We don't use phase restarts, as we use `preload_app`: https://github.com/puma/puma/blob/master/README.md#clustered-mode `:puma_phase` values will always be zero.
In some cases (during worker start) it's possible that Puma.stats returns an empty hash for worker's last status. In that case we just skip sampling of the worker until these stats are available.
This sampler gathers Puma-specific metrics which can be used by Prometheus then.