nixos/kubernetes: adding manual section for kubernetes and writing release note for NixOS 19.03

This commit is contained in:
Johan Thomsen 2018-09-17 14:12:44 +02:00 committed by Franz Pletz
parent e2380e79e1
commit 8d62d7972f
No known key found for this signature in database
GPG key ID: 846FDED7792617B4
3 changed files with 169 additions and 0 deletions

View file

@ -23,5 +23,6 @@
<xi:include href="linux-kernel.xml" />
<xi:include href="../generated/modules.xml" xpointer="xpointer(//section[@id='modules']/*)" />
<xi:include href="profiles.xml" />
<xi:include href="kubernetes.xml" />
<!-- Apache; libvirtd virtualisation -->
</part>

View file

@ -0,0 +1,127 @@
<chapter xmlns="http://docbook.org/ns/docbook"
xmlns:xlink="http://www.w3.org/1999/xlink"
xmlns:xi="http://www.w3.org/2001/XInclude"
version="5.0"
xml:id="sec-kubernetes">
<title>Kubernetes</title>
<para>
The NixOS Kubernetes module is a collective term for a handful of
individual submodules implementing the Kubernetes cluster components.
</para>
<para>
There are generally two ways of enabling Kubernetes on NixOS.
One way is to enable and configure cluster components appropriately by hand:
<programlisting>
services.kubernetes = {
apiserver.enable = true;
controllerManager.enable = true;
scheduler.enable = true;
addonManager.enable = true;
proxy.enable = true;
flannel.enable = true;
};
</programlisting>
Another way is to assign cluster roles ("master" and/or "node") to the host.
This enables apiserver, controllerManager, scheduler, addonManager,
kube-proxy and etcd:
<programlisting>
<xref linkend="opt-services.kubernetes.roles"/> = [ "master" ];
</programlisting>
While this will enable the kubelet and kube-proxy only:
<programlisting>
<xref linkend="opt-services.kubernetes.roles"/> = [ "node" ];
</programlisting>
Assigning both the master and node roles is usable if you want a single
node Kubernetes cluster for dev or testing purposes:
<programlisting>
<xref linkend="opt-services.kubernetes.roles"/> = [ "master" "node" ];
</programlisting>
Note: Assigning either role will also default both
<xref linkend="opt-services.kubernetes.flannel.enable"/> and
<xref linkend="opt-services.kubernetes.easyCerts"/> to true.
This sets up flannel as CNI and activates automatic PKI bootstrapping.
</para>
<para>
As of kubernetes 1.10.X it has been deprecated to open
non-tls-enabled ports on kubernetes components. Thus, from NixOS 19.03 all
plain HTTP ports have been disabled by default.
While opening insecure ports is still possible, it is recommended not to
bind these to other interfaces than loopback.
To re-enable the insecure port on the apiserver, see options:
<xref linkend="opt-services.kubernetes.apiserver.insecurePort"/>
and
<xref linkend="opt-services.kubernetes.apiserver.insecureBindAddress"/>
</para>
<note>
<para>
As of NixOS 19.03, it is mandatory to configure:
<xref linkend="opt-services.kubernetes.masterAddress"/>.
The masterAddress must be resolveable and routeable by all cluster nodes.
In single node clusters, this can be set to <literal>localhost</literal>.
</para>
</note>
<para>
Role-based access control (RBAC) authorization mode is enabled by default.
This means that anonymous requests to the apiserver secure port will
expectedly cause a permission denied error. All cluster components must
therefore be configured with x509 certificates for two-way tls communication.
The x509 certificate subject section determines the roles and permissions
granted by the apiserver to perform clusterwide or namespaced operations.
See also:
<link
xlink:href="https://kubernetes.io/docs/reference/access-authn-authz/rbac/">
Using RBAC Authorization</link>.
</para>
<para>
The NixOS kubernetes module provides an option for automatic certificate
bootstrapping and configuration,
<xref linkend="opt-services.kubernetes.easyCerts"/>.
The PKI bootstrapping process involves setting up a certificate authority
(CA) daemon (cfssl) on the kubernetes master node. cfssl generates a CA-cert
for the cluster, and uses the CA-cert for signing subordinate certs issued to
each of the cluster components. Subsequently, the certmgr daemon monitors
active certificates and renews them when needed. For single node Kubernetes
clusters, setting <xref linkend="opt-services.kubernetes.easyCerts"/> = true
is sufficient and no further action is required. For joining extra node
machines to an existing cluster on the other hand, establishing initial trust
is mandatory.
</para>
<para>
To add new nodes to the cluster:
On any (non-master) cluster node where
<xref linkend="opt-services.kubernetes.easyCerts"/> is enabled, the helper
script <literal>nixos-kubernetes-node-join</literal> is available on PATH.
Given a token on stdin, it will copy the token to the kubernetes
secrets directory and restart the certmgr service. As requested
certificates are issued, the script will restart kubernetes cluster
components as needed for them to pick up new keypairs.
</para>
<note>
<para>
Multi-master (HA) clusters are not supported by the easyCerts module.
</para>
</note>
<para>
In order to interact with an RBAC-enabled cluster as an administrator, one
needs to have cluster-admin privileges. By default, when easyCerts is
enabled, a cluster-admin kubeconfig file is generated and linked into
<literal>/etc/kubernetes/cluster-admin.kubeconfig</literal> as determined by
<xref linkend="opt-services.kubernetes.pki.etcClusterAdminKubeconfig"/>.
<literal>export KUBECONFIG=/etc/kubernetes/cluster-admin.kubeconfig</literal>
will make kubectl use this kubeconfig to access and authenticate the cluster.
The cluster-admin kubeconfig references an auto-generated keypair owned by
root. Thus, only root on the kubernetes master may obtain cluster-admin
rights by means of this file.
</para>
</chapter>

View file

@ -54,6 +54,13 @@
</itemizedlist>
<para>to <literal>false</literal> and enable your preferred display manager.</para>
</note>
<para>
A major refactoring of the Kubernetes module has been completed.
Refactorings primarily focus on decoupling components and enhancing
security. Two-way TLS and RBAC has been enabled by default for all
components, which slightly changes the way the module is configured.
See: <xref linkend="sec-kubernetes"/> for details.
</para>
</listitem>
</itemizedlist>
</section>
@ -564,6 +571,40 @@
provisioning.
</para>
</listitem>
<listitem>
<para>
The use of insecure ports on kubernetes has been deprecated.
Thus options:
<varname>services.kubernetes.apiserver.port</varname> and
<varname>services.kubernetes.controllerManager.port</varname>
has been renamed to <varname>.insecurePort</varname>,
and default of both options has changed to 0 (disabled).
</para>
</listitem>
<listitem>
<para>
Note that the default value of
<varname>services.kubernetes.apiserver.bindAddress</varname>
has changed from 127.0.0.1 to 0.0.0.0, allowing the apiserver to be
accessible from outside the master node itself.
If the apiserver insecurePort is enabled,
it is strongly recommended to only bind on the loopback interface. See:
<varname>services.kubernetes.apiserver.insecurebindAddress</varname>.
</para>
</listitem>
<listitem>
<para>
The option <varname>services.kubernetes.apiserver.allowPrivileged</varname>
and <varname>services.kubernetes.kubelet.allowPrivileged</varname> now
defaults to false. Disallowing privileged containers on the cluster.
</para>
</listitem>
<listitem>
<para>
The kubernetes module does no longer add the kubernetes package to
<varname>environment.systemPackages</varname> implicitly.
</para>
</listitem>
</itemizedlist>
</section>
</section>