Docker orchestration voxxed days berlin 2016

Docker Orchestration
An Overview
Dr. Halil-Cem Gürsoy
@hgutwit
adesso AG - Dortmund, Germany

About myself…
► Principal Architect @ adesso AG, DE
► since more than 15 years in commercial 
software development
> before that, wrote science software 
► Large distributed enterprise systems
► Persistence, build’n deployment
2

“It is not the
strongest or the
most intelligent who
will survive but
those who can best
manage change.”
C. Darwin

Why Docker orchestration?
4
https://www.flickr.com/photos/matijagrguric/4437187539

5

► Containers has to talk to each over
► Think about a microservice architecture with dozens of
containers
► And they have to find each other
6
www.flickr.com/photos/matijagrguric/4437187539/
www.flickr.com/photos/matijagrguric/4437187539/ www.flickr.com/photos/matijagrguric/4437187539/

Poor mans orchestration with Docker link
► The easiest way was to use Docker link functionality
docker run –d --name db42 dbimg
docker run –d --name app42 --link db42:db appimg
► But that about multiple Docker hosts?
7

Poor mans orchestration with Docker link
8

https://www.flickr.com/photos/zunami/3780220955/

You can implement your own network…
10
http://www.opencontrail.org/wp-content/uploads/2014/09/opencontrail-docker-figure-1.png

https://www.flickr.com/photos/r_rose/102766969/

Build-in solutions - Docker Machine
► Docker Machine as the lowest building block
► only needed for development environments
► in production you’ll provision with other tools…
docker-machine create  
--driver digitalocean  
--digitalocean-access-token=$DIGITAL_OCEAN_TOKEN  
machine-1 
► You can create as many docker hosts as you want (and
have resources / money)
12

Docker orchestration voxxed days berlin 2016

Docker Swarm
► Docker Swarm is the ‚built-in’ clustering solution
► Supports new overlay network (based on VXLAN)
► Easy to set up with Docker Machine
docker-machine create --driver digitalocean  
--swarm --swarm-master  
--swarm-discovery token://$SWARM_ID  
swarm-master 
docker-machine create --driver digitalocean  
--swarm --swarm-discovery token://$SWARM_ID  
swarm-node-01
14

Docker Swarm - Manager
► Manager needs a Discovery Service to handle nodes
> Consul, etcd, Zookeeper, static files, IP ranges etc.
► Acts like a proxy, implementing the Docker Remote API
► Manages distribution of starting containers to nodes
> Depending on CPU and / or memory constraints
> … labels on Docker hosts
> … affinity to ressources and other containers
15

Docker Swarm - Scheduler strategies
► Scheduler ranks nodes while starting containers
► Different strategies
> binpack
> spread (default)
> random
► binpack and spread depend on current node utilization
16

Docker Swarm - Filters
► The scheduler has various filters to choose a node
> Constraint - a label on a Docker host
> Affinity - be together with a specific resource
> Port - is a specific port free?
> Dependency - together with another container
> Health
$ docker daemon --label storage=ssd
$ docker run -d -P -e constraint:storage==ssd ...
17

Docker Swarm - Missing stuff
► Swarm is simply the low level clustering for Docker nodes
► Self healing / resilience
> restarting containers on other nodes on node failure
> comes with 1.1.0 (already included in RC, experimental)
► Moving containers during runtime
> only with 3rd party tools
► Auto-Scaling, may come with Docker Compose
► Monitoring
18

Taking a little trip: Overlay Network
► Allows communication between containers in the cluster
► Using libnetwork, based on vxlan
> uses plugin system: overlay, weave
► For overlay driver you need min Kernel 3.16
> have to be careful on Ubuntu
► Needs a K/V store like etcd, Zookeeper or Consul
► All containers attached to that network get an entry in 
/etc/hosts on all containers
19

Another trip: Discovery Services
► You may need a Discovery Service in a Microservice
architecture regardless of Docker
► ‚Did I need a Discovery Service even then I’ve a overlay
network?‘
> YES, you do!
> beyond ‚Hello World’ examples the challenge is the same
to find services in a large scale distributed system
20

Service Discovery
► Consider using registrator
> registers starting containers in Consul, etcd or SkyDNS
> but doesn’t work currently with overlay network
► Inject needed data into your container using something like
consul-template / envconsul, confd, Spring Cloud, …
21

http://github.com/hcguersoy/swarm-elastic-demo
https://www.flickr.com/photos/schwaber/3238977103

Docker Compose
► Former ‚Fig‘ project, now part of Docker’s core ecosystem
► implemented in Python
► „Compose is a tool for defining and running multi-container
applications with Docker.“
► Can be used to set up complex environments
► Using Docker Swarm as the low level clustering of the
Docker Hosts, Compose is for the orchestration of the
containers
23

Docker Compose
► Configuration using YAML:
myapp:
build: .
ports:
- "8080:8080"
volumes:
- ./conf:/etc/myapp/conf
links:
- mysql
mysql:
image: mysql
► Links supported only on single host system
► But latest version supports overlay network (experimental)
24

Docker Compose
► a more complex example: Kubernetes on top of Swarm
etcd:
image: gcr.io/google_containers/etcd:2.0.13
container_name: etcd
command: ['/usr/local/bin/etcd', '--bind-addr=0.0.0.0:4001', '--data-dir=/var/etcd/
data']
apiserver:
image: gcr.io/google_containers/hyperkube:v1.0.7
container_name: apiserver
ports:
- "8080"
command: ["/hyperkube", "apiserver", "--service-cluster-ip-range=172.17.17.1/24", "--
address=0.0.0.0", "--etcd_servers=http://etcd:4001", "--cluster_name=kubernetes", "--
v=2"]
controller:
image: gcr.io/google_containers/hyperkube:v1.0.7
command: ["/hyperkube", "controller-manager", "--address=0.0.0.0", "--master=http://
apiserver:8080", "--v=2"]
environment:
- "affinity:container==*apiserver*"
...
Source 'Swarm Frontend’ (https://github.com/docker/swarm-frontends)
25

https://www.flickr.com/photos/tomwahlin/with/3167491292/

Kubernetes
► Created by Google
► build to manage ‚oceans of user containers‘
► Cloud agnostics, supports different cloud and virtualization
platform
> but results in different installation directives
> easy playground setup using Docker Compose (s. above)
on top of Swarm
27

Kubernetes - Key components
► Kubernetes consists of different key components
► Master Server
> the main management system
> is build up using different tools
– etcd
– API Server, providing REST interface
– Controller Server
– Scheduler Server
28

► Minion
> the work units in Kubernetes, executing the commands
coming from the master server
> runs a Docker daemon
> Kubelete service
> Proxy Service
> cAdvisior
29

► Pod
> the smallest logical unit in Kubernetes
> containers belonging together are defined in a Pod - a
logical collection
> All containers of a Pod run on one Minion ( = one Host)
30

Kubernetes - System Arcitecture
31
Source: http://releases.k8s.io/release-1.1/docs/design/architecture.md

Kubernetes
► Build in resource monitoring, fail over  
and rescheduling
► Uses different API then Docker
► Uses different configuration files as Compose
► Complex architecture
► IMHO not very usable for ‚small‘ setups due to the overhead
> but may be very useful then you get really large
32

Spotify Helios
► Builds a cluster spanning multiple Docker hosts
► deploys and manages containers in this cluster
► Developed before rise of Swarm but does similar stuff
> …and still actively developed
> used by Spotify in production
► But missing something like Compose above this low layer
33

Apache Mesos / Marathon
► Apache Mesos is a ‚kernel‘ to set up large scale distributed
systems
> e.g. large Hadoop Clusters
► Comes with an 0.20.0 with Docker Containerizer
> still some limitations
► May be used together with Marathon
> e.g. supplies rescheduler
34

…and much more
► There is a whole bunch of more orchestration solutions
> Crane
> CoresOS / Fleetd
> RancherOS
> …
► …and much more hosted solutions
> Giantswarm (still alpha), Amazon ECS, Google CE
35

Summing up
► Docker Swarm together with Compose is a lightweight way
to orchestrate containers
> but need some maturity and additional features / services
► Kubernetes is the opposite, heavy weight solution
> good for large scenarios, but oversized for small setups
> breaks with Docker CLI
► Mesos / Marathon
> useful for some scenarios, breaks with Docker CLI
36

http://www.flickr.com/photos/an_untrained_eye/6630719431

http://www.flickr.com/photos/tcmorgan/7372944070
halil-cem.guersoy@adesso.de
https://twitter.com/hgutwit

Docker orchestration voxxed days berlin 2016

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Docker orchestration voxxed days berlin 2016 (20)

Recently uploaded (20)

Docker orchestration voxxed days berlin 2016