocked = m_sourceMutex->try_lock();
if( !isLocked )
{
std::this_thread::sleep_for(
std::chrono::milliseconds(SLEEP_TIME_MS));
}
}
if( !isLocked )
{
std::string message = "Timeout attempting to acquire mutex lock for ";
message += Conversion::toString(numWaitMilliseconds);
message += "ms";
throw MutexException(message);
}
}
Where now() is defined like this:
private:
unsigned long now() {
return std::chrono::duration_cast< std::chrono::milliseconds >(
std::chrono::system_clock::now().time_since_epoch() ).count();
}

Answer 2

Just a little bump-up for those who come late. And many thanks for help!
Got quite same behavior (only in std::shared_timed_mutex). After some digging found out that both try_lock_for() and try_lock_until() fail immediately if same thread already has exclusive lock on mutex, basically saving time on testing broken code.
Tested with gcc-9, gcc-10, clang-10 and clang-12.
Did NOT tested other possible combinations like requesting exclusive lock over shared lock or requesting shared lock over any of exclusive/shared locks.