should we add a 'keep running if you have a failure mode' option for `helm install`?

[jstrachan]

trafficstars

Right now helm install terminates if any kubectl create fails.

As we start to make composite apps with different RCs and Services that can overlap. New versions of a chart may add a new Service or RC for example.

Users might want a way to try to create all the kubernetes resources inside a chart; then summarise all the failures at the end; rather than failing fast. e.g. so that any resources that don't exist will be created. e.g. if you did

helm install cheese
# time passes a new RC called 'bar' is added to new version of cheese
helm update
helm install cheese

The second install fails with duplicates for all the things in the first version of cheese which have already been installed; but at least the 2nd install command will install 'bar'

Another option could be to try check if each resource exists already before doing the kubectl create so that we can avoid trying to create something that already exists and failing fast?

[arschles]

I like the last option you mentioned. having helm install fail fast, and providing a helm upgrade (as discussed in #292) makes the most sense to me, and follows what homebrew and yum (iirc, it's been a while) both do.

[technosophos]

Good idea. We already have the force flag. Should this behavior be rolled into that, or should we create a separate flag?

[jstrachan]

@arschles if you want to play, there's an implementation of helm upgrade that works as you describe in this PR: https://github.com/helm/helm/pull/323