This looks good to me. I'd just add one or a couple distribution metrics to count time spent blocked, waiting to load the iterable with the values (e.g. time of first iteration, time of all iterations). WDYT?
EDIT. Perhaps also measuring the time it takes to write the PCollection. [ Full content available at: https://github.com/apache/beam/pull/6361 ] This message was relayed via gitbox.apache.org for [email protected]
