stackabletech/hdfs-operator#175 (and others, like Kafka, will have the same problem later on) currently have an init job that reads the `Node' object to get the external IP. This embeds some annoying policy into the product container (such as which address to prefer), and requires the product Pod to have excessive privileges (you need a ClusterRoleBinding to be able to access Node objects).
I'd like to propose that we add a controller to the commons operator that pulls this information out into annotations on the Pod object (which we can then add a projected environment variable that reads).
The existing Pod.status.hostIP field is not sufficient, because:
- This typically prefers the internal IP rather than external
- Depending on cluster policy we'd like to be able to use hostnames here in the future (such as for Kerberos)
stackabletech/hdfs-operator#175 (and others, like Kafka, will have the same problem later on) currently have an init job that reads the `Node' object to get the external IP. This embeds some annoying policy into the product container (such as which address to prefer), and requires the product Pod to have excessive privileges (you need a ClusterRoleBinding to be able to access Node objects).
I'd like to propose that we add a controller to the commons operator that pulls this information out into annotations on the Pod object (which we can then add a projected environment variable that reads).
The existing Pod.status.hostIP field is not sufficient, because: