While using boost::threads I have come across this interruption problem. When I do a boost::thread_interrupt from thread A on thread B, while B has interrupts disabled (boost::this_thread::disable_interrupts di), the interrupt seems to be lost.
That is to say, that if I put a boost::thread::interruption_point() after interruption has been enabled, it does not throw the boost::thread_interrupted exception.
Is this the expected behavior or am I doing something wrong?
Thanks
Nothing in the documentation says that interruptions are re-triggered when thread B re-enables interruptions. I've tried a simple test program and can confirm the behaviour you describe.
After re-enabling interruptions, you can check this_thread::interruption_requested() to see if there was an interruption requested while interruptions were disabled. If an interruption was indeed requested, you can throw a thread_interrupted exception yourself.
Here's a working program that demonstrates this:
#include <boost/thread.hpp>
#include <iostream>
using namespace std;
using namespace boost;
void threadB()
{
int ticks = 0;
this_thread::disable_interruption* disabler = 0;
try
{
while (ticks < 20)
{
if (ticks == 5)
{
cout << "Disabling interruptions\n";
disabler = new this_thread::disable_interruption;
}
if (ticks == 15)
{
cout << "Re-enabling interruptions\n";
delete disabler;
if (this_thread::interruption_requested())
{
cout << "Interrupt requested while disabled\n";
throw thread_interrupted();
}
}
cout << "Tick " << ticks << "\n";
thread::sleep(get_system_time() + posix_time::milliseconds(100));
++ticks;
}
}
catch (thread_interrupted)
{
cout << "Interrupted\n";
}
}
int main()
{
thread b(&threadB);
thread::sleep(get_system_time() + posix_time::milliseconds(1000));
b.interrupt();
cout << "main -> Interrupt!\n";
b.join();
}
Hope this helps.
Related
I picked up the code from the conditional_variable::wait reference. The reference states that :
wait causes the current thread to block until the condition variable
is notified or a spurious wakeup occurs, optionally looping until some
predicate is satisfied.
The link also states that the wait is equivalent to :
while (!pred()) {
wait(lock); }
If that is the case, is there a reason why the following code does not work as the original code?
#include <condition_variable>
#include <iostream>
#include <mutex>
#include <thread>
std::condition_variable cv;
std::mutex cv_m; // This mutex is used for three purposes:
// 1) to synchronize accesses to i
// 2) to synchronize accesses to std::cerr
// 3) for the condition variable cv
int i = 0;
void waits()
{
std::unique_lock<std::mutex> lk(cv_m);
std::cerr << "Waiting... \n";
cv.wait(lk, [] {std::cout << "Check Done!" << std::endl; return i == 1; });
std::cerr << "...finished waiting. i == 1\n";
}
void signals()
{
std::this_thread::sleep_for(std::chrono::seconds(1));
{
std::lock_guard<std::mutex> lk(cv_m);
std::cerr << "Notifying...\n";
}
cv.notify_all();
std::this_thread::sleep_for(std::chrono::seconds(1));
{
std::lock_guard<std::mutex> lk(cv_m);
i = 1;
std::cerr << "Notifying again...\n";
}
//cv.notify_all();
}
int main()
{
std::thread t1(waits), t2(waits), t3(waits), t4(signals);
t1.join();
t2.join();
t3.join();
t4.join();
}
My callback for async_write_some is not called after a one second sleep. If I am starting an io_service worker thread for every write, why is the callback not being called?
header
boost::system::error_code error_1;
boost::shared_ptr <boost::asio::io_service> io_service_1;
boost::shared_ptr <boost::asio::ip::tcp::socket> socket_1;
connect
void eth_socket::open_eth_socket (void)
{
// 1. reset io services
io_service_1.reset();
io_service_1 = boost::make_shared <boost::asio::io_service> ();
// 2. create endpoint
boost::asio::ip::tcp::endpoint remote_endpoint(
boost::asio::ip::address::from_string("10.0.0.3"),
socket_1_port
);
// 3. reset socket
socket_1.reset(new boost::asio::ip::tcp::socket(*io_service_1));
// 4. connect socket
socket_1->async_connect(remote_endpoint,
boost::bind(
ð_socket::socket_1_connect_callback,
this, boost::asio::placeholders::error
)
);
// 5. start io_service_1 run thread after giving it work
boost::thread t(boost::bind(&boost::asio::io_service::run, *&io_service_1));
return;
}
write
void eth_socket::write_data (std::string data)
{
// 1. check socket status
if (!socket_1->is_open())
{
WARNING << "socket_1 is not open";
throw -3;
}
// 2. start asynchronous write
socket_1->async_write_some(
boost::asio::buffer(data.c_str(), data.size()),
boost::bind(
ð_socket::socket_1_write_data_callback,
this, boost::asio::placeholders::error,
boost::asio::placeholders::bytes_transferred
)
);
// 3. start io_service_1 run thread after giving it work
boost::thread t(boost::bind(&boost::asio::io_service::run, *&io_service_1));
return;
}
callback
void eth_socket::socket_1_write_data_callback (const boost::system::error_code& error, size_t bytes_transferred)
{
// 1. check for errors
if (error)
{
ERROR << "error.message() >> " << error.message().c_str();
return;
}
if (socket_1.get() == NULL || !socket_1->is_open())
{
WARNING << "serial_port_1 is not open";
return;
}
INFO << "data written to 10.0.0.3:1337 succeeded; bytes_transferred = " << bytes_transferred;
return;
}
test
open_eth_socket();
write_data("Hello"); // callback called
write_data("Hello"); // callback called
write_data("Hello"); // callback called
sleep(1);
write_data("Hello"); // callback not called after sleep
boost::thread t(boost::bind(&boost::asio::io_service::run, *&io_service_1));
That's weird for a number of reasons.
You should not "run" io_services for each operation. Instead, run them steadily while operations may be posted. Optionally use io_service::work to prevent run from returning.
You should not (have to) create threads for each operation. If anything, it's recipe for synchronization issues (Why do I need strand per connection when using boost::asio?)
When running io_service again after it returned (without error) you should call reset() first, as per documentation (Why must io_service::reset() be called?)
You destruct a non-detached thread - likely before it had completed. If you had used std::thread this would even have caused immediate abnormal program termination. It's bad practice to not-join non-detached threads (and I'd add it's iffy to use detached threads without explicit synchronization on thread termination). See Why is destructor of boost::thread detaching joinable thread instead of calling terminate() as standard suggests?
I'd add to these top-level concerns
the smell from using names like socket_1 (just call it socket_ and instantiate another object with a descriptive name to contain the other socket_). I'm not sure, but the question does raise suspicion these might even be global variables. (I hope that's not the case)
throw-ing raw integers, really?
You are risking full on data-races by destructing io_service while never checking that worker threads had completed.
More Undefined Behaviour here:
_sock.async_write_some(
ba::buffer(data.c_str(), data.size()),
You pass a reference to the parameter data which goes out of scope. When the async operation completes, it will be a dangling reference
There's some obvious copy/paste trouble going on here:
if (socket_1.get() == NULL || !socket_1->is_open())
{
WARNING << "serial_port_1 is not open";
return;
}
I'd actually say this stems from precisely the same source that lead to the variable names being serial_port_1 and socket_1
Some Cleanup
Simplify. There wasn't self-contained code, so nothing complete here, but at least see the many points of simplification:
Live On Coliru
#include <boost/asio.hpp>
#include <boost/thread.hpp>
#include <iostream>
namespace ba = boost::asio;
using ba::ip::tcp;
using boost::system::error_code;
#define ERROR std::cerr
#define WARNING std::cerr
#define INFO std::cerr
struct eth_socket {
~eth_socket() {
_work.reset();
if (_worker.joinable())
_worker.join(); // wait
}
void open(std::string address);
void write_data(std::string data);
private:
void connected(error_code error) {
if (error)
ERROR << "Connect failed: " << error << "\n";
else
INFO << "Connected to " << _sock.remote_endpoint() << "\n";
}
void written(error_code error, size_t bytes_transferred);
private:
ba::io_service _svc;
boost::optional<ba::io_service::work> _work{ _svc };
boost::thread _worker{ [this] { _svc.run(); } };
std::string _data;
unsigned short _port = 6767;
tcp::socket _sock{ _svc };
};
void eth_socket::open(std::string address) {
tcp::endpoint remote_endpoint(ba::ip::address::from_string(address), _port);
_sock.async_connect(remote_endpoint, boost::bind(ð_socket::connected, this, _1));
}
void eth_socket::write_data(std::string data) {
_data = data;
_sock.async_write_some(ba::buffer(_data), boost::bind(ð_socket::written, this, _1, _2));
}
void eth_socket::written(error_code error, size_t bytes_transferred) {
INFO << "data written to " << _sock.remote_endpoint() << " " << error.message() << ";"
<< "bytes_transferred = " << bytes_transferred << "\n";
}
int main() {
{
eth_socket s;
s.open("127.0.0.1");
s.write_data("Hello"); // callback called
s.write_data("Hello"); // callback called
s.write_data("Hello"); // callback called
boost::this_thread::sleep_for(boost::chrono::seconds(1));
s.write_data("Hello"); // callback not called after sleep
} // orderly worker thread join here
}
My problems are now fixed thanks to sehe's help and prayer.
This line in open_eth_socket:
boost::thread t(boost::bind(&boost::asio::io_service::run, *&io_service_1));
is now this:
boost::shared_ptr <boost::thread> io_service_1_thread; // in header
if (io_service_1_thread.get()) io_service_1_thread->interrupt();
io_service_1_thread.reset(new boost::thread (boost::bind(ð_socket::run_io_service_1, this)));
I added this function:
void eth_socket::run_io_service_1 (void)
{
while (true) // work forever
{
boost::asio::io_service::work work(*io_service_1);
io_service_1->run();
io_service_1->reset(); // not sure if this will cause problems yet
INFO << "io_service_1 run complete";
boost::this_thread::sleep (boost::posix_time::milliseconds (100));
}
return;
}
I want to know when dispatchhas finished with some specific work
service.dispatch(&some_work);
I want to know this because I need to restart some_work if it has finished.
struct work
{
std::shared_ptr<asio::io_service> io_service;
bool ready;
std::mutex m;
template <class F>
void do_some_work(F&& f)
{
if (io_service && ready) {
m.lock();
ready = false;
m.unlock();
io_service->dispatch([&f, this]() {
f();
m.lock();
ready = true;
m.unlock();
});
}
}
work(std::shared_ptr<asio::io_service> io_service)
: io_service(io_service)
, ready(true)
{
}
};
int
main()
{
auto service = std::make_shared<asio::io_service>();
auto w = std::make_shared<asio::io_service::work>(*service);
std::thread t1([&] { service->run(); });
work some_work{ service };
for (;;) {
some_work.do_some_work([] {
std::cout << "Start long draw on thread: " << std::this_thread::get_id()
<< std::endl;
std::this_thread::sleep_for(std::chrono::seconds(5));
std::cout << "End long draw on thread: " << std::this_thread::get_id()
<< std::endl;
});
}
w.reset();
t1.join();
}
There are some problems with the code, for example if some_workgoes out of scope, then the running taskwould still write to ready.
I am wondering if something like this already exists in Asio?
For lifetime issues, the common idiom is indeed to use shared pointers, examples:
Ensure no new wait is accepted by boost::deadline_timer unless previous wait is expired
Boost::Asio Async write failed
Other than that, the completion handler is already that event. So you would do:
void my_async_loop() {
auto This = shared_from_this();
socket_.async_read(buffer(m_buffer, ...,
[=,This](error_code ec, size_t transferred) {
if (!ec) {
// do something
my_async_loop();
}
}
);
}
This will re-schedule an (other?) async operation once the previous has completed.
On the subject of threadsafety, see Why do I need strand per connection when using boost::asio?
I'm trying to run an example of websocket++ that consists in receive messages from websocket clients and broadcast to all connected clients, but i having problems with thread synchronization.
In the code example the method process_messages waits for message on a std:queue
boost::unique_lock<boost::mutex> lock(m_action_lock);
while(m_actions.empty()) {
m_action_cond.wait(lock);
}
And the on_message handler locks the queue before to push a new message received from client, but when it try to notify_one(), the program fail with an Segmentation fault 11.
void on_message(connection_hdl hdl, server::message_ptr msg) {
// queue message up for sending by processing thread
{
boost::unique_lock<boost::mutex> lock(m_action_lock);
m_actions.push(action(MESSAGE,msg));
lock.unlock();
}
m_action_cond.notify_one();
}
The only way that the program works is commenting the wait(lock) but i not sure if this is safe.
Some body could help me to find de segmentation fault cause?
The complete code is:
#include <websocketpp/config/asio_no_tls.hpp>
#include <websocketpp/server.hpp>
#include <iostream>
#include <boost/thread.hpp>
#include <boost/thread/mutex.hpp>
#include <boost/thread/condition_variable.hpp>
typedef websocketpp::server<websocketpp::config::asio> server;
using websocketpp::connection_hdl;
using websocketpp::lib::placeholders::_1;
using websocketpp::lib::placeholders::_2;
using websocketpp::lib::bind;
/* on_open insert connection_hdl into channel
* on_close remove connection_hdl from channel
* on_message queue send to all channels
*/
enum action_type {
SUBSCRIBE,
UNSUBSCRIBE,
MESSAGE
};
struct action {
action(action_type t, connection_hdl h) : type(t), hdl(h) {}
action(action_type t, server::message_ptr m) : type(t), msg(m) {}
action_type type;
websocketpp::connection_hdl hdl;
server::message_ptr msg;
};
class broadcast_server {
public:
broadcast_server() {
// Initialize Asio Transport
m_server.init_asio();
// Register handler callbacks
m_server.set_open_handler(bind(&broadcast_server::on_open,this,::_1));
m_server.set_close_handler(bind(&broadcast_server::on_close,this,::_1));
m_server.set_message_handler(bind(&broadcast_server::on_message,this,::_1,::_2));
}
void run(uint16_t port) {
// listen on specified port
m_server.listen(port);
// Start the server accept loop
m_server.start_accept();
// Start the ASIO io_service run loop
try {
m_server.run();
} catch (const std::exception & e) {
std::cout << e.what() << std::endl;
} catch (websocketpp::lib::error_code e) {
std::cout << e.message() << std::endl;
} catch (...) {
std::cout << "other exception" << std::endl;
}
}
void on_open(connection_hdl hdl) {
boost::unique_lock<boost::mutex> lock(m_action_lock);
//std::cout << "on_open" << std::endl;
m_actions.push(action(SUBSCRIBE,hdl));
lock.unlock();
m_action_cond.notify_one();
}
void on_close(connection_hdl hdl) {
boost::unique_lock<boost::mutex> lock(m_action_lock);
//std::cout << "on_close" << std::endl;
m_actions.push(action(UNSUBSCRIBE,hdl));
lock.unlock();
m_action_cond.notify_one();
}
void on_message(connection_hdl hdl, server::message_ptr msg) {
// queue message up for sending by processing thread
boost::unique_lock<boost::mutex> lock(m_action_lock);
//std::cout << "on_message" << std::endl;
m_actions.push(action(MESSAGE,msg));
lock.unlock();
m_action_cond.notify_one();
}
void process_messages() {
while(1) {
boost::unique_lock<boost::mutex> lock(m_action_lock);
while(m_actions.empty()) {
m_action_cond.wait(lock);
}
action a = m_actions.front();
m_actions.pop();
lock.unlock();
if (a.type == SUBSCRIBE) {
boost::unique_lock<boost::mutex> lock(m_connection_lock);
m_connections.insert(a.hdl);
} else if (a.type == UNSUBSCRIBE) {
boost::unique_lock<boost::mutex> lock(m_connection_lock);
m_connections.erase(a.hdl);
} else if (a.type == MESSAGE) {
boost::unique_lock<boost::mutex> lock(m_connection_lock);
con_list::iterator it;
for (it = m_connections.begin(); it != m_connections.end(); ++it) {
m_server.send(*it,a.msg);
}
} else {
// undefined.
}
}
}
private:
typedef std::set<connection_hdl,std::owner_less<connection_hdl>> con_list;
server m_server;
con_list m_connections;
std::queue<action> m_actions;
boost::mutex m_action_lock;
boost::mutex m_connection_lock;
boost::condition_variable m_action_cond;
};
int main() {
broadcast_server server;
// Start a thread to run the processing loop
boost::thread(bind(&broadcast_server::process_messages,&server));
// Run the asio loop with the main thread
server.run(9002);
}
I can reproduce this behavior when Boost is compiled using g++ and libstdc++ but the program linking to it is compiled using clang and libc++. The libstdc++ and libc++ standard libraries are not ABI compatible, so you will need to build everything with one or everything with the other.
Details on how to compile Boost in C++11 mode with clang/libc++:
How to compile/link Boost with clang++/libc++?
After porting a code segment from Windows to Mac OS X, I found it to consume a whole CPU core while running; the responsible call for the CPU consumption is boost::interprocess::interprocess_semaphore::timed_wait.
Here follows the code portion which reproduces this behaviour.
#include <boost/interprocess/sync/interprocess_semaphore.hpp>
#include <boost/interprocess/shared_memory_object.hpp>
#include <boost/interprocess/mapped_region.hpp>
#include <boost/thread/thread_time.hpp>
#include <iostream>
static bool gStopRequested(false);
struct ShmObj
{
boost::interprocess::interprocess_semaphore mSemaphore;
ShmObj() : mSemaphore(0) {};
~ShmObj() {};
};
int main(char* argc, const char** argv)
{
boost::interprocess::shared_memory_object* lShmObj = NULL;
std::string lShmObjName("My_Boost_Interprocess_Test");
boost::interprocess::mapped_region* lRegion;
ShmObj* lObj;
//Create shared segment
try
{
lShmObj = new boost::interprocess::shared_memory_object(boost::interprocess::create_only, lShmObjName.c_str(), boost::interprocess::read_write);
}
catch (boost::interprocess::interprocess_exception &ex)
{
if (ex.get_error_code() != boost::interprocess::already_exists_error)
{
std::cerr << "Some error" << std::endl;
exit(1);
}
else
{
std::cerr << "Already exists, just taking it back." << std::endl;
try
{
lShmObj = new boost::interprocess::shared_memory_object(boost::interprocess::open_only, lShmObjName.c_str(), boost::interprocess::read_write);
}
catch (boost::interprocess::interprocess_exception &ex2)
{
std::cerr << "D'oh !" << std::endl;
exit(1);
}
}
}
if (!lShmObj)
{
exit(1);
}
lShmObj->truncate(sizeof(ShmObj));
lRegion = new boost::interprocess::mapped_region(*lShmObj, boost::interprocess::read_write);
lObj = new (lRegion->get_address()) ShmObj;
// The loop
while (!gStopRequested)
{
boost::system_time lDeadlineAbsoluteTime = boost::get_system_time() + boost::posix_time::milliseconds(500);
if (lObj->mSemaphore.timed_wait(lDeadlineAbsoluteTime))
{
std::cout << "acquired !" << std::endl;
}
else
{
std::cout << "tick" << std::endl;
}
}
}
Then, I read that unnamed semaphores were not available under Mac OS X, so I thought it could be because unnamed semaphores were not efficiently emulated... I then tried the following, unsucessfully:
#include <boost/interprocess/sync/named_semaphore.hpp>
#include <boost/thread/thread_time.hpp>
#include <iostream>
static bool gStopRequested(false);
int main(char* argc, const char** argv)
{
boost::interprocess::named_semaphore::remove("My_Boost_Interprocess_Test");
boost::interprocess::named_semaphore lMySemaphore(boost::interprocess::open_or_create, "My_Boost_Interprocess_Test", 1);
// The loop
while (!gStopRequested)
{
boost::system_time lDeadlineAbsoluteTime = boost::get_system_time() + boost::posix_time::milliseconds(500);
if (lMySemaphore.timed_wait(lDeadlineAbsoluteTime))
{
std::cout << "acquired !" << std::endl;
}
else
{
std::cout << "tick" << std::endl;
}
}
}
I was actually expecting a better behaviour of boost::interprocess on Mac OS X because of the available Posix primitives, but it is actually not. Any idea for a resolution? Thanks a lot.
I successfully Used Mach semaphores instead of the ones of boost::interprocess... see http://pkaudio.blogspot.com/2010/05/mac-os-x-no-timed-semaphore-waits.html