If request handlers block (by calling MySQL among other things), then one instance of Tornado per core may not be sufficient to take advantage of a machine's re

If request handlers block (by calling MySQL among other things), then one instance of Tornado per core may not be sufficient to take advantage of a machine's resources. Tornado includes optional code to fork more than one instance where the number of instances defaults to the number of cores. Tornado's forking code has issues: - There's no code to restart dead child processes. - It's not possible to do a rolling upgrade because all instances of the server must be restarted at the same time. I'd stay away from the forking code in Tornado. I suspect that it's not used by FriendFeed because the forking code was added well after the initial release of Tornado.