Description
The current PeriodicFlusher helper flushes samples synchronously and it delegates the eventual concurrency to the specific Output. This is something required because the out-of-order sometimes is not accepted (e.g. writing files). This approach has the downside for the flusher to make it very dependent on the output's rate achieved.
Instead, there are cases (e.g InfluxDB), where out-of-order is not relevant so we can add a safe concurrent flush for them and speed up things when few flush requests will get a bad response time.
Suggested implementation
Add a new field to the PeriodicFlusher struct for explicit the required concurrency limit, then spawn in run method the relative amount of goroutines.
A dedicated NewAsyncPeriodicFlusher(limit int) must be implemented to avoid a breaking change for the current API.