I have two threads. The first creates a Logic object, detaching a second thread to spin, blocking on OpenSSL socket to receive messages:
struct Logic
{
Logic()
{
std::thread t1(&Logic::run, this);
t1.detach();
}
void run()
{
while(true)
{
// Gets data from SSL (blocking socket)
// Processes data
// Updates timestamp
}
}
uint64_t timestamp;
};
The first thread returns, enters a while loop and continually checks if the detached thread is still running (or whether its blocked permanently).
while(true)
{
Logic logic();
while(true)
{
if(timestamp_not_updated)
{
break; // Break, destroy current Logic object and create another
}
}
}
If the timestamp stops being updated, the inner while loop breaks, causing the Logic object to be destroyed and a new one created.
When this restart behaviour triggers I get a seg fault. thread apply all bt shows 3 threads, not 2. The original detached thread (blocking on OpenSSL) still exists. I thought this would get destroyed due to the object.
How do I stop a detached thread which is blocking/waiting on a resource, so I can restart my class? I need the blocking behaviour because I don't have anything else to do (besides receive the packet) and it's better for performance, than to keep calling in to OpenSSL.
Please see the following code:
std::mutex mutex;
std::condition_variable cv;
std::atomic<bool> terminate;
// Worker thread routine
void work() {
while( !terminate ) {
{
std::unique_lock<std::mutex> lg{ mutex };
cv.wait(lg);
// Do something
}
// Do something
}
}
// This function is called from the main thread
void terminate_worker() {
terminate = true;
cv.notify_all();
worker_thread.join();
}
Is the following scenario can happen?
Worker thread is waiting for signals.
The main thread called terminate_worker();
The main thread set the atomic variable terminate to true, and then signaled to the worker thread.
Worker thread now wakes up, do its job and load from terminate. At this step, the change to terminate made by the main thread is not yet seen, so the worker thread decides to wait for another signal.
Now deadlock occurs...
I wonder this is ever possible. As I understood, std::atomic only guarantees no race condition, but memory order is a different thing. Questions:
Is this possible?
If this is not possible, is this possible if terminate is not an atomic variable but is simply bool? Or atomicity has nothing to do with this?
If this is possible, what should I do?
Thank you.
I don't believe, what you describe is possible, as cv.notify_all() afaik (please correct me if I'm wrong) synchronizes with wait(), so when the worker thread awakes, it will see the change to terminate.
However:
A deadlock can happen the following way:
Worker thread (WT) determines that the terminate flag is still false.
The main thread (MT) sets the terminate flag and calls cv.notify_all().
As no one is curently waiting for the condition variable that notification gets "lost/ignored".
MT calls join and blocks.
WT goes to sleep ( cv.wait()) and blocks too.
Solution:
While you don't have to hold a lock while you call cv.notify, you
have to hold a lock, while you are modifying terminate (even if it is an atomic)
have to make sure, that the check for the condition and the actual call to wait happen while you are holding the same lock.
This is why there is a form of wait that performs this check just before it sends the thread to sleep.
A corrected code (with minimal changes) could look like this:
// Worker thread routine
void work() {
while( !terminate ) {
{
std::unique_lock<std::mutex> lg{ mutex };
if (!terminate) {
cv.wait(lg);
}
// Do something
}
// Do something
}
}
// This function is called from the main thread
void terminate_worker() {
{
std::lock_guard<std::mutex> lg(mutex);
terminate = true;
}
cv.notify_all();
worker_thread.join();
}
In the following example(not all the code included just the necessary portions):
class A
{
public:
void FlushToDisk(char* pData, unsigned int uiSize)
{
char* pTmp = new char[uiSize];
memcpy(pTmp, pData, uiSize);
m_Thread = boost::thread(&CSimSwcFastsimExporter::WriteToDisk, this, pTmp, uiSize);
}
void WriteToDisk(char* pData, unsigned int uiSize)
{
m_Mtx.lock();
m_ExportFile.write(pData, uiSize);
delete[] pData;
m_Mtx.unlock();
}
boost::thread m_Thread;
boost::mutex m_Mtx
}
is it safe to use the m_Thread that way since the FlushToDisk method can be called while the created thread is executing the WriteToDisk method.
Or should I do something like:
m_Thread.join();
m_Thread = boost::thread(&CSimSwcFastsimExporter::WriteToDisk, this, pTmp, uiSize);
Would this second solution be slower than the first?
From what i saw at http://www.boost.org/doc/libs/1_59_0/doc/html/thread/thread_management.html#thread.thread_management.tutorial
"When the boost::thread object that represents a thread of execution is destroyed the thread becomes detached. Once a thread is detached, it will continue executing until the invocation of the function or callable object supplied on construction has completed, or the program is terminated".
So in my case the threads should not be interrupted or?
Thanks in advance.
The second solution will pause the main thread to wait until the writer thread completes. You would be able to remove mutex if you go this way. You are guaranteed to have one file writing thread.
The first solution is going to allow main thread to continue, and will create an uncontrolled writing thread - serialized on the mutex. While you might believe this is better (main thread will not wait) I do not like this solution for several reasons.
First, you do not have any control over the number of created threads. If the function is called often, and the operation is slow, you can easily run out of threads! Second, and much more important, you will accumulate a backlog of detached threads waiting on mutex. If your main application decides to exit, all those threads will be silently killed and the updates will be lost.
The following process function reads data off a queue and processes it. The wait_and_pop function of masterQueue performs a blocking call. Therefore, control does not move ahead until there exists data on the queue that can be read.
class Context
{
void launch()
{
boost::thread thread1(boost::bind(&Context::push,this ) );
boost::thread thread2(boost::bind(&Context::process,this ) );
std::cout<<"Joining Thread1"<<std::endl;
thread1.join();
std::cout<<"Joining Thread2"<<std::endl;
thread2.join();
}
void process()
{
Data data;
while(status)
{
_masterQueue.wait_and_pop(data); //Blocking Call
//Do something with data
}
}
void push()
{
while(status)
{
//Depending on some internal logic, data is generated
_masterQueue.push(data);
}
}
};
status is a boolean(in global scope). This boolean is set to true by default. It is only changed to false when a signal is caught such as SIGINT, SIGSESV etc. In such a case, the while loop is exited and the program can be exited safely.
bool status = true;
void signalHandler(int signum)
{
std::cout<<"SigNum"<<signum;
status = false;
exit(signum);
}
int main()
{
signal(SIGABRT, signalHandler);
signal(SIGINT, signalHandler);
signal(SIGSEGV, signalHandler);
Context context;
context.launch();
}
Since, no new data is pushed by thread2 when a signal is thrown, control in thread1 is stuck at
_masterQueue.wait_and_pop(data);
How do I force this blocking call to be interrupted?
Is it possible to implement this without changing the internal workings of wait_and_pop
Placing a timeout is not an option, since data may arrive on the queue once in a couple of hours or multiple times a second
Do I push a specific type of data on receiving a signal, e.g INT_MAX/INT_MIN, which the process function is coded to recognise and it exits the loop.
Timeout actually is your answer
You break your loop on getting an answer or having an interrupt
You could also spoof things a bit by having the inturrupt push a noop to the queue
You can try to .interrupt() thread when need to finish it.
If .wait_and_pop() uses standard boost mechanism for wait(condition variable or like), it will definitely be interrupted even in blocked state via throwing boost::thread_interrupted exception. If your masterQueue class is reliable wrt exceptions, then such interruption is safe.
Here is a skeleton of my thread class:
class MyThread {
public:
virutal ~MyThread();
// will start thread with svc() as thread entry point
void start() = 0;
// derive class will specialize what the thread should do
virtual void svc() = 0;
};
Somewhere in code I create an instance of MyThread and later I want to destroy it.
In this case MyThread~MyThread() is called. MyThread:svc() is still running and using the object's data members. So I need a way politely inform MyThread:svc() to stop spinning, before proceeding with the destructor.
What is the acceptable way to destroy the thread object?
Note: I'm looking for platform agnostic solution.
UPD: It's clear that the root of problem is that there's no relationship between C++ object representing thread and OS thread. So the question is: in context of object destuction, is there an acceptable way to make thread object behave like an ordinary C++ object or should it be treated as an unusual one (e.g. should we call join() before destoying it?
Considering your additional requirements posted as comment to Checkers' reply (which is the
most straightforward way to do that):
I agree that join in DTor is problematic for various reasons. But from that the lifetime of your thread object is unrelated to the lifetime of the OS thread object.
First, you need to separate the data the thread uses from the thread object itself. They are distinct entities with distinct lifetime requirements.
One approach is to make the data refcounted, and have any thread that wants to access it hold a strong reference to the data. This way, no thread will suddenly grab into the void, but the data will be destroyed as soon as noone touches it anymore.
Second, about the thread object being destroyed when the thread joins:
I am not sure if this is a good idea. The thread object is normally a way to query the state of a thread - but with a thread object that dies as soon as the thread finishes, noone can tell you wether the thread finished.
Generally, I'd completely decouple the lifetime of the thread object from the lifetime of the OS thread: Destroying your thread object should not affect the thread itself. I see two basic approaches to this:
Thread Handle Object - reference counted again, returned by thread creator, can be released as early as one likes without affecting the OS thread. It would expose methods such as Join, IsFinished, and can give access to the thread shared data.
(If the thread object holds relevant execution state, the threafFunc itself could hold a reference to it, thereby ensuring the instance won't be released before the thread ends)
Thin Wrapper - You simply create a temporary around an OS thread handle. You could not hold additional state for the thread easily, but it might be just enough to make it work: At any place, you can turn an OS thread handle into an thread object. The majority of communication - e.g. telling the thread to terminate - would be via the shared data.
For your code example, this means: separate the start() from the svc()
You'd roughly work with this API (XxxxPtr could be e.g. boost::shared_ptr):
class Thread
{
public:
bool IsFinished();
void Join();
bool TryJoin(long timeout);
WorkerPtr GetWorker();
static ThreadPtr Start(WorkerPtr worker); // creates the thread
};
class Worker
{
private:
virtual void Svc() = 0;
friend class Thread; // so thread can run Svc()
}
Worker could contain a ThreadPtr itself, giving you a guarantee that the thread object exists during execution of Svc(). If multiple threads are allowed to work on the same data, this would have to be a thread list. Otherwise, Thread::Start would have to reject Workers that are already associated with a thread.
Motivation: What to do with rogue threads that block?
Assuming a thread fails to terminate within time for one reason or another, even though you told it to. You simply have three choices:
Deadlock, your applicaiton hangs. That usually happens if you join in the destructor.
Violently terminate the thread. That's potentially a violent termination of the app.
Let the thread run to completion on it's own data - you can notify the user, who can safely save & exit. Or you simply let the rogue thread dance on it's own copy of the data (not reference by the main thread anymore) until it completes.
Usually any OS-specific threads API will allow you to "join" a thread. That is, to block indefinitely on a thread handle until the thread functions returns.
So,
Signal the thread function to return (e.g. by setting a flag in its loop to false).
Join the thread, to make sure the actual thread terminates before you try to delete the thread object.
Then you can proceed with destruction of the thread object (you may also join in the destructor, though some people object to blocking destructors.).
I've had a project before with a similar "thread worker" class and a corresponding "work item" class (a-la Java's Thread and Runnable, except thread does not terminate but waits for a new Runnable object to be executed).
In the end, there was no difference if you join in a separate "shutdown" function or in the destructor, except a separate function is a bit more clear.
If you join in a destructor and a thread blocks, you will wait indefinitely.
If you join in a separate function and a thread blocks, you will wait indefinitely.
If you detach the thread and let it finish on its own, it will usually block application from exiting, so you will wait indefinitely.
So there is no straightforward way to make a thread behave like a regular C++ object and ignore its OS thread semantics, unless you can guarantee that your thread code can terminate almost immediately when notified to do so.
You could havee somthing like this in your svc method
while (alive){ //loops}
//free resources after while.
In your destructor, you could set the alive member to false. Or, you could have a pleaseDie() method, that sets the alive member to false, and can be called from the outside requesting the Thread instance to stop processing.
void
Thread::pleaseDie()
{
this->alive = false;
}
You first need a way to communicate with the thread to tell it to shut down. The best mechanism to do this depends on what svc() is doing. If, for example, it is looping on a message queue, you could insert a "please stop" message in that queue. Otherwise, you could simply add a member bool variable (and synchronize access to it) that is periodically checked by the svc(), and set by the thread wanting to destroy the object. Your could add a pure virtual stop() function to your base class, giving the implementor a clear signal that it has to implement svc() to make its class "runnable", and to implement stop() to make it "stoppable".
After asking the thread to stop, you must wait for it to exit before destroying the object. Again, there are several ways to do this. One is to make the stop() function blocking. It could wait, for example, for a "ok, I'm really stopped now" condition variable to be set by the thread running svc(). Alternatively, the caller could "wait" on the thread running svc(). The way to "wait" is platform dependent.
Most thread systems allow you to send a signal to a thead.
Example: pthreads
pthread_kill(pthread_t thread, int sig);
This will send a signall to a thread.
You can use this to kill the thread. Though this can leave a few of the resources hanging in an undefined state.
A solution to the resource problem is to install a signall handler.
So that when the signal handler is called it throws an exception. This will cause the thread stack to unwind to the entry point where you can then get the thread to check a variable about weather it is sill alive.
NOTE: You should never allow an exception to propogate out of a thread (this is so undefined my eyes bleed thinking about it). Basically catch the exception at the thread entry point then check some state variable to see if the thread should really exit.
Meanwhile the thread that sends the signal should wait for the thread to die by doing a join.
The only issues are that when you throw out of signal handler function you need to be careful. You should not use a signal that is asynchronus (ie one that could have been generated by a signal in another thread). A good one to use is SIGSEGV. If this happens normally then you have accessed invalid memory any you thread should think about exiting anyway!
You may also need to specify an extra flag on some systems to cope.
See This article
A working example using pthreads:
#include <pthread.h>
#include <iostream>
extern "C" void* startThread(void*);
extern "C" void shouldIexit(int sig);
class Thread
{
public:
Thread();
virtual ~Thread();
private:
friend void* startThread(void*);
void start();
virtual void run() = 0;
bool running;
pthread_t thread;
};
// I have seen a lot of implementations use a static class method to do this.
// DON'T. It is not portable. This is because the C++ ABI is not defined.
//
// It currently works on several compilers but will break if these compilers
// change the ABI they use. To gurantee this to work you should use a
// function that is declared as extern "C" this guarantees that the ABI is
// correct for the callback. (Note this is true for all C callback functions)
void* startThread(void* data)
{
Thread* thread = reinterpret_cast<Thread*>(data);
thread->start();
}
void shouldIexit(int sig)
{
// You should not use std::cout in signal handler.
// This is for Demo purposes only.
std::cout << "Signal" << std::endl;
signal(sig,shouldIexit);
// The default handler would kill the thread.
// But by returning you can continue your code where you left off.
// Or by throwing you can cause the stack to unwind (if the exception is caught).
// If you do not catch the exception it is implementation defined weather the
// stack is unwound.
throw int(3); // use int for simplicity in demo
}
Thread::Thread()
:running(true)
{
// Note starting the thread in the constructor means that the thread may
// start before the derived classes constructor finishes. This may potentially
// be a problem. It is started here to make the code succinct and the derived
// class used has no constructor so it does not matter.
if (pthread_create(&thread,NULL,startThread,this) != 0)
{
throw int(5); // use int for simplicity in demo.
}
}
Thread::~Thread()
{
void* ignore;
running = false;
pthread_kill(thread,SIGSEGV); // Tell thread it may want to exit.
pthread_join(thread,&ignore); // Wait for it to finish.
// Do NOT leave before thread has exited.
std::cout << "Thread Object Destroyed" << std::endl;
}
void Thread::start()
{
while(running)
{
try
{
this->run();
}
catch(...)
{}
}
std::cout << "Thread exiting" << std::endl;
}
class MyTestThread:public Thread
{
public:
virtual void run()
{
// Unless the signal causes an exception
// this loop will never exit.
while(true)
{
sleep(5);
}
}
};
struct Info
{
Info() {std::cout << "Info" << std::endl;}
~Info() {std::cout << "Done: The thread Should have exited before this" << std::endl;}
};
int main()
{
signal(SIGSEGV,shouldIexit);
Info info;
MyTestThread test;
sleep(4);
std::cout << "Exiting About to Exit" << std::endl;
}
> ./a.exe
Info
Exiting About to Exit
Signal
Thread exiting
Thread Object Destroyed
Done: The thread Should have exited before this
>
You should add dedicated thread management class (i.e. MyThreadMngr), that handles this and other tasks, like book keeping, owning the thread handles etc. The Thread itself should somehow signal to the thread manager that its going to terminate and MyThreadMngr should i.e. have a loop like Tom proposed.
There will probably be more actions that suite into such a thread manager class.
I reckon the easiest way to do this is to wrap the thread execution code in a loop
while(isRunning())
{
... thread implementation ...
}
You can also stop your thread by doing specific calls, for instance when you're using a WIN32 thread you can call TerminateThread on the thread handle in the destructor.
i give a simple and clean design, no signal, no sync, no kill needed.
per your MyThread, i suggest renaming and adding as below:
class MyThread {
public:
virutal ~MyThread();
// will be called when starting a thread,
// could do some initial operations
virtual bool OnStart() = 0;
// will be called when stopping a thread, say calling join().
virtual bool OnStop() = 0;
// derive class will specialize what the thread should do,
// say the thread loop such as
// while (bRunning) {
// do the job.
// }
virtual int OnRun() = 0;
};
the thread interface user will control the lifetime of MyThread.
and actually the real thread object is as below:
class IThread
{
public:
virtual API ~IThread() {}
/* The real destructor. */
virtual void Destroy(void) = 0;
/* Starts this thread, it will call MyThread::OnStart()
* and then call MyThread::OnRun() just after created
* the thread. */
virtual bool Start(void) = 0;
/* Stops a thread. will call MyThread::OnStop(). */
virtual void Stop(void) = 0;
/* If Wait() called, thread won't call MyThread::OnStop().
* If could, it returns the value of MyThread::OnRun()
* returned */
virtual int Wait(void) = 0;
/* your staff */
virtual MyThread * Command(void) = 0;
};
/* The interface to create a thread */
extern IThread * ThrdCreate(MyThread *p);
See the complete interfaces
http://effoaddon.googlecode.com/svn/trunk/devel/effo/codebase/addons/thrd/include/thrd_i.h
Coding Examples
Case 1. Controlled thread loop
class ThreadLoop : public MyThread
{
private:
bool m_bRunning;
public:
virtual bool OnStart() { m_bRunning = true; }
virtual bool OnStop() { m_bRunning = false; }
virtual int OnRun()
{
while (m_bRunning) {
do your job;
}
}
};
int main(int argc, char **argv)
{
ThreadLoop oLoop;
IThread *pThread = ThrdCreate(&oLoop);
// Start the thread, it will call Loop::OnStart()
//and then call Loop::OnRun() internally.
pThread->Start();
do your things here. when it is time to stop the thread, call stop().
// Stop the thread, it will call Loop::OnStop(),
// so Loop::OnRun() will go to the end
pThread->Stop();
// done, destroy the thread
pThread->Destroy();
}
Case 2. Don't know when the thread will stop
class ThreadLoop : public MyThread
{
public:
virtual bool OnStart() { }
virtual bool OnStop() { }
virtual int OnRun()
{
do your job until finish.
}
};
int main(int argc, char **argv)
{
ThreadLoop oLoop;
IThread *pThread = ThrdCreate(&oLoop);
// Start the thread, it will call Loop::OnStart()
//and then call Loop::OnRun() internally.
pThread->Start();
do your things here. Since you don't know when the job will
finish in the thread loop. call wait().
// Wait the thread, it doesn't call Loop::OnStop()
pThread->Wait();
// done, destroy the thread
pThread->Destroy();
}
A complete IThread implementation:
see
http://effoaddon.googlecode.com/svn/trunk/devel/effo/codebase/addons/thrd/src/thrd/thrd.cpp