Polling versus events for Node background work

SetInterval schedules background work by start time, which can cause jobs to overlap and pile up when a task takes longer than its interval. This leads to a growing overlap until the pool of resources is exhausted. To address this issue, use setTimeout instead, creating a gap between runs instead of a strict schedule. However, polling still has downsides: each idle check costs a query, round trip, and wake-up, which can add up quickly, especially on metered databases or battery-powered systems.

Events, on the other hand, remove the idle cost but add the responsibility of handling delivery guarantees. Missed notifications result in jobs that never run, creating a reputation for occasional failures. The recommended approach is to use both methods, leveraging events for latency and a slow poll as a safety net. Each background worker should start with setInterval to check for tasks, but also include a slow poll for reliability.

Implement safe concurrent execution by using a draining flag to prevent overlapping tasks. This layered approach provides the benefits of both methods, ensuring low latency, minimal idle cost, and reliable job execution without the risk of silent failures.