Check in scheduler

[unflxw]

Part of appsignal/integration-guide#169.

Implement check-in scheduler

Implement a check-in event scheduler, which schedules check-in events
to be transmitted in a separate thread, with a minimum wait period
of ten seconds between requests, and a wait period of a tenth of a
second before the first request, referred to as "debounce" periods.

This is a relatively minor improvement for the existing cron
check-ins, but it is a requirement for the heartbeat check-ins, both
to avoid slowing down customers' applications with blocking requests,
and to avoid misuse of the feature from spamming our servers.

The scheduler also acts as a deduplicator, removing "similar enough"
check-in events -- again, not particularly interesting for cron
check-ins, but a requirement to minimise damage when heartbeat
check-ins are misused.

Implement utility functions for serialising and deserialising NDJSON
payloads.

When an event is scheduled, the scheduler sets a timeout for the
duration of the timeout period, after which the events stored in
the scheduler are transmitted.

When Appsignal.stop is called and awaited, the scheduler is
gracefully stopped, awaiting the transmission of all scheduled
check-in events. During this time, new check-in events cannot be
scheduled.

Avoid probes interval blocking shutdown

The probes instance starts an interval timer on initialisation.
By default, timers cause the Node.js engine to await their
completion on shutdown -- which, this being an interval timer, is
never actually completed. (See #418)

We fixed this in the past by implementing Appsignal.stop, which
clears the interval. But a preferable solution is to mark the
probes' interval as .unref(), telling the Node.js engine that it
does not need to be awaited.

[backlog-helper]

• This Pull Request needs more reviews. @jeffkreeftmeijer @luismiramirez - (More info)

---

This is a message from the daily scheduled checks.

New issue guide | Backlog management | Rules | Feedback