mvn fabric8:apply / mvn fabric8:recreate never seems to delete any pods now

[jstrachan]

seems to be regression since the new kubernetes client.

Only tested on kubernetes so far. Normally when you run 'mvn fabric8:apply' or 'mvn fabric8:recreate' then after an RC is recreated/updated all the pods are deleted

[jstrachan]

I wonder if its due to changes in helper methods; like finding all the pods for an RC so they can be deleted?

[jimmidyson]

We should add in functionality to scale rc to 0 before deleting rc. Optional I guess but default to scale down to mimic upstream delete IIRC.

[jstrachan]

I guess we'd also need to wait for the scale to zero to take effect before recreating/updating the RC too. But yeah - nice idea!

[jstrachan]

BTW we have the same issue with updating an RC too btw. Typically if you update an RC (e.g. to use a different set of env vars or docker image) we need to scale down the RC before applying the update

[jimmidyson]

We need to ensure compatibility with kubernetes client to get consistent experience but can add as option. Important to think of zero downtime, meaning no scale down, update, scale up - need to think of rolling updates.

[jstrachan]

Yes, using a rolling update is another option we should support; though we probably want a separate maven goal for that as we'd not be recreating / updating a single RC but probably having a new RC for the new version etc

[jimmidyson]

Following the convention in kubectl, update the client should default to deleting cascaded resources (e.g. pods when updating an RC). update is now deprecated btw & called replace - follow fabric8io/kubernetes-client#44 for that change.

So it should be fine to do a client.replace(...) & related resources should be deleted & replaced too.

Rolling update will come along soon too.