WHAT

As a consequence of the workaround put in place for MGDSTRM-9006, whenever a kafka instance is provisioned or deprovisioned, the established connections of other kafka instances (including those belonging to other users), will be dropped. The specific part of the workaround that introduces this issue is the use of hard-stop-after 5s to ensure after a haproxy reconfiguration event (such as configuring a new OpenShift Route), that the old haproxy process terminates. If this weren't so, the way that stats are collected from haproxy would mean that we'd fail to account for traffic through the existing connections after the reconfiguration event.

HOW

To solve this either we need a solution like NE-879 (which it is hoped would mean haproxy is reconfigured on the fly without being restarted), or MGDSTRM-9021 which would give us an ingress/egress stats directly from kafka itself.

If we had MGDSTRM-9021, the need to collect statistic from haproxy would go, so hard-stop-after could be made much less aggressive (there'd still be an argument for keeping hard-stop-after to help prevent excessive numbers of haproxy children accumulating in the container).