La capacità di rispondere in poche frazioni di secondo alle richieste degli utenti - indipendentemente dal loro numero - è un fattore determinante per il successo dei servizi sul web. Secondo Amazon, bastano 100 millisecondi di latenza nella risposta per generare una perdita economica di circa l'1% sul
fatturato [1]. In base alle statistiche di Google AdWords, inoltre, il 2015 ha sancito l’ufficiale superamento del numero di interazioni mobile rispetto a quelle desktop [2], con la conseguente riduzione della durata media delle sessioni di navigazione web.
In uno scenario di questo tipo, la razionalizzazione dell’utilizzo delle risorse hardware e la capacità di scalare rispetto al numero di utenti sono fattori determinanti per il successo del business.
In questo talk racconteremo la nostra esperienza di migrazione di soluzioni e-commerce di tipo enterprise in Magento da un’architettura basata su VM tradizionali ad una di tipo software-defined basata su Kubernetes, Flannel e Docker. Discuteremo, quindi, delle reali difficoltà da noi incontrate nel porting su container di soluzioni in produzione e daremo evidenza di come, alla fine di questo lungo viaggio, i nostri sforzi siano stati concretamente premiati dall’aumento di resilienza, affidabilità e automazione della soluzione finale.
A supporto della conversazione, mostreremo i risultati dei benchmark da noi condotti per valutare la scalabilità della nuova architettura presentando delle evidenze delle reali capacità di Kubernetes come strumento di orchestrazione di servizi erogati in Docker container.
Concluderemo l’intervento presentando il nostro progetto di distribuzione geografica dei nodi master di Kubernetes facendo uso di reti SD-WAN per garantire performance e continuità di servizio della soluzione.
2. Outline
● Introduction to Purple Ocean
● Magento
○ The good
○ The bad
○ The ugly
● Project Picasso
○ Software Defined Overlay Networking with Flannel
○ Cluster Mangement with Kubernetes
● What’s Next
○ Geographical Distribution
○ Hybrid Cloud
5. Magento
“The high performance, scalable
eCommerce solution for fast-growing
and large businesses.”
https://magento.com/products/overview
6. Magento: the goods
1. Wide range of features and
functionalities
2. Integrations
3. Large community
4. Open source
5. Thousands of agencies and developers
6. Enterprise edition available
https://www.elementarydigital.co.uk/why-use-magento-for-ecommerce/
7. Magento
“Every 100ms of latency costs Amazon
1% of profit”
http://radar.oreilly.com/2008/08/radar-theme-web-ops.html
“Google says more searches now on
mobile than on desktop”
http://searchengineland.com/its-official-google-says-more-
searches-now-on-mobile-than-on-desktop-220369
8. Magento
“Every 100ms of latency costs Amazon
1% of profit”
http://radar.oreilly.com/2008/08/radar-theme-web-ops.html
“Every 100ms of latency costs Amazon
1% of profit”
http://radar.oreilly.com/2008/08/radar-theme-web-ops.html
PERFORMANCES
MATTER
9. Magento: the bad
Running medium-size businesses on
Magento, could require a lot of caching
layers and tricks.
14. Project Picasso
“Docker is an open source project
to pack, ship and run any
application as a lightweight
container. Docker containers are
both hardware-agnostic and
platform-agnostic.”
https://github.com/docker/docker
“Flannel is a virtual network that
gives a subnet to each host for
use with container runtimes.”
https://coreos.com/flannel/docs/latest/
“Kubernetes is an open-source
system for automating
deployment, operations, and
scaling of containerized
applications.”
http://kubernetes.io/
20. Persistent Volume
A PersistentVolume (PV) is a piece of networked storage in the cluster that has
been provisioned by an administrator. It is a resource in the cluster just like a
node is a cluster resource.
http://kubernetes.io/docs/user-guide/persistent-volumes/
# Allocates 5GB of storage on the given NFS server
apiVersion: v1
kind: PersistentVolume
metadata:
name: web-storage
spec:
capacity:
storage: 5Gi
accessModes:
- ReadWriteMany
nfs:
server: kube-master-1
path: "/root/nfs"
21. Persistent Volume
A PersistentVolume (PV) is a piece of networked storage in the cluster that has
been provisioned by an administrator. It is a resource in the cluster just like a
node is a cluster resource.
http://kubernetes.io/docs/user-guide/persistent-volumes/
# kubectl create -f web-storage-pv.yaml
persistentvolume "web-storage" created
# kubectl describe pv web-storage
Name: web-storage
Labels: <none>
Status: Available
Claim:
Reclaim Policy: Retain
Access Modes: RWX
Capacity: 5Gi
Message:
Source:
Type: NFS (an NFS mount that lasts the lifetime of a pod)
Server: kube-master-1
Path: /root/nfs
ReadOnly: false
22. Persistent Volume Claim
A PersistentVolumeClaim (PVC) is a request for storage by a user. It is similar
to a pod. Pods consume node resources and PVCs consume PV resources.
# This PersistentVolumeClaim claims 5GB of storage
# for the web-application.
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: web-storage-claim
spec:
accessModes:
- ReadWriteMany
resources:
requests:
storage: 5Gi
http://kubernetes.io/docs/user-guide/persistent-volumes/
23. Persistent Volume Claim
A PersistentVolumeClaim (PVC) is a request for storage by a user. It is similar
to a pod. Pods consume node resources and PVCs consume PV resources.
# kubectl create -f web-storage-pvc.yaml
persistentvolumeclaim "web-storage-claim" created
http://kubernetes.io/docs/user-guide/persistent-volumes/
# kubectl describe pvc web-storage
Name: web-storage-claim
Namespace: default
Status: Bound
Volume: web-storage
Labels: <none>
Capacity: 5Gi
Access Modes: RWX
24. Pod & ReplicationController
A ReplicationController (RC) ensures that a specified number of pod “replicas” are
running at any one time. In other words, a replication controller makes sure that a pod
or homogeneous set of pods are always up and available. If there are too many pods,
it will kill some. If there are too few, the replication controller will start more
http://kubernetes.io/docs/user-guide/replication-controller/
apiVersion: v1
kind: ReplicationController
metadata:
name: web-frontend
spec:
replicas: 2
selector:
role: web-frontend
template:
metadata:
labels:
role: web-frontend
spec:
containers:
- name: web-frontend
image: php:5.5-apache
ports:
- name: web
containerPort: 80
volumeMounts:
- name: nfs
mountPath: "/var/www/html"
volumes:
- name: nfs
persistentVolumeClaim:
claimName: web-storage-claim
25. # kubectl create -f web-frontend-rc.yaml
replicationcontroller "web-frontend" created
# kubectl describe rc web-frontend
Name: web-frontend
Namespace: default
Image(s): php:5.5-apache
Selector: role=web-frontend
Labels: role=web-frontend
Replicas: 2 current / 2 desired
Pods Status: 2 Running / 0 Waiting / 0 Succeeded / 0 Failed
Volumes:
nfs:
Type:PersistentVolumeClaim (a reference to a PersistentVolumeClaim in the same namespace)
ClaimName: web-storage-claim
ReadOnly: true
[...]
Pod & ReplicationController
A ReplicationController (RC) ensures that a specified number of pod “replicas” are
running at any one time. In other words, a replication controller makes sure that a pod
or homogeneous set of pods are always up and available. If there are too many pods,
it will kill some. If there are too few, the replication controller will start more
http://kubernetes.io/docs/user-guide/replication-controller/
26. # Expose the web-frontend on each Minion
apiVersion: v1
kind: Service
metadata:
name: web-frontend
spec:
ports:
- port: 80
type: NodePort
selector:
role: web-frontend
Service
A Kubernetes Service (SVC) is an abstraction which defines a logical set of
Pods and a policy by which to access them - sometimes called a micro-service.
http://kubernetes.io/docs/user-guide/services/
27. Service
A Kubernetes Service (SVC) is an abstraction which defines a logical set of
Pods and a policy by which to access them - sometimes called a micro-service.
http://kubernetes.io/docs/user-guide/services/
# kubectl create -f web-frontend-svc.yaml
You have exposed your service on an external port on all nodes
in your cluster. If you want to expose this service to the external
internet, you may need to set up firewall rules for the service
port(s) (tcp:31468) to serve traffic.
# kubectl describe svc web-frontend
Name: web-frontend
Namespace: default
Labels: <none>
Selector: role=web-frontend
Type: NodePort
IP: 10.254.101.55
Port: <unnamed> 80/TCP
NodePort: <unnamed> 31468/TCP
Endpoints: 172.16.50.3:80,172.16.82.3:80
28. # kubectl scale --replicas=10 rc/web-frontend
replicationcontroller "web-frontend" scaled
# kubectl describe rc web-frontend
Name: web-frontend
Namespace: default
Image(s): php:5.5-apache
Selector: role=web-frontend
Labels: role=web-frontend
Replicas: 10 current / 10 desired
Pods Status: 10 Running / 0 Waiting / 0 Succeeded / 0 Failed
Volumes:
nfs:
Type:PersistentVolumeClaim (a reference to a PersistentVolumeClaim in the same namespace)
ClaimName: web-storage-claim
ReadOnly: true
[...]
Set a new size for a Replication Controller
http://kubernetes.io/docs/user-guide/kubectl/kubectl_scale/
Manual Scaling
29. HorizontalPodAutoscaler
HorizontalPodAutoscaler (HPA) allows to automatically scale the number of
pods in a replication controller, deployment or replica set based on observed
CPU utilization.
apiVersion: extensions/v1beta1
kind: HorizontalPodAutoscaler
metadata:
name: web-frontend
spec:
cpuUtilization:
targetPercentage: 50
maxReplicas: 10
minReplicas: 3
scaleRef:
kind: ReplicationController
name: web-frontend
subresource: scale
http://kubernetes.io/docs/user-guide/horizontal-pod-autoscaling/
30. HorizontalPodAutoscaler
HorizontalPodAutoscaler (HPA) allows to automatically scale the number of
pods in a replication controller, deployment or replica set based on observed
CPU utilization.
http://kubernetes.io/docs/user-guide/horizontal-pod-autoscaling/
# kubectl create -f web-frontend-hpa.yaml
horizontalpodautoscaler "web-frontend" created
# kubectl describe hpa web-frontend
Name: web-frontend
Namespace: default
Labels: <none>
CreationTimestamp: Thu, 14 Apr 2016 10:06:05 -0400
Reference: ReplicationController/web-
frontend/scale
Target CPU utilization: 50%
Current CPU utilization: <not available>
Min replicas: 3
Max replicas: 10
ReplicationController pods:2 current / 2 desired
