We setup two new clusters with replication and we were seeing really bad performance when reading from backlog.
We then discovered some bookkeepers were at 100% CPU usage but apparently doing nothing, almost no Disk or Net IO.
I then proceeded to downgrading one of the bookkeeper to 1.14 and it seems to be working just fine.

Any ideas of what could be going on?

System configuration

Pulsar version: 1.15.2