Skip adding stable/incubator on init flag

[mattfarina]

In Helm v2 when helm init is run it adds the stable repo and an update happens. Even if you specify the tiller image. This is a problem for air gaped environments. They can specify a local tiller image but still have issues due to the upstream charts repo.

Since we are extending the Helm v2 support window it may be worth adding a flag to init to skip trying to add the stable repo.

Note, this was captured in the CNCF Air Gapped Working Group.

[bacongobbler]

Seems reasonable. There's helm init --skip-refresh which will skip refreshing the repository cache file, but I don't believe there is a feature flag to skip adding the stable repository entirely.

[technosophos]

I think that this is a necessary change in order to allow Helm2 users to continue using Helm when the Stable and Incubator repositories go away.

[hickeyma]

Sounds reasonable to me.

[rimusz]

why not just to remove of adding stable/incubator repos in helm v2 with init?
then document it as we have it for the helm v3?

[bacongobbler]

That would be a breaking change.

[rimusz]

the change will be needed sooner or later anyway :)

[bacongobbler]

Helm 2's support window will be closed by the time the stable chart repository is shut down, @rimusz. It will not be necessary to remove the stable repository from helm init during Helm 2's lifetime. That's why it'd be best as a feature flag so users can continue to use Helm 2 after the support window has closed.

Now normally, we have mentioned that we would not accept new features for the Helm 2 branch. However, this does seems like a necessary feature we need to get in before the stable repository is shut down. I'm less concerned about providing this to help support the air-gapped scenario given our current support window for Helm 2, but it certainly would help out in that regard.

Have there been any suggestions on a flag name from the Air Gapped WG? Perhaps helm init --skip-repos?

[rimusz]

then flag with a name helm init --skip-repos is needed for sure.

[mattfarina]

@bacongobbler helm init --skip-repos works for me. I can start dev on this.

[rjbez17]

For those waiting for this, this is the hack we've used internally to work around the issue. Run before helm init, then you can do a helm repo remove stable after the init.

# Make helm work without internet
# We need to make a dummy repository so helm init doesn't try to add
# an existing valid stable repo.
mkdir -p $HOME/.helm/repository/cache
mkdir -p $HOME/.helm/repository/local
mkdir -p $HOME/.helm/cache/archive

DATE=$(date +%F)
cat <<EOF > $HOME/.helm/repository/repositories.yaml
apiVersion: v1
generated: ${DATE}T12:00:00.111111-06:00
repositories:
- caFile: ""
  cache: /root/.helm/repository/cache/stable-index.yaml
  certFile: ""
  keyFile: ""
  name: stable
  password: ""
  url: localhost
  username: ""
EOF
# End hack

[bacongobbler]

@rjbez17 I think helm init --client-only --skip-refresh && helm repo remove stable will accomplish the same task, if I'm not mistaken.

[rjbez17]

@bacongobbler could you then follow that up with a helm init to deploy tiller without issue?

[bacongobbler]

Yep, though if you want to deploy tiller, you can also just omit the --client-only flag from the above example. --client-only prevents Helm from installing tiller in Kubernetes.

So, there's two ways you can do it:

helm init --skip-refresh --client-only
helm repo remove stable
helm init --skip-refresh

Or

helm init --skip-refresh
helm repo remove stable

Both accomplish the same thing.

[mattfarina]

@rjbez17 there is a flag to specify a different alternative location for Tiller, too.

[bridgetkromhout]

@mattfarina's PR in #8344 said "closes" and it's merged, so as far as I can tell, we can probably close this issue.