title | menu_order |
---|---|
Customising the deployment |
20 |
The deployment installs Flux and its dependencies. First, change to the directory with the examples configuration.
First, you need to connect Flux to the repository with Kubernetes
manifests. This is achieved by setting the --git-url
and
--git-branch
arguments in the
flux-deployment.yaml
manifest.
Flux uses memcache to cache docker registry requests.
kubectl create -f memcache-dep.yaml -f memcache-svc.yaml
You will need to create a secret in which Flux will store its SSH key. The daemon won't start without this present.
The flux
logs should show that it has now connected to the
repository and synchronised the cluster.
When using Kubernetes, this key is stored as a Kubernetes secret. You
can restart flux
and it will continue to use the same key.
Flux connects to the repository using an SSH key.
The SSH key must be configured to have R/W access to the repository.
More specifically, in the case of standalone Flux, the ssh key must be able to
create and update tags. E.g. in Gitlab, that means it requires Maintainer
permissions. The Developer
permission can create tags, but not update them.
You have two options:
If you don't specify a key to use, Flux will create one for you. Obtain the public key through fluxctl:
Create a Kubernetes Secret from a private key:
kubectl create secret generic flux-git-deploy --from-file=identity=/path/to/private_key
The Kubernetes deployment configuration file
flux-deployment.yaml runs the
Flux daemon, but you'll need to edit it first, at least to supply your
own configuration repo (the --git-repo
argument).
$EDITOR flux-deployment.yaml
kubectl create -f flux-deployment.yaml
You will need to provide fluxd with a service account which can access
the namespaces you want to use Flux with. To do this, consult the
example service account given in
flux-account.yaml (which
puts essentially no constraints on the account) and the
RBAC documentation,
and create a service account in whichever namespace you put fluxd
in. You may need to alter the namespace: default
lines, if you adapt
the example.
Using an SSH key allows you to maintain control of the repository. You
can revoke permission for flux
to access the repository at any time
by removing the deploy key.
If you're using your own git host -- e.g., your own installation of
gitlab, or bitbucket server -- you will need to add its host key to
~/.ssh/known_hosts
in the flux daemon container.
First, run a check that you can clone the repo. The following assumes
that your git server's hostname (e.g., githost
) is in $GITHOST
and
the URL you'll use to access the repository (e.g.,
user@githost:path/to/repo
) is in $GITREPO
.
$ # Find the fluxd daemon pod:
$ kubectl get pods --all-namespaces -l name=flux
NAMESPACE NAME READY STATUS RESTARTS AGE
weave flux-85cdc6cdfc-n2tgf 1/1 Running 0 1h
$ kubectl exec -n weave flux-85cdc6cdfc-n2tgf -ti -- \
env GITHOST="$GITHOST" GITREPO="$GITREPO" PS1="container$ " /bin/sh
container$ git clone $GITREPO
Cloning into <repository name>...
No ECDSA host key is known for <GITHOST> and you have requested strict checking.
Host key verification failed.
fatal: Could not read from remote repository
container$ # ^ that was expected. Now we'll try with a modified known_hosts
container$ ssh-keyscan $GITHOST >> ~/.ssh/known_hosts
container$ git clone $GITREPO
Cloning into '...'
...
If git clone
doesn't succeed, you'll need to check that the SSH key
has been installed properly first, then come back. ssh -vv $GITHOST
from within the container may help debug it.
If it did work, you will need to make it a more permanent arrangement. Back in that shell, create a configmap for the cluster. To make sure the configmap is created in the namespace of the Flux or weave deployment, the namespace is set explicitly:
container$ kubectl create configmap flux-ssh-config --from-file=$HOME/.ssh/known_hosts -n $(cat /var/run/secrets/kubernetes.io/serviceaccount/namespace)
configmap "flux-ssh-config" created
To use the ConfigMap every time the Flux daemon restarts, you'll need to mount it into the container. The example deployment manifest includes an example of doing this, commented out. Uncomment those two blocks:
- name: ssh-config
configMap:
name: flux-ssh-config
- name: ssh-config
mountPath: /root/.ssh
It assumes you used flux-ssh-config
as name of the ConfigMap and then reapply the
manifest.
Another alternative is to create the configmap from a template. This could be something like:
apiVersion: v1
data:
known_hosts: |
# github
192.30.253.112 ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==
# github
192.30.253.113 ssh-rsa AAAAB3NzaC1yc2EAAAABIwAAAQEAq2A7hRGmdnm9tUDbO9IDSwBK6TbQa+PXYPCPy6rbTrTtw7PHkccKrpp0yVhp5HdEIcKr6pLlVDBfOLX9QUsyCOV0wzfjIJNlGEYsdlLJizHhbn2mUjvSAHQqZETYP81eFzLQNnPHt4EVVUh7VfDESU84KezmD5QlWpXLmvU31/yMf+Se8xhHTvKSCZIFImWwoG6mbUoWf9nzpIoaSjB+weqqUUmpaaasXVal72J+UX2B+2RPW3RcT0eOzQgqlJL3RKrTJvdsjE3JEAvGq3lGHSZXy28G3skua2SmVi/w4yCE6gbODqnTWlg7+wC604ydGXA8VJiS5ap43JXiUFFAaQ==
# private gitlab
gitlab.________ ssh-rsa AAAAB3N...
kind: ConfigMap
metadata:
name: flux-ssh-config
namespace: <OPTIONAL NAMESPACE (if not default)>
You will need to explicitly tell fluxd to use that service account by
uncommenting and possible adapting the line # serviceAccountName: flux
in the file fluxd-deployment.yaml
before applying it.