Github user kkhatua commented on the issue:
https://github.com/apache/drill/pull/1160
I think I need to put up some tooltips.
We're getting the metrics from the `drill.allocator.root.[peak|used]`
values. I haven't seen a metric that shows the actual Direct allocation from
the system. What Drill seems to report is the current **actively** used Direct
memory. So, when running queries on the cluster, these numbers move up. The
percentage for Direct is the current usage as a percent of the peak memory that
was reported by the root allocator.
Check out any active Drillbit's `#metrics` page. The `drill.allocator.*`
entries are what I was referring to. JVM itself doesn't appear to expose the
Direct memory information readily.
---
