You are browsing a read-only backup copy of Wikitech. The live site can be found at wikitech.wikimedia.org

Kubernetes/Clusters/New

From Wikitech-static
Jump to navigation Jump to search

Intro

This is a guide for setting up or reinitializing a new cluster from scratch or almost scratch, using all the already present wikimedia infrastructure. A quick primer:

A vanilla kubernetes is made up of the following components:

  • Control plane
    • etcd
    • kube-apiserver
    • kube-controller-manager
    • kube-scheduler
  • Node
    • kube-proxy
    • kubelet

Note that upstream documents also refer to another control-plane component, namely cloud-controller-manager. We don't run cloud-controller-manager as we are not in a cloud.

In our infrastructure the first 3 components (kube-apiserver, kube-controller-manager, kube-scheduler) are assumed to be collocated on the same servers and talk over localhost. Kubelet and kube-proxy are assumed to be collocated on every kubernetes node. etcd is assumed to be on 3 nodes that are dedicated and different from all the others. Those assumptions might be attacked at some point and things changed, these docs will be updated when that happens.

Our services/main cluster also uses calico as CNI (container networking interface) and helm as a deployment tool. Those are covered as well in the networking and deployment sections.

Versions

Kubernetes versioning is important and brutal. You might want to have a peek at our kubernetes components upgrade policy Kubernetes/Kubernetes_Infrastructure_upgrade_policy

This guide currently covers kubernetes 1.16, calico 3.16, helm 2.17

Prerequisites

  • Make sure you accept the restrictions about the versions above.
  • Allocate IP spaces for your cluster.
    • Calculate the maximum amount of pods you want to support and figure out using a subnet calculator (e.g. sipcalc) what IPv4 subnet you require (e.g. if you want a 100 pods, 128 pod IPs should be ok, so a /25 is enough). If you plan on max 1000 pods, you need 4 /24s (256 IPs) so a /22. Allocate them as active in Netbox. We can always add more pools after, but with IPv4 it's better to keep things a bit tidied. Don't forget IPv6. Allocate a /64. It should be enough regardless of amount of pods and will allow for growth.
    • Calculate the maximum amount of services you want to have (obviously it will be smaller than the amount of pods. Unless you plan to expose >250 services a /24 should be more than enough). Allocate it in Netbox. Don't forget IPv6. Allocate a /64. It should be enough regardless of growth

helmfile.d structure

We use extensively helmfile for all deployments, including creating all the cluster configuration.

Clone "https://gerrit.wikimedia.org/r/operations/deployment-charts" and navigate to helmfile.d/admin_ng/values hierarchy. The directories there are 1 per cluster. Copy one of those and amend it to fit your cluster.

Important things that WILL require alteration are:

File calico-values.yaml

# This is before coredns works, we can't rely on internal DNS, so use the external one
kubernetesApi:
 host: <myclusterdns e.g kubestagemaster.svc.codfw.wmnet> # You must have already a certificate by cergen for that
 port: 6443
BGPConfiguration:
 asNumber: 64602
IPPools:
  # These are the IP spaces you reserved for the cluster. It of course varies per DC
  ipv4-1:
     cidr: "myipv4/24"
  ipv6:
     cidr: "myipv6/64"
BGPPeers:
 # This actually per DC. It represents the IPv4+IPv6 IP of the core routers. Make sure to have the correct ones (which should happen if you copied the correct DC to start with)
 cr1-codfw-ipv4:
   asNumber: 14907
   peerIP: "208.80.153.192"
 cr2-codfw-ipv4:
   asNumber: 14907
   peerIP: "208.80.153.193"
 cr1-codfw-ipv6:
   asNumber: 14907
   peerIP: "2620:0:860:ffff::1"
 cr2-codfw-ipv6:
   asNumber: 14907
   peerIP: "2620:0:860:ffff::2"

File coredns-values.yaml

# This is before coredns works, we can't rely on internal DNS, so use the external one
kubernetesApi:
  host: <myclusterdns>
  port: 6443
service:
  # This is the cluster level IP that coredns will listen on. It MUST be in the service ip range you reserved previously and it MUST NOT be the very first one (.1) as that is internally used by kubernetes
  clusterIP: X.Y.Z.W


Components

etcd

etcd is a distributed datastore using the Raft algorithm for consensus. It is used by kubernetes to store cluster configuration as well as deployment data. In WMF it is also used for pybal, so there is some knowledge.

Depending on the critically of your new cluster, request an odd (recommended value is 3) number of small VMs on phabricator vm-requests project via SRE_Team_requests#Virtual_machine_requests_(Production). Then use Ganeti to create those VMs, followed by the guide in the dedicated page Etcd

Control-plane

Servers

The control plane houses kube-apiserver, kube-controller-manager, kube-scheduler. For this guide kube-controller-manager and kube-scheduler are assumed to talk to localhost kube-apiserver. If > 1 control-plane nodes exists, those 2 components will perform elections over the API about which is the main one at any given point in time (detection and failover is automatic).

Depending on the criticality of having the control plane always working request 1 or 2 small VMs on phabricator vm-requests project. Then use Ganeti to create those VMs.

Puppet/hiera

In our setup puppet roles are the way we instruct hiera to do lookups, but they don't have any functionality themselves (see Puppet_coding#Organization for a primer.

Create a new role for your nodes. The best way forward is to copy role::kubernetes::staging::master and set a proper system::role description. . Something like the following should be good enough

class role::foo::main {
   include ::profile::standard
   include ::profile::base::firewall

   # Sets up docker on the machine
   include ::profile::kubernetes::master

   system::role { 'foo::main':
       description => 'foo control plane server',
   }
}

If you are going to have >1 control plane nodes, add profile::lvs::realserver to the list of profiles included

Create the proper hiera files corresponding to your new role. e.g. if your new role is called role::foo::main then you want the following hiera files

  • hieradata/role/common/foo/master.yaml. This is where non-DC specific hiera values go. You can copy hieradata/role/common/kubernetes/staging/master.yaml, make sure to change keys, lvs configuration
  • hieradata/role/codfw/foo/master.yaml. This is codfw specific data. Mostly service cluster ip ranges and etcd things should be in there. Make sure to set the correct cluster service IP range that you reserved earlier as well as a list of the etcd hosts you created previously.
  • hieradata/role/eqiad/foo/master.yaml. This is eqiad specific data. Same rules apply as above
  • Create the corresponding private puppet repo and labs/private tokens. It should be just profile::kubernetes::master::controllermanager_token:. You can obtain them from the repos themselves (remember that labs/private is full of dummy tokens)
  • Create the certificates using Cergen in the puppet private repo.
  • Put the public cert that was obtained from the above step in the public repo under the files/ssl directory with the proper name.

Apply the above role to your new node(s)

All of the above can be done in 1 patch while using the puppet compiler

LVS

Follow LVS#Add a new load balanced service

Users/tokens

Our user/token populating process is currently hardwired to work across all clusters the same way. You will get all the users that the main services kubernetes clusters have. That is a limitation of our lack of a proper authentication layer that we have not yet solved.


Node (worker)

This setup is meant (and achieves) to provide a hands off approach to node provisioning/reprovisioning/imaging etc. That is from the moment the node is declared ready to be put in service and the puppet role (and respective hiera) has been applied, a single re-image should suffice for the node the registered to the API and be ready to receive traffic.

Notes

  • The setup has only been tested with the specific partman recipe present in partman/custom/kubernetes-node.cfg. It creates a specific vg the is meant to be deleted and recreated by puppet on first role apply.
  • docker is mean to be used as the CRE. Other runtime engines aren't currently supported
  • Currently docker is using the lvm devicemapper. This graph driver is deprecated. When we move to buster or bullseye we expect to not use the devicemapper graph driver and rather rely on overlay graph driver
  • The CNI of choice is calico and it is deployed via a Kubernetes Daemonset. A node component is running on every node and is the one providing connectivity to pods. Failure of that components means pods have no connectivity
  • This setup is tested both with version 4.9 as well as 4.19 of the linux kernel. 4.19 is the recommended one currently.

General Puppet/hiera setup

In our setup puppet roles are the way we instruct hiera to do lookups, but they don't have any functionality themselves (see Puppet_coding#Organization for a primer.

Create a new role for your nodes. The best way forward is to copy role::kubernetes::staging::worker and set a proper system::role description. Something like the following should be good enough

class role::foo::worker {
   include ::profile::standard
   include ::profile::base::firewall
   include ::profile::base::linux419

   # Sets up docker on the machine
   include ::profile::docker::storage
   include ::profile::docker::engine
   # Setup kubernetes stuff
   include ::profile::kubernetes::node
   # Setup calico
   include ::profile::calico::kubernetes

   system::role { 'foo::worker':
       description => 'foo worker node',
   }
}

In case you expect to expose services via LVS, add profile::lvs::realserver in the list of profiles you include.

Create the proper hiera files corresponding to your new role. e.g. if your new role is called role::foo::worker then you want the following hiera files

  • hieradata/role/common/foo/worker.yaml. This is where non-DC specific hiera values go. You can copy hieradata/role/common/kubernetes/staging/worker.yaml. It should mostly not require changes.
  • hieradata/role/codfw/foo/worker.yaml. This is codfw specific data. You need to update:
# Enter your control plain DNS
profile::kubernetes::master_fqdn: <foo> 
# The list of control-plain nodes. This is used to open up firewall rules
profile::kubernetes::master_hosts:
  - main1
  - main2
# The IP coredns will listen on. It needs to be in your service IP cluster range. Don't use .1 it's used internally by kubernetes
profile::kubernetes::node::kubelet_cluster_dns: X.Y.Z.W
# Enter your control plain DNS
profile::rsyslog::kubernetes::kubernetes_url: <foo>

  • hieradata/role/eqiad/foo/worker.yaml. This is eqiad specific data. Same rules apply as above

Make sure to create the corresponding private puppet repo and labs/private tokens. You don't get to generate them on your own currently as they are shared across all cluster until we can have a better solution. So reuse what the services k8s cluster users. Things to define:

profile::kubernetes::node::kubeproxy_token: dummytoken1
profile::kubernetes::node::kubelet_token: dummytoken2
profile::rsyslog::kubernetes::token: dummytoken3
profile::calico::kubernetes::calico_cni::token: dummytoken4
profile::calico::kubernetes::calicoctl::token: dummytoken5

Access to restricted docker images

If your nodes need access to restricted docker images (see: T273521 for context), you have provide credentials for the docker registry to your nodes. This can be done by adding the hiera key profile::kubernetes::node::docker_kubernetes_user_password to the file hieradata/role/common/foo/worker.yaml in in the private puppet repository.

See Docker-registry#Access_control on how to find the correct password.


Adding Nodes

For adding nodes (based on the generic setup described above) please follow Kubernetes/Clusters/Add_or_remove_nodes

Apply RBAC rules and PSPs

If you have your helmfile.d/admin_ng ready you can apply at least RBAC and Pod Security Policies

Note: these commands need to be run as logged-in root (just prefixing them with sudo will not work).

$ deploy100X:/srv/deployment-charts/helmfile.d/admin_ng# helmfile -e <my_cluster> -l name=rbac-rules sync
$ deploy100X:/srv/deployment-charts/helmfile.d/admin_ng# helmfile -e <my_cluster> -l name=pod-security-policies sync

After this stage your nodes will registered to the API, but will not be ready to receive pods, cause you lack the next section.

Label Kubernetes Masters

For some clusters there is the need to add specific node labels to identify roles, for example the master nodes part of the control plane:

Note: these commands need to be run as logged-in root (just prefixing them with sudo will not work). As root, you may need to run kube_env admin somecluster as well

kubectl label nodes ml-serve-ctrl1001.eqiad.wmnet node-role.kubernetes.io/master=""
kubectl label nodes ml-serve-ctrl1002.eqiad.wmnet node-role.kubernetes.io/master=""

Due to https://github.com/kubernetes/kubernetes/issues/84912#issuecomment-551362981, we cannot add the above labels to the ones set by the Kubelet when registering the node, so this step needs to be done manually when bootstrapping the cluster. The labels will be useful for NetworkPolicies, for example to identify traffic coming from the master nodes towards a certain pod (likely a webhook).

Networking

First of all, have a look in Network design for how a DC (not a caching pop) is cable network wise. It will help get an idea of what it is you are going to be doing in this section.

What we are going to do in this section is have the nodes talk BGP to the cr*-<site> core routers (aka the juniper routers) and vice versa (it's a bidirectional protocol).

Core routers

  • Craft a change like the following https://gerrit.wikimedia.org/r/c/operations/homer/public/+/661055 (before proceeding make sure that the Network Engineers are aware of what you are going to do).
  • Apply it to the core routers using Homer
  • Don't worry about BGP alerts, the important bit is doing this step and the next one one after the other (to establish iBGP sessions).

Calico node/controllers

Now you can deploy all calico components

At this stage you can probably deploy the entire helmfile.d structure in 1 go but since RBAC/PSPs are already covered above we are going to just mention calico here.

# remember to root-login via sudo -i first
$ deploy100X:/srv/deployment-charts/helmfile.d/admin_ng$ helmfile -e <my_cluster> -l name=calico-crds sync
$ deploy100X:/srv/deployment-charts/helmfile.d/admin_ng$ helmfile -e <my_cluster> -l name=calico sync

Note: if the second command times out, you may need to sync the namespaces first.

There are dependencies between the 2 so you don't really need this level of release by relase, but for clarity:

* The CRDs (Custom Resource Definitions) are calico's way of storing its' data in the Kubernetes API
* The calico release itself will setup a calico-node pod in every node with hostNetwork: true (that is it will not have its own IP address but rather share it with the host), 1 calico typha pod and 1 calico kube-controllers pod. 

If this succeeds, you are almost ready to deploy workloads, but have a look for 2 rather crucial cluster tools below.

To check if calico works:

root@deploy1002:/srv/deployment-charts/helmfile.d/admin_ng# kube_env admin ml-serve-eqiad
root@deploy1002:/srv/deployment-charts/helmfile.d/admin_ng# kubectl -n kube-system get deployment,daemonset
NAME                                      READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/calico-kube-controllers   1/1     1            1           2m29s
deployment.apps/calico-typha              1/1     1            1           2m29s

NAME                         DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR            AGE
daemonset.apps/calico-node   4         4         4       4            4           kubernetes.io/os=linux   2m29s

And if you want to check on the routers, ssh to one of them (like cr1-eqiad.wikimedia.org) and run the following:

$ show bgp neighbor
[..]
  Description: ml-serve1002             
  Group: Kubemlserve4          Routing-Instance: master
  Forwarding routing-instance: master  
  Type: External    State: Established    Flags: <Sync>
  Last State: OpenConfirm   Last Event: RecvKeepAlive
  Last Error: None
[..]

You should see an established session for all the k8s workers of your cluster.

Cluster tools

There are 2 cluster level tools you probably want:

CoreDNS

CoreDNS is the deployment and service that provides outgoing DNS resolution to pods as well as internal DNS discovery. It is NOT used by deployments that are hostNetwork: true (e.g. calico-node) in our setup on purpose.

Assuming you populated the helmfile.d/admin_ng/values/<cluster>/ it can be populated with

helmfile -e <mycluster> -l name=coredns sync

To check that everything is up and running as expected:

root@deploy1002:/srv/deployment-charts/helmfile.d/admin_ng# kube_env admin ml-serve-eqiad
root@deploy1002:/srv/deployment-charts/helmfile.d/admin_ng# kubectl -n kube-system get deployment,daemonset
NAME                                      READY   UP-TO-DATE   AVAILABLE   AGE
deployment.apps/calico-kube-controllers   1/1     1            1           15m
deployment.apps/calico-typha              1/1     1            1           15m
deployment.apps/coredns                   4/4     4            4           49s

NAME                         DESIRED   CURRENT   READY   UP-TO-DATE   AVAILABLE   NODE SELECTOR            AGE
daemonset.apps/calico-node   4         4         4       4            4           kubernetes.io/os=linux   15m

After the deployment of the coredns pods, you are free to merge a change like https://gerrit.wikimedia.org/r/c/operations/puppet/+/673985 to configure the coredns service ip to all kubelets/workers.

Eventrouter

Eventrouter aggregates and sends to logstash kubernetes events.

Deploy it with on the deployment host:

# remember to root-login via sudo -i first
helmfile -e <mycluster> -l name=eventrouter sync

To check that everything is up and running as expected, see what's done for coredns.

Namespaces

Namespaces are created using helmfile and the main clusters (production + staging) all share them, however they are overridable per cluster. The main key is at helmfile.d/admin_ng. An example of augmenting it is at helmfile.d/admin_ng/staging

The same structure also holds limitRanges and resourceQuotas. Note that it's a pretty opinionated way

Creating them is done with the following command on the deployment host:

# remember to root-login via sudo -i first
helmfile -e staging-codfw -l name=namespaces sync

Which means that if you don't want the main namespaces populated (which makes sense), your best bet is to skip running that command. Alternatively override the main values for your cluster.

Prometheus

Prometheus talks to the api and discovers the API server, nodes, pods, endpoints and services. In WMF we only scrape the API server, the nodes and the pods. We have 2 nodes per DC doing the scraping. Those will need to be properly configured to scrape the new cluster.

This happens via the files:

  • hieradata/role/eqiad/prometheus.yaml
  • hieradata/role/codfw/prometheus.yaml

An example stanza is pasted below, hopefully it's self documenting.

# A hash containing configuration for kubernetes clusters.
profile::prometheus::kubernetes::clusters:
  k8s:
    enabled: true
    master_host: 'kubemaster.svc.codfw.wmnet'
    port: 9906
    class_name: role::kubernetes::worker
  k8s-staging:
    enabled: false
    master_host: 'kubestagemaster.svc.codfw.wmnet'
    port: 9907
    class_name: role::kubernetes::staging::worker
# In the private repo a stanza like the following is required
# profile::prometheus::kubernetes::cluster_tokens:
# k8s:
#   client_token: eqiaddummy
# k8s-staging:
#   client_token: eqiaddummystaging

The above will only add the config for the new Prometheus instance on the master nodes:

  • prometheus100[3,4].eqiad.wmnet
  • prometheus200[3,4].codfw.wmnet

Please verify on them that the new systemd units are working as expected. Once done, you can follow up with:

The first code review also need a puppet run on the grafana nodes to pick up the new config. Once done, you should be able to see the new cluster in the Kubernetes Grafana dashboards!

LVM creation

This is unfortunately currently manual, requiring the creation of lvm volumes on multiple prometheus nodes:

  • prometheus100[3,4].eqiad.wmnet
  • prometheus200[3,4].codfw.wmnet

Please follow up with a member of the Observability team first to let them know what you are doing, so they are aware.

Decide what kind of speed class and disk space you want (essentially HDD vs SSD) and run on the correct nodes (the ones having prometheus role the commands

Set the size, the name of the k8s cluster and speed class

SIZE=X
CLUSTER_NAME=CLUSTER_NAME
VG=vg-hdd

Then run (careful, this IS NOT idempotent)

lvcreate --size ${SIZE}GB --name prometheus-${CLUSTER_NAME} ${VG}
mkfs.ext4 /dev/mapper/$(echo $VG | sed -e 's/-/--/')-prometheus--$(echo $CLUSTER_NAME | sed -e s/-/--/)
mkdir /srv/prometheus/${CLUSTER_NAME}
echo "/dev/${VG}/prometheus-${CLUSTER_NAME}	/srv/prometheus/${CLUSTER_NAME}	ext4	defaults	0	0" >> /etc/fstab
mount /srv/prometheus/${CLUSTER_NAME}

And you should be good to go.