That's good to know and makes sense. I have noticed larger time frames are less prone to the render error. While doing pipeline troubleshooting it is often the most recent time window (~15 minutes) which is most useful to look for changes in throughput. That's why it is more annoying while the larger time frame windows may not be as buggy. It would be nice if the metric assumed a 0 rather than a large value when the partial information causes unreliability. This would have less impact on the graph scale, I'd think?
I'm actually not concerned about the gap. Logstash was simply stopped during this time as this cluster is currently in development.