Suggestion: Consider changing worker polling to deadline-based scheduling

[bhamiltoncx]

Each pg-boss worker currently polls the database every pollingIntervalSeconds seconds (defaulting to 2) checking for new work. This is pretty heavy-weight on the CPU (not to mention having to open a DB connection, etc.).

As an alternative for efficiency (and testability), consider having each worker do a single pass at startup to calculate the next deadline at which work is due, then sleep until either:

1. The deadline is reached, or
2. A new job is scheduled, or
3. An existing job is modified / cancelled

For 2) and 3), we can recalculate the next deadline and update the affected worker(s).

I don't know if this might be impacted by database replication, of course, so we'd have to take that into account.

[timgit]

work(), which creates a poller that calls your functions is a convenience abstraction over fetch() and complete(). It's very difficult to make this perfect for all use cases, especially when you have multiple instances distributed and hosted elsewhere.