Go Memcached CSV improvements (discussions)

[camilamacedo86]

This task is for we track the current discussion over how should we dealing with.
This thread started at: https://github.com/operator-framework/operator-sdk-samples/pull/82/files#r341319166

• Should we version all CSV files according to the Memcached released?
• Should we change the name of the channels?

[camilamacedo86]

From @joelanford:

To be totally honest, I think it's a little weird to have CSVs per SDK version since that isn't how a "real" operator would do things. +1 @camilamacedo86

Regards the CSV the suggestion shows be changed the name for the spec.version in the CSV and does as follows.

channels:
- currentCSV: memcached-operator.v0.11.0
  name: v0.11.x
- currentCSV: memcached-operator.v0.10.0
  name: v0.10.x
defaultChannel: v0.11.x

[camilamacedo86]

From @estroz (last questions so far)

@joelanford what you're implying is that we should update the alpha channel with the new CSV version on each release?

An issue crops up here with the CSV's spec.replaces. If OLM can't resolve a CSV specified in spec.replaces, then installation will fail. So each spec.replaces must reference a CSV name contained in the package manifest's channels. That is why I suggested we use versioned channel names. We may have to remove spec.replaces from all CSV versions if we do not want to also expand the number of channels.

[estroz]

In reference to ^: I recently found out about spec.skips and metadata.annotations["olm.skipRange"], both allowing a CSV to specify CSV versions to skip. I recommend using the latter so we can say "ignore CSV's below v0.11.0" with:

metadata:
    annotations:
        olm.skipRange: '<0.11.0'

We can still remove spec.replaces so we can always have one alpha channel for the current CSV in the package manifest. Just figured I'd point out this useful feature.

[camilamacedo86]

Closing since it is sorted out already.