Hi Drew,
Thank you for the response!
>> The Armada API accepts a Tiller hostname as a safeguard in the event that Armada is unable to find a running Tiller pod
Looking at the code [1] it seems to be that Armada will try to find Tiller pod in the event Shipyard doesn't find one.
Was there some problem with Armada API not being able to retrieve Tiller endpoint so it required a fallback? They both seem to use k8s API for that.
>> This can be changed per API request by providing the hostname of a different Tiller pod.
From architecture point of view, do we expect it to work with multiple Tillers within a single Airship installation?
I'm asking these questions to explore the options for resolving GRPC (which is used by Tiller client) [2] constraints, it does not handle CIDR in no_proxy environment variables [3]. One of the options is:
1. [high priority] Expose Tiller to create a DNS name for the pod, and use it when making requests from Armada API to Tiller, in this case Aramada API POD can have ".cluster.local" in NO_PROXY and everything would work just fine (i.e. requests to Tiller won't go through the proxy).
2. [low priority] Reduce coupling between Tiller/k8s and Shipyard by configuring Tiller endpoints in Armada API directly without requiring Shipyard to pass one.
Prior to submitting any changes I would like to verify with you if this makes sense or if I should explore other options.
Thanks,