Ensure minutely probes do not drift

[tombruijn]

As agreed upon in https://github.com/appsignal/integration-guide/pull/124, make sure that the minutely probes don't drift, so they don't actually calculate metrics over 1 minute and 5 seconds, for example.

TODO

• Research if the probes system can drift
  • Account for probe drift better, it now only schedules after the probe has done sending all the metrics.
• Fix drift if needed

[tombruijn]

The minutely probes thread doesn't drift, as long as all probes are done within a minute. We use the wait_time method to calculate the next 0 second in a minute and run the next iteration over the probes then. This follows the minutely probes integrations guide spec.

Individual probes can drift. They're all run in sequence, if probe 1 is slow, probe 2 runs later. That's fine if probe 1 is always slow, but if it's only intermeditatly slow, probes that run after it could drift a bit back and forth. I don't think we can cover this case unless we run each probe in its own thread.

I'm okay with how it works now.

Things we can do to improve this is to:

• Track if the total of all probes take more than a minute and log an error to the AppSignal log.
• Track how long individual probes take and log a warning if it takes longer than some arbitrary number, like 5 seconds