Right now the scheduler has to wake up periodically (currently every 10th of a second) to call the has_to_stop() function to see if it should shut down. This condition can be set without any change to the pool, which means it bypasses the libev signalling mechanism.

There are a few minor problems with this approach:

• wakes up the thread more frequently than is strictly needed
• makes xstream_join() functions take longer to complete than necessary
• waking the scheduler up periodically may obfuscate bugs that would otherwise manifest as a scheduler hang