Flush metrics on exit

[aabmass]

For #664, in PushController before exit, flush the meter by calling tick().

Metrics are batched in the meter and flushed each tick() of the PushController. The fix is to additionally call tick() before exiting.

[aabmass]

@lzchen I have one concern about this flushing. On exit, the PushController's time interval would be cut short, creating an aggregation over a shorter period. This could accidentally look like an outlier e.g. if you were measuring QPS with a Counter and PushController interval of 1 second, but the flushed measurement was cut at 0.1 seconds.

Many aggregators have a last_update_timestamp that the exporter could forward on to the metric backend to account for this. Does that handle this case? Or maybe the meter should know if it needs to normalize the measurement somehow?

[lzchen]

LGTM. Just a non-blocking comment.

[lzchen]

@aabmass
Yes that is a good concern. It is up to the ingestion service to handle outliers. We provide the timestamp as you've mentioned, and that should be sufficient enough information for whatever backend is analyzing this data. Our responsibility is to be transparent in the behaviour of SDK (it must be clear to users that there is a force flush before exit so they can readily handle these cases).

[codeboten]

LGTM, thanks for the contribution!