Unusual memory allocation pattern in RPZ "stress" tests
Sparked by a graph in #4670 that looks between roughly the 3rd and 13th minute less than stellar. Nearly all of the memory should be allocated in the first minute of the test.
This is present in RPZ "stress" tests on BIND 9.18 and 9.19 FreeBSD and Linux. Not on BIND 9.16; authoritative and recursive modes are fine as well. (Flamethrowers on FreeBSD can behave oddly, but this is unlikely to be a case of tooling.)
It needs to be figured out when this started and what's happening between 3. and 13. minute.
This is how it should look like (BIND 9.19.14):