Hmm, interesting read. Some stuff like this seems to have evolved
organically - I'd say TaskRealtimeMetricsMonitor is an example of this
technique.
It seems like a reasonable pattern when you're doing multiple different
kinds of instrumentation (like, both logs and metrics), if the thing you're
There is a recent discussion of logging (including its code-level
noisiness): https://github.com/apache/incubator-druid/issues/7362
This article describes an interesting pattern that we can probably apply in
some parts of the Druid codebase: