select is definitely how I would do it, for several key reasons:
- It is architecturally uncomplicated: the process is actually doing only one thing at a time, even though it is unpredictable what it might do next.
- You never have to worry about things being incomplete ... about accidentally sending someone a partial-message that you haven’t finished reading yet.
- You can easily prioritize what it is doing, according to the business requirement.
- Messy timing issues are generally avoided completely.
The program itself is really a dispatcher, connected to everything else with nice, flexible pipes and sockets. It goes to sleep, wakes up with a “honey-do list,” does its chores in whatever sequence it best sees fit, then goes back to sleep again. As long as none of these things take a significant time to actually do, esp. relative to the others, it all works great.