The Karaf health check / readiness check (https://github.com/jboss-fuse/fabric8/blob/3.0.11.redhat-7.x/components/fabric8-karaf/fabric8-karaf-checks/src/main/java/io/fabric8/karaf/checks/internal/ReadinessCheckServlet.java) uses the same servlet container as all of the rest of the http traffic coming into the container.

If there is any kind of longer running request or a slow down or down stream processing delay, the worker queue can start to accumulate messages.

When this happens, health and readiness checks become blocked at the end of queue which leads to probe failures and the container getting killed.