Performance improvements for checking schedule intervals

[drcapulet]

When working with large schedules of jobs (~100) that run infrequently (weekly or less) our Sidekiq server can take 8 to 10 minutes to start due to the check to ensure jobs don't run more frequently than the scheduler.

For Cron jobs, Fugit supports up to 1 second cron resolution, the entire check is skipped if the scheduler frequency is below that. Then based on a simple heuristic if the minimum delta between the 2nd, 3rd, and 4th runtimes is greater than a week the brute_frequency method is used. Otherwise the old enumeration logic over next_time_from is used.

I added simple specs to ensure adding both frequent and infrequent cron additions take less than a second (on master the infrequent spec takes 12 seconds).

This reduces the schedule parsing time for us to just under 20s.

[jmettraux]

Hello,

this looks excellent. Give me time to fully wake up and integrate that.

Thanks!

[RyanBalfanz]

I approve of this change.

[jmettraux]

May I get a Square sticker?

[jmettraux]

@drcapulet What if I introduce a scheduler option to tell it not to check the frequency at all (trusting you the user to feed it with schedules that match)?

[jmettraux]

@drcapulet

Is using Fugit::Cron#rough_frequency instead of your logic OK for you? It should cut down the #check_frequency time even more.

Thanks in advance.

[drcapulet]

@jmettraux Ah yeah - looks great, thanks!