Scheduler 01: The Tick Loop

A person glancing at a clock and a list of jobs: anything due now? Do it, then note when it’s next due. It sits inside the project lesson Build Your Own Distributed Scheduler - From Scratch, which lays out the six stages this build climbs.

Before you start. This series builds a small scheduler, a program whose only job is running other programs at the right moment instead of a person watching a clock. Build Your Own Distributed Scheduler - From Scratch lays out the six stages ahead; this first one is the smallest version of that idea - one loop, on one machine, with every job’s plan held in memory (the computer’s fast working space, wiped when the program stops) rather than saved to disk.

The idea in 30 seconds

A scheduler is one loop: what’s due, run it, when’s it next due, sleep - everything else is armour bolted onto that.
Sleeping for a length of time always overshoots a little, and that overshoot piles up (drift) unless you sleep until an exact moment instead (a deadline).
A job that runs longer than the gap until its next run forces a choice - skip it, queue it, or let it overlap - and there is no default right answer.
Apollo 11’s guidance computer hit exactly that choice mid-landing, and had already decided, years earlier, to keep the highest-priority jobs and drop the rest.
Checking every job on the list every time the loop wakes (a scan) is free with a short list in memory and expensive once that list lives in a database.
None of this needs a second machine or a database. It is all true of one loop, on one computer, holding everything in memory.

Read the story · Skip to the numbers