We'd like to set up monitoring for historical disk usage. It's easy to
see it on the coordinator console and to fetch it with (eg)
http://COORDINATOR/druid/coordinator/v1/servers?simple, but am I
missing something or is it not actually emitted as a monitor anywhere?
I mean I can totally write a tool that polls /servers?simple, but
we've already got monitoring set up...
“jvm/bufferpool/used” should correspond to the size of the segments present on an historical
Huh, really? I would have thought that BufferPool is an in-memory
thing, vs the disk storage of segments...
Segments loaded by historicals are mmaped, which the JVM represents as MappedByteBuffers. Worth noting that I believe this metric will also include the size of DirectByteBuffers used for various queries, these are fairly static but worth checking your own metrics to see.