Comment by adrienjarthon

In my case (updown.io), I use a "scheduler" process (one per machine) which runs the following loop: every 5 seconds it fetches the checks that needs to execute in the next 5 seconds, ordered by time of run ascending. Then it iterates through them one by one, wait with a small sleep until the precise running time, and then push the job into a background job queue (sidekiq / redis in my case). This allows for a very precise timing of execution even if there is hundreds jobs per seconds, and a good distribution over time (instead of firing 100 jobs every seconds in spikes, it can schedule one every 10ms for example)