use a global merkle thread pool to avoid creation and destruction of threads every block #1412
Conversation
|
|
||
| struct merkle_pool { | ||
| merkle_pool() { | ||
| thread_pool.start(4, [](const fc::exception&){}, [](size_t){}); |
There was a problem hiding this comment.
idk what to do with exceptions off hand here. Need to think about it
There was a problem hiding this comment.
Log the exception and std::exit(-2);. Exceptions should be completely unexpected. If they can happen then they should be caught and handled as a bad hash in calculate_merkle_pow2.
There was a problem hiding this comment.
I had to check it worked like I would have guess -- and yeah, when using use_future exceptions thrown inside of the dispatched work do get packaged in to the future. So if you're only post()ing use_futures then it is completely unexpected for anything to leak out.
It might be useful for named_thread_pool to provide some pre-defined functions one can pick up and use. thread_pool.start(4, named_thread_pool::abort_on_exception,...) or something. Will just do something locally here for now. I'm not sure exit() is the right choice since it seems like you'd want something more aborty
There was a problem hiding this comment.
std::abort() does make sense here.
It does seem having a named_thread_pool::abort_on_exception would be handy.
I like this better than the original. I feel that a benefit of having the thread pool in appbase, and our custom |
| abort(); | ||
| }; | ||
| } | ||
|
|
There was a problem hiding this comment.
I wanted to make a make_on_except_quit_appbase() and then replace a number of usages of that across the code but libchain does not depend on appbase so I can't do that here. oh well 😞
Like #1398 but using this approach #1398 (comment)