Container Days Boston - Kubernetes in production

Kubernetes in Production,
From the Ground Up
Mike Splain | @mikesplain

About Me
• Lead DevOps Engineer at Barkly
• Formerly PayPal / eBay & OurStage
• Running applications at scale with the least amount of human interaction
• Current Technologies: Kubernetes, Docker, Terraform, Ansible
• Previous Technologies: Chef, Test-Kitchen….

Resources
• docs.kubernetes.io
• coreos.com/kubernetes/docs
• bostonkubernetes.com
• slack.kubernetes.io
• github.com/mikesplain/etcd_demo

Our requirements for running our apps
• Fast recovery without human intervention
• Nodes are ephemeral
• Autoscaling
• Testable on developer machines
• Infrastructure as an artifact

Technology choices for running K8s

Hasn’t someone else already built this?
… March 2015

K8s official scripts
• Simple bash scripts
• “Just Works!”
• Sets up Autoscaling group for
minions
• Uses Salt
Problems? (at the time)
• No etcd HA
• No Master HA
• Salt Master is coupled with K8s
Master
• Ubuntu / Fedora

CoreOS’s official scripts
• Cool go app to start it!
• Or Cloud formation?
• But what now?
Problems? (at the time)
• No etcd HA
• No Master HA
• Lots of magic

“It can’t be that hard right?”

Container Days Boston - Kubernetes in production

Note to self… you’re going too slow.

etcd
• Easy.
• “Just works”
• Cluster discovery:
• Discovery Service
• DNS
• ?
#cloud-config
coreos:
etcd2:
advertise-client-urls: "http://$public_ipv4:2379"
initial-advertise-peer-urls: "http://$private_ipv4:2380"
listen-client-urls: "http://0.0.0.0:2379,http://0.0.0.0:4001"
listen-peer-urls: "http://$private_ipv4:2380,http://$private_ipv4:7001"
discovery-token: “<token here>”
units:
- name: etcd2.service
command: start
update:
reboot-strategy: none

etcd
• etcd-aws-cluster 
https://github.com/MonsantoCo/
etcd-aws-cluster
• Uses Autoscaling groups for
discovery
• Requires IAM Instance Roles
#cloud-config
coreos:
etcd2:
advertise-client-urls: "http://$public_ipv4:2379"
initial-advertise-peer-urls: "http://$private_ipv4:2380"
listen-client-urls: "http://0.0.0.0:2379,http://0.0.0.0:4001"
listen-peer-urls: "http://$private_ipv4:2380,http://$private_ipv4:7001"
units:
command: stop
- name: etcd-peers.service
command: start
content: |
[Unit]
Description=Write a file with the etcd peers that we should
bootstrap to
Requires=docker.service
After=docker.service
[Service]
Restart=on-failure
RestartSec=10
TimeoutStartSec=300
ExecStartPre=/usr/bin/docker pull registry.barklyprotects.com/
kubernetes/etcd-aws-cluster:latest
ExecStartPre=/usr/bin/docker run --rm=true -v /etc/sysconfig/:/etc/
sysconfig/ registry.barklyprotects.com/kubernetes/etcd-aws-cluster:latest
ExecStart=/usr/bin/systemctl start etcd2
write_files:
- path: /etc/systemd/system/etcd2.service.d/30-etcd_peers.conf
permissions: 0644
content: |
[Service]
# Load the other hosts in the etcd leader autoscaling group from file
EnvironmentFile=/etc/sysconfig/etcd-peers

Terraform to launch etcd
• References static cloud-init files
resource "aws_launch_configuration" "terraform_etcd" {
name_prefix = "${var.environment}_etcd_conf-"
image_id = "${var.coreos_ami}"
instance_type = "t2.small"
key_name = "${var.key_name}"
security_groups = ["$
{aws_security_group.terraform_etcd2_sec_group.id}"]
user_data = "${file("../cloud-config/output/etcd.yml")}"
enable_monitoring = true
ebs_optimized = false
iam_instance_profile = "$
{aws_iam_instance_profile.terraform_etcd_role_profile.id}"
root_block_device {
volume_size = 20
}
lifecycle {
create_before_destroy = true
}
}
resource "aws_autoscaling_group" "terraform_etcd" {
name = "${var.environment}_etcd"
launch_configuration = "$
{aws_launch_configuration.terraform_etcd.name}"
availability_zones = ["us-east-1c"]
max_size = "${var.capacities_etcd_max}"
min_size = "${var.capacities_etcd_min}"
health_check_grace_period = 300
desired_capacity = "${var.capacities_etcd_desired}"
vpc_zone_identifier = ["${aws_subnet.etcd.id}"]
force_delete = true
tag {
key = "Name"
value = "${var.environment}_etcd"
propagate_at_launch = true
}
lifecycle {
}
}

Master Node
• Kubelet service
• API Server
• Controller Manager
• Scheduler
• Proxy
• Flannel

Consider Master pods as
additional artifact
• Docker container that takes env
variables
• Outputs templated pods to disk for
Kubelet to load
• We use j2cli to template these files
with little overhead
apiVersion: v1
kind: Pod
metadata:
name: kube-podmaster
namespace: kube-system
spec:
hostNetwork: true
containers:
- name: scheduler-elector
image: gcr.io/google_containers/podmaster:1.1
imagePullPolicy: Always
command:
- /podmaster
- --etcd-servers={{ ETCD_ENDPOINTS }}
- --key=scheduler
- --whoami={{ ADVERTISE_IP }}
- --source-file=/src/manifests/kube-scheduler.yaml
- --dest-file=/dst/manifests/kube-scheduler.yaml
volumeMounts:
- mountPath: /src/manifests
name: manifest-src
readOnly: true
- mountPath: /dst/manifests
name: manifest-dst

Master Cloud Config
• Setup box for services needed for
real Docker to run
• Get etcd server IPs and write to file
• Start flannel with those IPs
#cloud-config
coreos:
units:
- name: etcd.service
command: stop
command: stop
- name: early-docker.service
command: start
- name: kub_get_etcd.service
command: start
content: |
[Unit]
Description= Write K8s etcd urls to disk.
Requires=early-docker.service
After=early-docker.service
Before=early-docker.target
[Service]
Type=oneshot
Environment="DOCKER_HOST=unix:///var/run/early-docker.sock"
ExecStart=/usr/bin/sh -c "/usr/bin/docker pull
registry.barklyprotects.com/kubernetes/kub-get-etcd"
ExecStart=/usr/bin/sh -c "/usr/bin/docker run --net=host -v /etc/
barkly/:/etc/barkly/ registry.barklyprotects.com/kubernetes/kub-get-etcd {{
KUB_ETCD_ASG }} > /etc/etcd_servers.env"
- name: flanneld.service
command: start
drop-ins:
- name: 10-environment_vars.conf
content: |
[Unit]
After=kub_get_etcd.service
[Service]
ExecStartPre=/usr/bin/sh -c "/usr/bin/echo -n
FLANNELD_ETCD_ENDPOINTS= > /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/sh -c "/usr/bin/cat /etc/etcd_servers.env
>> /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/sh -c "/usr/bin/echo FLANNELD_IFACE=
$private_ipv4 >> /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/ln -sf /etc/flannel_etcd_servers.env /run/
flannel/options.env
Restart=always
RestartSec=10

Other config
• Grab certs from S3
• Terraform only allows
permissions to specific files
• Format Master pod files to disk
- name: kub_certs.service
command: start
content: |
[Unit]
Description=Writes kubernetes cluster certs to disk.
Before=kubelet.service
[Service]
Type=oneshot
ExecStart=/usr/bin/sh -c /usr/bin/mkdir -p /etc/kubernetes/ssl
ExecStart=/usr/bin/docker run --net=host -v /etc/kubernetes/ssl:/
ssl registry.barklyprotects.com/ops/awscli s3 cp s3://
our_k8s_cluster_bucket/ca.pem /ssl
our_k8s_cluster_bucket/apiserver.pem /ssl
our_k8s_cluster_bucket/apiserver-key.pem /ssl
- name: kub_pods.service
command: start
content: |
[Unit]
Description=Writes kubernetes pod files to disk.
Before=kubelet.service
[Service]
Type=oneshot
ExecStart=/usr/bin/sh -c "/usr/bin/mkdir -p /etc/kubernetes/ssl"
ExecStart=/usr/bin/docker run --net=host -v /etc/barkly/:/etc/
barkly/ -e K8S_VERSION='1.1.7' -e CLOUD_PROVIDER='--cloud-provider=aws' -e
SERVICE_IP_RANGE="10.3.0.0/16" -e ADVERTISE_IP="$private_ipv4" -e
ETCD_AUTOSCALE_GROUP_NAME="our_etcd_autoscaling_group_name" -v /srv/
kubernetes/manifests:/output_src -v /etc/kubernetes/manifests:/output_dst
registry.barklyprotects.com/kubernetes/kub-master-pods

Start Docker & Kubelet
• Kubelet will wait for docker and
flannel to be ready
• Kubelet will load manifests for
Master services outputted from
previous container
- name: docker.service
command: start
drop-ins:
- name: 40-flannel.conf
content: |
[Unit]
Requires=flanneld.service
After=flanneld.service
- name: kubelet.service
command: start
content: |
[Unit]
After=fluentd-elasticsearch.service
[Service]
ExecStartPre=/usr/bin/mkdir -p /var/log/containers
ExecStart=/etc/bin/kubelet
--hostname-override="$private_ipv4"
--api_servers=http://127.0.0.1:8080
--register-node=false
--allow-privileged=true
--config=/etc/kubernetes/manifests
--cluster-dns=10.3.0.10
--cluster-domain=cluster.local
--cloud-provider=aws
--v=4
Restart=always
RestartSec=10
[Install]
WantedBy=multi-user.target

Get Kubelet
• /usr/bin & /usr/local/bin are read
only in CoreOS
command: start
drop-ins:
- name: 10-download-binary.conf
content: |
[Service]
ExecStartPre=/bin/bash -c "/etc/bin/download-k8s-binary kubelet"
write_files:
# Since systemd needs these files before it will start
- path: /etc/bin/download-k8s-binary
permissions: '0755'
content: |
#!/usr/bin/env bash
export K8S_VERSION=“v1.2.4"
mkdir -p /etc/bin
FILE=$1
if [ ! -f /usr/bin/$FILE ]; then
curl -sSL -o /etc/bin/$FILE https://s3.amazonaws.com/barkly-
kubernetes-builds/${K8S_VERSION}/bin/$FILE
chmod +x /etc/bin/$FILE
else
# we check the version of the binary
INSTALLED_VERSION=$(/etc/bin/$FILE --version)
MATCH=$(echo "${INSTALLED_VERSION}" | grep -c "${K8S_VERSION}")
if [ $MATCH -eq 0 ]; then
# the version is different
curl -sSL -o /etc/bin/$FILE https://s3.amazonaws.com/barkly-
kubernetes-builds/${K8S_VERSION}/bin/$FILE
chmod +x /etc/bin/$FILE
fi
fi

Terraform to build
• Similar to as before.. we reference
our cloudinit script
resource "aws_launch_configuration" "terraform_master" {
name_prefix = "${var.environment}_master_conf-"
image_id = "${var.coreos_ami}"
instance_type = "t2.medium"
key_name = "${var.key_name}"
security_groups = ["$
{aws_security_group.terraform_master_sec_group.id}"]
user_data = "${file("../cloud-config/output/master.yml")}"
enable_monitoring = true
ebs_optimized = false
iam_instance_profile = "$
{aws_iam_instance_profile.terraform_master_role_profile.id}"
root_block_device {
volume_size = 20
}
lifecycle {
}
}
resource "aws_autoscaling_group" "terraform_master" {
name = "${var.environment}_master"
launch_configuration = "$
{aws_launch_configuration.terraform_master.name}"
availability_zones = ["us-east-1c"]
max_size = "${var.capacities_master_max}"
min_size = "${var.capacities_master_min}"
health_check_grace_period = 300
desired_capacity = "${var.capacities_master_desired}"
vpc_zone_identifier = ["${aws_subnet.master.id}"]
force_delete = true
load_balancers = ["${aws_elb.terraform_master.name}"]
tag {
key = "Name"
value = "${var.environment}_master"
propagate_at_launch = true
}
lifecycle {
}
}

Minion Node (now just Nodes)
• Kubelet Service manages all other services.
• Proxy
• Pods

Early Docker / Flannel
• Same as master!
#cloud-config
coreos:
units:
- name: etcd.service
command: stop
command: stop
- name: early-docker.service
command: start
- name: kub_get_etcd.service
command: start
content: |
[Unit]
Description= Write K8s etcd urls to disk.
[Service]
Type=oneshot
ExecStart=/usr/bin/sh -c "/usr/bin/docker pull
registry.barklyprotects.com/kubernetes/kub-get-etcd"
ExecStart=/usr/bin/sh -c "/usr/bin/docker run --net=host -v /etc/
barkly/:/etc/barkly/ registry.barklyprotects.com/kubernetes/kub-get-etcd {{
KUB_ETCD_ASG }} > /etc/etcd_servers.env"
- name: flanneld.service
command: start
drop-ins:
- name: 10-environment_vars.conf
content: |
[Unit]
After=kub_get_etcd.service
[Service]
ExecStartPre=/usr/bin/sh -c "/usr/bin/echo -n
FLANNELD_ETCD_ENDPOINTS= > /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/sh -c "/usr/bin/cat /etc/etcd_servers.env
>> /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/sh -c "/usr/bin/echo FLANNELD_IFACE=
$private_ipv4 >> /etc/flannel_etcd_servers.env"
ExecStartPre=/usr/bin/ln -sf /etc/flannel_etcd_servers.env /run/
flannel/options.env
Restart=always
RestartSec=10

Kubelet
• Similar to master
- name: docker.service
command: start
drop-ins:
- name: 40-flannel.conf
content: |
[Unit]
Requires=flanneld.service
After=flanneld.service
command: start
content: |
[Unit]
After=fluentd-elasticsearch.service
[Service]
ExecStartPre=/usr/bin/mkdir -p /var/log/containers
ExecStart=/etc/bin/kubelet
--api_servers=https://ourk8smaster.barkly.com
--hostname-override="$private_ipv4"
--register-node=true
--allow-privileged=true
--config=/etc/kubernetes/manifests
--cluster-dns=10.3.0.10
--cluster-domain=cluster.local
--kubeconfig=/etc/kubernetes/worker-kubeconfig-kubelet.yaml
--tls-cert-file=/etc/kubernetes/ssl/worker.pem
--tls-private-key-file=/etc/kubernetes/ssl/worker-key.pem
--cloud-provider=aws
--v=4
Restart=always
RestartSec=10
[Install]
WantedBy=multi-user.target
drop-ins:
- name: 10-download-binary.conf
content: |
[Service]
ExecStartPre=/bin/bash -c "/etc/bin/download-k8s-binary kubelet"

Manifests can be hard coded
• Minion manifests are far less
dynamic and can be hard coded
as CloudInit files
write_files:
- path: "/etc/kubernetes/manifests/kube-proxy.yaml"
content: |
apiVersion: v1
kind: Pod
metadata:
name: kube-proxy
namespace: kube-system
spec:
hostNetwork: true
containers:
- name: kube-proxy
image: registry.barklyprotects.com/kubernetes/hyperkube:v1.1.8
command:
- /hyperkube
- proxy
- --master=https://ourk8smaster.barkly.com
- --kubeconfig=/etc/kubernetes/worker-kubeconfig-proxy.yaml
- --proxy-mode=iptables
- --v=4
securityContext:
privileged: true
volumeMounts:
- mountPath: /etc/ssl/certs
name: "ssl-certs"
- mountPath: /etc/kubernetes/worker-kubeconfig-proxy.yaml
name: "kubeconfig"
readOnly: true
- mountPath: /etc/kubernetes/ssl
name: "etc-kube-ssl"
readOnly: true
volumes:
- name: "ssl-certs"
hostPath:
path: "/usr/share/ca-certificates"
- name: "kubeconfig"
hostPath:
path: "/etc/kubernetes/worker-kubeconfig-proxy.yaml"
- name: "etc-kube-ssl"
hostPath:
path: "/etc/kubernetes/ssl"

Manifests can be hard coded
• Minion manifests are far less
dynamic and can be hard coded
as CloudInit files
- path: "/etc/kubernetes/worker-kubeconfig-proxy.yaml"
content: |
apiVersion: v1
kind: Config
clusters:
- name: local
cluster:
certificate-authority: /etc/kubernetes/ssl/ca.pem
users:
- name: kubelet
user:
client-certificate: /etc/kubernetes/ssl/worker.pem
client-key: /etc/kubernetes/ssl/worker-key.pem
contexts:
- context:
cluster: local
user: kubelet
name: kubelet-context
current-context: kubelet-context
- path: "/etc/kubernetes/worker-kubeconfig-kubelet.yaml"
content: |
apiVersion: v1
kind: Config
clusters:
- name: local
cluster:
certificate-authority: /etc/kubernetes/ssl/ca.pem
users:
- name: kubelet
user:
client-certificate: /etc/kubernetes/ssl/worker.pem
client-key: /etc/kubernetes/ssl/worker-key.pem
contexts:
- context:
cluster: local
user: kubelet
name: kubelet-context
current-context: kubelet-context

K8s does not so well:
• Still relatively young,
• Not extremely google-able
• Leaves managing manifests to user with little guidance
• Docker only (rkt coming soon)
• No DNS support (coming soon)
• SaltStack by default

K8s does really well:
• Keeping applications running
• Managing failover, liveness,readiness of applications
• Mounting volumes, load balancers (cloud resources)
• Internal DNS, Namespacing
• Basic secret / config management
• Easily Horizontally Scalable services
• Rolling updates / rollback via Deployments
• Support via Github / Slack chat
• Easy debugging of actual app problems via kubectl logs and kubectl exec

https://github.com/mikesplain/etcd_demo
Questions?

Container Days Boston - Kubernetes in production

Recommended

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to Container Days Boston - Kubernetes in production (20)

Recently uploaded (20)

Container Days Boston - Kubernetes in production