The data compression job needs to provide some form of back pressure. In its current form it is unstable and easily fails and/or causes other requests to fail. I was running my server with a single Cassandra 3.7 node on my Fedora laptop with this gatling load testing scenario:
and I got errors on the second run of the job. And this is on a decent dev laptop with SSD, 16 GB of RAM, and 8 cores.