Dara Hayes David Martin Wojciech Trocki I've been working on the side-car approach and here are some of my thoughts on it. The code I've been using for the side-car is ( https://github.com/StevenTobin/k8s-grafana-watcher ) a slightly modified fork of a project called k8s-grafana-watcher.

• Using a sidecar will mean we have to have a container calling out into the cluster using kubrenetes. To enable this, the side-car needs to run with a serviceaccount with permissions to get configmaps.
• We'll have to support and test this side-car and the code in it. The k8s-grafana-watcher is only rough proof of concept at the moment so we will most likely have to build our own version of it to accomodate testing and the robustness we would want.
• k8s-grafana-watcher does not currently support renaming or removing datasources or dashboards.
• This is functionality which really should be in grafana and the side-car approach feels like a bit of a hack.