Managers, queues and workers

Which manager runs what, how many can run at once, and WHY A REQUEST CAN SIT PENDING ON AN IDLE SYSTEM. The pending-on-idle case is the one worth understanding first, because it defeats every intuition about capacity. The system is doing nothing, the host is bored, and a request has been queued for twenty minutes. EVERY MANAGER HAS A NUMBER OF PROCESSES, AND THAT NUMBER IS THE REAL CAPACITY LIMIT RATHER THAN THE HOST'S. A manager configured with five processes runs five requests at once and queues the sixth — on a host with plenty of spare capacity, because the constraint is the configuration and not the machine. AND EVERY MANAGER ACCEPTS ONLY CERTAIN REQUESTS, WHICH IS THE OTHER HALF OF THE ANSWER. Specialisation rules — Lesson 3 — decide what each will pick up. So a request can be pending while managers sit idle, because the idle ones will…

In this module: Module 8 · Concurrent processing

1. 1What a concurrent request is
2. 2Managers, queues and workers
3. 3Specialisation rules
4. 4Work shifts
5. 5A request that will never finishFree preview
6. 6Reading a request log
7. 7Killing a request safely
8. 8Managers that will not start
9. 9Output, printing and delivery
10. 10Purging: the table that ate the database
11. 11What breaks: four concurrent processing failures