Magic Namespace provides an easy, comprehensive option for cluster operators to manage namespaces and observe good security practices in multi-tenant, RBAC-enabled Kubernetes clusters.
So you've got a multi-tenant cluster? Let's assume your cluster is RBAC-enabled. If it isn't, go fix that first. You're playing with fire. Until you fix that, you don't need Magic Namespace. Go fix it. We'll wait...
In a multi-tenant cluster, a cluster operator (someone with full, unrestricted privileges across the entire cluster), will manage users, groups, service accounts, roles, and user/group bindings to roles-- all to either permit or prevent subjects from performing certain actions in different namespaces.
A common paradigm that has emerged is that teams are given their own namespace and some degree of latitude to administer that namespace, whilst not being permitted to perform actions on other teams' namespaces.
Now bring Helm/Tiller into the equation. In an RBAC-enabled cluster, Tiller is
so often granted the
cluster-admin role-- which gives it "root" access to the
entire cluster. While such a Tiller may be suitable for use by a cluster
operator, it's not suitable for use by other teams, as it presents them with
an easy avenue for escalating their privileges.
To compensate for this, a pattern that has emmerged to complement the namespace-per-team pattern is the tiller-per-namespace pattern. This has been widely adopted in multi-tenant, RBAC-enabled clusters. Until now, cluster operators have tended to create their own bespoke scripts for performing all requisite setup to implement these patterns.
Magic Namespace takes the pain out of this setup. It offers cluster operators an easy, comprehensive avenue for using their Tiller to manage namespaces, service accounts, other Tillers, and role bindings for their consituent teams. Magic Namespace permits cluster operators to manage all of this using familiar Helm-based workflows.
How it Works
By default, Magic Namespace creates a service account for Tiller in the
designated namespace and binds it to the
admin role for that namespace. It
also creates a deployment that utilizes this service account. This can be
disabled or configured further, but the default behavior is sensible. In fact,
the defaults closes a variety of known Tiller-based attack vectors.
Magic Namespace also offers cluster operators to define additional service accounts and role bindings for use within the namespace. Typically, it would be a good idea to define at least one role binding that grants a user or group administrative privileges in the namespace. Absent this, the namespace's own Tiller will function, but no user (other than the cluster operator) will be capable of interacting with it via Helm.
- A Kubernetes cluster with RBAC enabled
Installing the Chart
To install the chart to create the
foo namespace (if it doesn't already exist)
and useful resources (Tiller, service accounts, etc.) within that namespace:
$ helm install stable/magic-namespace --name foo --namespace foo
Typically, you will want to bind at least one user or group to the
in this namespace. Here are steps to follow:
First, make a copy of the default
$ helm inspect values stable/magic-namespace > ~/my-values.yaml
~/my-values.yaml accordingly. Here is a sample role binding:
... roleBindings: - name: admin-group-admin role: ## Valid values are "Role" or "ClusterRole" kind: ClusterRole name: admin subject: ## Valid values are "User", "Group", or "ServiceAccount" kind: Group name: <group> ...
Deploy as follows:
$ helm install stable/magic-namespace \ --name foo \ --namespace foo \ --values ~/my-values.yaml
Uninstalling the Chart
Deleting a release of a Magic Namespace will not delete the namespace,
unless you have used the optional
namespace setting. It will
only delete the Tiller, service accounts, role bindings, etc. from that
namespace. This is actually desirable behavior, as anything the team has
deployed within that namespace is likely to be unaffected, though further
deployments to and management of that namespace will not be possible by anyone
other than the cluster operator.
If you have used the
namespace setting, deleting the release will cleanup
all releases deployed with the tiller in the Magic Namespace, along with the
namespace. If other tillers, such as the one in
deployed charts into the Magic Namespace, they will get orphaned when the namespace is
removed, but they can still be removed with the standard
helm delete <name> --purge command.
$ helm delete foo --purge
The following table lists the most common, useful, and interesting configuration
parameters of the Magic Namespace chart and their default values. Please
reference the default
values.yaml to understand further options.
||Whether to include a Tiller in the namespace||
||The number of Tiller replicas to run||
||The Docker image to use for Tiller, minus version/label||
||The specific version/label of the Docker image used for Tiller||
||The pull policy to utilize when pulling Tiller images from a Docker repsoitory||
||The maximum number of releases Tiller should remember. A value of
||Identify the kind of role (
||Identify the name of the
||This deploys a service resource for Tiller. This is not generally needed. Please understand the security implications of this before overriding the default.||
||This prevents Tiller from binding to
||An optional array of names of additional service account to create||
||An optional array of objects that define role bindings||
||Identify the kind of role (
||Identify the name of the role to be used in the role binding|
||Identify the kind of subject (
||Identify the name of the subject to be used in the role binding|
||Specify a namespace to be created and used, overriding the one on the command line|
||Specify annotations to be attached to the namespace|
||Specify labels to be attached to the namespace|