You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am not sure this can be fixed in any way, but I have a simulation in which actors do wait_any_for() a lot, and also kill each other brutally a lot. As a result, in wait_any_for() the dynar allocated as
leaks if the actor is killed why it is waiting (the dynar is freed only at the end of the wait_any_for() code).
Again, this seems difficult to fix and after all killing actors brutally may not be good design, but I thought I'd mention it just in case Mt can do some magic ;)
The text was updated successfully, but these errors were encountered:
mquinson
changed the title
Memory Leak in Comm::wait_any_for()
Memory Leak in Comm::wait_any_for() when killing the waiting actor during the wait
Oct 16, 2019
Hello @henricasanova, if you get bored in the future, could you please check whether the issue got fixed in between, as I think? If not, could we maybe close the bug since we fail to reproduce and think it's fixed anyway?
Sorry to have missed this. It's been 3 years now that I've posted this. I think I was able to reproduce it using an old MWE about this, and valgrind confirms that there are no leaks. So I believe it's safe to close. Doing so right now.
I am not sure this can be fixed in any way, but I have a simulation in which actors do wait_any_for() a lot, and also kill each other brutally a lot. As a result, in wait_any_for() the dynar allocated as
xbt_dynar_t comms = xbt_dynar_new(sizeof(simgrid::kernel::activity::ActivityImpl*), [](void* ptr) {
intrusive_ptr_release(*(simgrid::kernel::activity::ActivityImpl**)ptr);
});
leaks if the actor is killed why it is waiting (the dynar is freed only at the end of the wait_any_for() code).
Again, this seems difficult to fix and after all killing actors brutally may not be good design, but I thought I'd mention it just in case Mt can do some magic ;)
The text was updated successfully, but these errors were encountered: