I’ve been looking into this, and I believe that this issue occurs because of an interaction between the performance schema and thread-pooling. Since PMM polls the server frequently, this leak is more easily seen with PMM. (It’s not a leak exactly since the memory does get freed up when the server exits).
So, here are the relevant options
performance_schema=ON
thread-handling=pool-of-threads
A workaround is to disable the performance schema or the thread pooling.
[url]https://bugs.launchpad.net/percona-server/+bug/1693511[/url]