When we implemented the tracking table sweep, we started seeing this behaviour.

I saw one instance where the test failed like so:

[2021/07/08 22:12:26.588] populate: nothing done.
[2021/07/08 22:12:26.588] run: FAILED: runtime_monitor: Cache usage exceeded during test! Limit: 100 usage: 121.183916: error return: -1
[2021/07/08 22:12:26.588] run: process aborting
[2021/07/08 22:12:26.588] FAIL run (exit status: 134)

But the t2 data looks ok from what I can tell. The cache usage is very high but the cache fill ratio calculation is under 100%. And the "page" and "other" metrics add up to the "bytes currently in cache" stat.

I'm not 100% sure whether it's something wrong with the runtime monitor's calculation or whether we're seeing a different value from the stat cursor than what is ending up in FTDC (we've had issues like this before).