Github user simonellistonball commented on the issue:
https://github.com/apache/metron/pull/867
Should the size limit on the sample really be a cut off? In a likely usage
scenario a users would sample over a window in a profile. Limiting the size is
likely to skew to time at the beginning of the window rather than being
genuinely uniform. Would a random replacement strategy make more sense when
over the limit? This could be a lot heavier in terms of performance, but may be
more mathematically sound.
---
