Docker Tutorial - Mars 2019

Intro Containers I/O Images Builder Security Ecosystem Future
Docker Tutorial
Anthony Baire
Université de Rennes 1 / UMR IRISA
March 28, 2019
This tutorial is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 France License
1 / 83
Summary
1. Introduction
2. Managing docker containers
3. Inputs/Outputs
4. Managing docker images
5. Building docker images
6. Security considerations
7. The ecosystem & the future
2 / 83
Part 1.
Introduction
3 / 83
What is Docker (1/3)
“Docker is an open platform for developers and sysadmins to build, ship, and run distributed applications.
Consisting of Docker Engine, a portable, lightweight runtime and packaging tool, and Docker Hub, a cloud service
for sharing applications and automating workflows, Docker enables apps to be quickly assembled from components
and eliminates the friction between development, QA, and production environments. As a result, IT can ship faster
and run the same app, unchanged, on laptops, data center VMs, and any cloud.”
source: https://www.docker.com/whatisdocker/
4 / 83
• a container manager
• lightweight virtualisation
(host and guest systems share the same kernel)
• based on linux namespaces and cgroups
• massively copy-on-write
• immutable images
• instant deployment
• suitable for micro-services (one process, one container)
→ immutable architecture
5 / 83
• a build system
• images may be build from sources
• using a simple DSL (Dockerfile)
• a set of REST APIs

• Engine API (control the docker engine)
• Plugin API (extend the engine → network, storage,
authorisation)
• Registry API (publish/download images)
• Swarm API (manage a clusted of docker machines)
6 / 83
How Docker helps?
• normalisation: same environment (container image) for

• development
• jobs on the computing grid
• continuous integration
• peer review
• demonstrations, tutorials
• technology transfer
• archival (ever tried to reuse old codes)

• source → Dockerfile = recipe to rebuild the env from scratch
• binary → docker image = immutable snapshot of the software
with its runtime environment
→ can be rerun it at any time later
7 / 83
In practice
A docker image is an immutable snapshot of the filesystem
A docker container is
• a temporary file system
• layered over an immutable fs (docker image)
• fully writable (copy-on-write1 )
• dropped at container’s end of life (unless a commit is made)
• a network stack
• with its own private address (by defaut in 172.17.x.x)
• a process group
• one main process launched inside the container
• all sub-process SIGKILLed when the main process exits
1
several possible methods: overlayfs (default), btrfs, lvm, zfs, aufs
8 / 83
Installation
https://docs.docker.com/engine/installation/
Native installation:
• requires linux kernel >3.8
Docker Machine:
• a command for provisionning an managing docker nodes
deployed:
• in a local VM (virtualbox)
• remotely (many cloud API supported)
9 / 83
Part 2.
Managing containers
• create/start/stop/remove containers
• inspect containers
• interact, commit new images
10 / 83
Lifecycle of a docker container
11 / 83
Container management commands

command description
docker create image [ command ] create the container
docker run image [ command ] = create + start
docker rename container new name rename the container
docker update container update the container config
docker start container. . . start the container
docker stop container. . . graceful2 stop
docker kill container. . . kill (SIGKILL) the container
docker restart container. . . = stop + start
docker pause container. . . suspend the container
docker unpause container. . . resume the container
docker rm [ -f3 ] container. . . destroy the container
2
send SIGTERM to the main process + SIGKILL 10 seconds later
3
-f allows removing running containers (= docker kill + docker rm)
12 / 83
Notes about the container lifecycle
• the container filesystem is created in docker create and

dropped in docker rm
• it is persistent across stop/start
• the container configuration is mostly static

• config is set in create/run
• docker update may change only a few parameters
(eg: cpu/ram/blkio allocations)
• changing other parameters require destroying and re-creating
the container
• other commands are rather basic
13 / 83

Usage: docker create [OPTIONS] IMAGE [COMMAND] [ARG...] Usage: docker start [OPTIONS] CONTAINER [CONTAINER...]
Create a new container Start one or more stopped containers
-a, --attach=[] Attach to STDIN, STDOUT or STDERR -a, --attach=false Attach STDOUT/STDERR and forward signals
--add-host=[] Add a custom host-to-IP mapping (host:ip) --help=false Print usage
--blkio-weight=0 Block IO (relative weight), between 10 and 1000 -i, --interactive=false Attach container's STDIN
--cpu-shares=0 CPU shares (relative weight)
--cap-add=[] Add Linux capabilities
--cap-drop=[] Drop Linux capabilities
--cgroup-parent= Optional parent cgroup for the container
--cidfile= Write the container ID to the file Usage: docker stop [OPTIONS] CONTAINER [CONTAINER...]
--cpu-period=0 Limit CPU CFS (Completely Fair Scheduler) period
--cpu-quota=0 Limit CPU CFS (Completely Fair Scheduler) quota Stop a running container.
--cpuset-cpus= CPUs in which to allow execution (0-3, 0,1) Sending SIGTERM and then SIGKILL after a grace period
--cpuset-mems= MEMs in which to allow execution (0-3, 0,1)
--device=[] Add a host device to the container --help=false Print usage
--disable-content-trust=true Skip image verification -t, --time=10 Seconds to wait for stop before killing it
--dns=[] Set custom DNS servers
--dns-opt=[] Set DNS options
--dns-search=[] Set custom DNS search domains
-e, --env=[] Set environment variables
--entrypoint= Overwrite the default ENTRYPOINT of the image Usage: docker restart [OPTIONS] CONTAINER [CONTAINER...]
--env-file=[] Read in a file of environment variables
--expose=[] Expose a port or a range of ports Restart a container
--group-add=[] Add additional groups to join
-h, --hostname= Container host name --help=false Print usage
--help=false Print usage -t, --time=10 Seconds to wait for stop before killing the container
-i, --interactive=false Keep STDIN open even if not attached
--ipc= IPC namespace to use
--kernel-memory= Kernel memory limit
-l, --label=[] Set meta data on a container
--label-file=[] Read in a line delimited file of labels Usage: docker kill [OPTIONS] CONTAINER [CONTAINER...]
--link=[] Add link to another container
--log-driver= Logging driver for container Kill a running container
--log-opt=[] Log driver options
--lxc-conf=[] Add custom lxc options --help=false Print usage
-m, --memory= Memory limit -s, --signal=KILL Signal to send to the container
--mac-address= Container MAC address (e.g. 92:d0:c6:0a:29:33)
--memory-reservation= Memory soft limit
--memory-swap= Total memory (memory + swap), '-1' to disable swap
--memory-swappiness=-1 Tuning container memory swappiness (0 to 100)
--name= Assign a name to the container Usage: docker rm [OPTIONS] CONTAINER [CONTAINER...]
--net=default Set the Network for the container
--oom-kill-disable=false Disable OOM Killer Remove one or more containers
-P, --publish-all=false Publish all exposed ports to random ports
-p, --publish=[] Publish a container's port(s) to the host -f, --force=false Force the removal of a running container (uses SIGKILL)
--pid= PID namespace to use --help=false Print usage
--privileged=false Give extended privileges to this container -l, --link=false Remove the specified link
--read-only=false Mount the container's root filesystem as read only -v, --volumes=false Remove the volumes associated with the container
--restart=no Restart policy to apply when a container exits
--security-opt=[] Security Options
--stop-signal=SIGTERM Signal to stop a container, SIGTERM by default
-t, --tty=false Allocate a pseudo-TTY
-u, --user= Username or UID (format: <name|uid>[:<group|gid>]) Usage: docker pause [OPTIONS] CONTAINER [CONTAINER...]
--ulimit=[] Ulimit options
--uts= UTS namespace to use Pause all processes within a container
-v, --volume=[] Bind mount a volume
--volume-driver= Optional volume driver for the container --help=false Print usage
--volumes-from=[] Mount volumes from the specified container(s)
-w, --workdir= Working directory inside the container 14 / 83

docker run — Run a container

https://docs.docker.com/reference/run/
docker run [ options ] image [ arg0 arg1...]
→ create a container and start it
• the container filesystem is initialised from image image

• arg0..argN is the command run inside the container (as PID 1)

$ docker run debian /bin/hostname
f0d0720bd373
$ docker run debian date +%H:%M:%S
17:10:13
$ docker run debian true ; echo $?
0
$ docker run debian false ; echo $?
1

15 / 83
docker run — Foreground mode vs. Detached mode
• Foreground mode is the default

• stdout and stderr are redirected to the terminal
• docker run propagates the exit code of the main process
• With -d, the container is run in detached mode:

• displays the ID of the container
• returns immediately

$ docker run debian date
Tue Jan 20 17:32:07 UTC 2015
$ docker run -d debian date
4cbdefb3d3e1331ccf7783b32b47774fefca426e03a2005d69549f3ff06b9306
$ docker logs 4cbdef
Tue Jan 20 17:32:16 UTC 2015

16 / 83
docker run — TTY allocation

Use -t to allocate a pseudo-terminal for the container
→ without a tty

$ docker run debian ls
bin
boot
dev
...
$ docker run debian bash
$

→ with a tty (-t)

$ docker run -t debian ls
bin dev home lib64 mnt proc run selinux sys usr
boot etc lib media opt root sbin srv tmp var
$ docker run -t debian bash
root@10d90c09d9ac:/#

17 / 83
docker run — interactive mode

• By default containers are non-interactive
• stdin is closed immediately
• terminal signals are not forwarded4

$ docker run -t debian bash
root@6fecc2e8ab22:/# date
ˆC
$

• With -i the container runs interactively
• stdin is usable
• terminal signals are forwarded to the container

$ docker run -t -i debian bash
root@78ff08f46cdb:/# date
Tue Jan 20 17:52:01 UTC 2015
root@78ff08f46cdb:/# ˆC
root@78ff08f46cdb:/#

4
ˆC only detaches the terminal, the container keeps running in background
18 / 83
docker run — override defaults (1/2)
user (-u)

$ docker run debian whoami
root
$ docker run -u nobody debian whoami
nobody

working directory (-w)

$ docker run debian pwd
/
$ docker run -w /opt debian pwd
/opt

19 / 83
docker run — override defaults (2/2)
environment variables (-e)

$ docker run debian sh -c 'echo $FOO $BAR'
$ docker run -e FOO=foo -e BAR=bar debian sh -c 'echo $FOO $BAR'

foo bar

hostname (-h)

$ docker run debian hostname
830e47237187
$ docker run -h my-nice-container debian hostname
my-nice-hostname

20 / 83
docker run — set the container name

--name assigns a name for the container
(by default a random name is generated)

$ docker run -d -t debian
da005df0d3aca345323e373e1239216434c05d01699b048c5ff277dd691ad535
$ docker run -d -t --name blahblah debian
0bd3cb464ff68eaf9fc43f0241911eb207fefd9c1341a0850e8804b7445ccd21
$ docker ps
CONTAINER ID IMAGE COMMAND CREATED .. NAMES
0bd3cb464ff6 debian:7.5 "/bin/bash" 6 seconds ago blahblah
da005df0d3ac debian:7.5 "/bin/bash" About a minute ago drunk_darwin
$ docker stop blahblah drunk_darwin

Note: Names must be unique

$ docker run --name blahblah debian true
2015/01/20 19:31:21 Error response from daemon: Conflict, The name blahblah is already assigned
to 0bd3cb464ff6. You have to delete (or rename) that container to be able to assign blahblah to a
container again.

21 / 83
docker run — autoremove

By default the container still exists after command exit

$ docker run --name date-ctr debian date
Tue Jan 20 18:38:21 UTC 2015
$ docker start date-ctr
date-ctr
$ docker logs date-ctr
Tue Jan 20 18:38:21 UTC 2015
Tue Jan 20 18:38:29 UTC 2015
$ docker rm date-ctr
date-ctr
$ docker start date-ctr
Error response from daemon: No such container: date-ctr
2015/01/20 19:39:27 Error: failed to start one or more containers

With --rm the container is automatically removed after exit

$ docker run --rm --name date-ctr debian date
Tue Jan 20 18:41:49 UTC 2015
$ docker rm date-ctr
Error response from daemon: No such container: date-ctr
2015/01/20 19:41:53 Error: failed to remove one or more containers

22 / 83
Common rm idioms
Launch an throwaway container for debugging/testing purpose

$ docker run --rm -t -i debian
root@4b71c9a39326:/#

Remove all zombie containers

$ docker ps -a
CONTAINER ID IMAGE COMMAND CREATED STATUS
2b291251a415 debian:7.5 "hostname" About a minute ago Exited (0) About a mi
6d36a2f07e18 debian:7.5 "false" 2 minutes ago Exited (1) 2 minutes
0f563f110328 debian:7.5 "true" 2 minutes ago Exited (0) 2 minutes
4b57d0327a20 debian:7.5 "uname -a" 5 minutes ago Exited (0) 5 minutes
$ docker container prune
WARNING! This will remove all stopped containers.
Are you sure you want to continue? [y/N] y
Deleted Containers:
2b291251a415
6d36a2f07e18
0f563f110328
4b57d0327a20

23 / 83
Inspecting the container

command description
docker ps list running containers
docker ps -a list all containers
docker logs [ -f5 ] container show the container output
(stdout+stderr)
docker top container [ ps options ] list the processes running
inside the containers6
docker stats [ container ] display live usage statistics7
docker diff container show the differences with
the image (modified files)
docker port container list port mappings
docker inspect container. . . show low-level infos
(in json format)
5
with -f, docker logs follows the output (à la tail -f)
6
docker top is the equivalent of the ps command in unix
7
docker stats is the equivalent of the top command in unix
24 / 83
Interacting with the container
command description
docker attach container attach to a running container
(stdin/stdout/stderr)
docker cp container:path hostpath|- copy files from the container
docker cp hostpath|- container:path copy files into the container
docker export container export the content of
the container (tar archive)
docker exec container args. . . run a command in an existing
container (useful for debugging)
docker wait container wait until the container terminates
and return the exit code
docker commit container image commit a new docker image
(snapshot of the container)
25 / 83
docker commit example

$ docker run --name my-container -t -i debian
root@3b397d383faf:/# cat >> /etc/bash.bashrc <<EOF
> echo 'hello!'
> EOF
root@3b397d383faf:/# exit
$ docker start --attach my-container
my-container
hello!
root@3b397d383faf:/# exit
$ docker diff my-container
C /etc
C /etc/bash.bashrc
A /.bash_history
C /tmp
$ docker commit my-container hello
a57e91bc3b0f5f72641f19cab85a7f3f860a1e5e9629439007c39fd76f37c5dd
$ docker rm my-container
my-container
$ docker run --rm -t -i hello
hello!
root@386ed3934b44:/# exit
$ docker images -t
511136ea3c5a Virtual Size: 0 B
af6bdc397692 Virtual Size: 115 MB
667250f9a437 Virtual Size: 115 MB Tags: debian:wheezy, debian:latest
a57e91bc3b0f Virtual Size: 115 MB Tags: hello:latest

26 / 83
Part 3.
Inputs/Outputs
• Data volumes (persistent data)
• mounted from the host filesystem
• named volumes (interal + volume plugins)
• Devices
• Links
• Publishing ports (NAT)
27 / 83
docker run — mount external volumes
docker run -v /hostpath:/containerpath[:ro] ...
-v mounts the location /hostpath from the host filesystem at the

location /containerpath inside the container
With the “:ro” suffix, the mount is read-only
Purposes:
• store persistent data outside the container

• provide inputs: data, config files, . . . (read-only mode)
• inter-process communicattion (unix sockets, named pipes)
28 / 83
mount examples (1/2)

Persistent data

$ docker run --rm -t -i -v /tmp/persistent:/persistent debian
root@0aeedfeb7bf9:/# echo "blahblah" >/persistent/foo
root@0aeedfeb7bf9:/# exit
$ cat /tmp/persistent/foo
blahblah
$ docker run --rm -t -i -v /tmp/persistent:/persistent debian
root@6c8ed008c041:/# cat /persistent/foo
blahblah

Inputs (read-only volume)

$ mkdir /tmp/inputs
$ echo hello > /tmp/inputs/bar
$ docker run --rm -t -i -v /tmp/inputs:/inputs:ro debian
root@05168a0eb322:/# cat /inputs/bar
hello
root@05168a0eb322:/# touch /inputs/foo
touch: cannot touch `/inputs/foo': Read-only file system

29 / 83
mount examples (2/2)
Named pipe

$ mkfifo /tmp/fifo
$ docker run -d -v /tmp/fifo:/fifo debian sh -c 'echo blah blah> /fifo'
ff0e44c25e10d516ce947eae9168060ee25c2a906f62d63d9c26a154b6415939
$ cat /tmp/fifo
blah blah

Unix socket

$ docker run --rm -t -i -v /dev/log:/dev/log debian
root@56ec518d3d4e:/# logger blah blah blah
root@56ec518d3d4e:/# exit
$ sudo tail /var/log/messages | grep logger
Jan 21 08:07:59 halfoat logger: blah blah blah

30 / 83
docker run — named volumes

Named volumes
• stored inside /var/lib/docker

• lifecycle managed with the docker volume command
• plugin API to provide shared storage over a cluster/cloud8

$ docker volume create my-volume
my-volume
$ docker volume ls
DRIVER VOLUME NAME
local my-volume
$ docker run --rm -t -i -v my-volume:/vol busybox
/ # echo foo > /vol/bar
/ # ˆD
$ docker volume inspect my-volume|grep Mountpoint
"Mountpoint": "/var/lib/docker/volumes/my-volume/_data",
$ docker run --rm -t -i -v my-volume:/vol busybox cat /vol/bar
foo
$ docker volume rm my-volume
my-volume

8
https://docs.docker.com/engine/tutorials/dockervolumes/
31 / 83
docker run — grant access to a device

By default devices are not usable inside the container

$ docker run --rm debian fdisk -l /dev/sda
root@dcba37b0c0bd:/# fdisk -l /dev/sda
fdisk: cannot open /dev/sda: No such file or directory
$ docker run --rm debian sh -c 'mknod /dev/sda b 8 0 && fdisk -l /dev/sda'

fdisk: cannot open /dev/sda: Operation not permitted
$ docker run --rm -v /dev/sda:/dev/sda debian fdisk -l /dev/sda

fdisk: cannot open /dev/sda: Operation not permitted

They can be whitelisted with --device
docker run --device /hostpath[:/containerpath ] ...

$ docker run --rm --device /dev/sda debian fdisk -l /dev/sda
Disk /dev/sda: 250.1 GB, 250059350016 bytes
...

32 / 83
docker run — inter-container links (legacy links9 )

Containers cannot be assigned a static IP address (by design)
→ service discovery is a must
Docker “links” are the most basic way to discover a service
docker run --link ctr:alias . . .
→ container ctr will be known as alias inside the new container

$ docker run --name my-server debian sh -c 'hostname -i && sleep 500' &
172.17.0.4
$ docker run --rm -t -i --link my-server:srv debian

root@d752180421cc:/# ping srv
PING srv (172.17.0.4): 56 data bytes
64 bytes from 172.17.0.4: icmp_seq=0 ttl=64 time=0.195 ms

9
since v1.9.0, links are superseded by user-defined networks
33 / 83
Legacy links
deprecated feature
34 / 83
User-defined networks (since v1.9.0)

• by default new containers are connected to the main network
(named “bridge”, 172.17.0.0/16)
• the user can create additional networks:

docker network create NETWORK
• newly created containers are connected to one network:

docker run --net=NETWORK
• container may be dynamically attached/detached to any

network:
docker network connect NETWORK CONTAINER
docker network disconnect NETWORK CONTAINER
• networks are isolated from each other, communications is

possible by attaching a container to multiple networks
35 / 83
User-defined networks example
36 / 83
36 / 83
36 / 83
36 / 83
36 / 83
36 / 83
docker run — publish a TCP port
Containers are deployed in a private network, they are not

reachable from the outside (unless a redirection is set up)
docker run -p [ipaddr:]hostport:containerport
→ redirect incoming connections to the TCP port hostport of the

host to the TCP port containerport of the container
The listening socket binds to 0.0.0.0 (all interfaces) by default or
to ipaddr if given
37 / 83
publish example
38 / 83
publish example
bind to all host addresses

$ docker run -d -p 80:80 nginx
52c9105e1520980d49ed00ecf5f0ca694d177d77ac9d003b9c0b840db9a70d62
$ wget -nv http://localhost/

2016-01-12 18:32:52 URL:http://localhost/ [612/612] -> "index.html" [1]
$ wget -nv http://172.17.42.1/
2016-01-12 18:33:14 URL:http://172.17.42.1/ [612/612] -> "index.html" [1]

bind to 127.0.0.1

$ docker run -d -p 127.0.0.1:80:80 nginx
4541b43313b51d50c4dc2722e741df6364c5ff50ab81b828456ca55c829e732c
$ wget -nv http://localhost/

2016-01-12 18:37:10 URL:http://localhost/ [612/612] -> "index.html.1" [1]
$ wget http://172.17.42.1/
--2016-01-12 18:38:32-- http://172.17.42.1/
Connecting to 172.17.42.1:80... failed: Connection refused.

39 / 83
The whole picture
40 / 83
The whole picture
40 / 83
Part 4.
Managing docker images
41 / 83
Docker images
A docker image is a snapshot of the filesystem + some metadata
• immutable
• copy-on-write storage
• for instantiating containers
• for creating new versions of the image (multiple layers)
• identified by a unique hex ID (hashed from the image content)
• may be tagged10 with a human-friendly name

eg: debian:wheezy debian:jessie debian:latest
10
possibly multiple times
42 / 83
Image management commands
command description
docker images list all local images
docker history image show the image history
(list of ancestors)
docker inspect image. . . show low-level infos
(in json format)
docker tag image tag tag an image
docker commit container image create an image
(from a container)
docker import url|- [tag] create an image
(from a tarball)
docker rmi image. . . delete images
43 / 83
Example: images & containers
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
44 / 83
Images vs. Layers
45 / 83
Image tags
A docker tag is made of two parts: “REPOSITORY:TAG”
The TAG part identifies the version of the image. If not provided,
the default is “:latest”

$ docker images
REPOSITORY TAG IMAGE ID CREATED VIRTUAL SIZE
debian 8 835c4d274060 2 weeks ago 122.6 MB
debian 8.0 835c4d274060 2 weeks ago 122.6 MB
debian jessie 835c4d274060 2 weeks ago 122.6 MB
debian rc-buggy 350a74df81b1 7 months ago 159.9 MB
debian experimental 36d6c9c7df4c 7 months ago 159.9 MB
debian 6.0.9 3b36e4176538 7 months ago 112.4 MB
debian squeeze 3b36e4176538 7 months ago 112.4 MB
debian wheezy 667250f9a437 7 months ago 115 MB
debian latest 667250f9a437 7 months ago 115 MB
debian 7.5 667250f9a437 7 months ago 115 MB
debian unstable 24a4621560e4 7 months ago 123.6 MB
debian testing 7f5d8ca9fdcf 7 months ago 121.8 MB
debian stable caa04aa09d69 7 months ago 115 MB
debian sid f3d4759f77a7 7 months ago 123.6 MB
debian 7.4 e565fbbc6033 9 months ago 115 MB
debian 7.3 b5fe16f2ccba 11 months ago 117.8 MB

46 / 83
Tagging conventions (1/2)
Local tags may have arbitrary names, however the docker push
and docker pull commands expect some conventions
The REPOSITORY identifies the origin of the image, it may be:
• a name (eg: debian)

→ refers to a repository on the official registry
→ https://store.docker.com/
• a hostname+name (eg: some.server.com/repo)

→ refers to an arbitrary server supporting the registry API
→ https://docs.docker.com/reference/api/registry_api/
47 / 83
Tagging conventions (2/2)
Use slashes to delimit namespaces (for subprojects):

image name description
debian (semi-)official debian images
fedora official fedora images
fedora/apache apache images provided
by the fedora project
fedora/couchdb couchdb images provided
by the fedora project
48 / 83
Image transfer commands
Using the registry API

docker pull repo[:tag]. . . pull an image/repo from a registry
docker push repo[:tag]. . . push an image/repo from a registry
docker search text search an image on the official registry
docker login . . . login to a registry
docker logout . . . logout from a registry
Manual transfer
docker save repo[:tag]. . . export an image/repo as a tarbal
docker load load images from a tarball
docker-ssh11 . . . proposed script to transfer images
between two daemons over ssh
11
https://github.com/a-ba/docker-utils/
49 / 83
Transferring images
50 / 83
Part 5.
Docker builder
51 / 83
What is the Docker builder ?
Docker’s builder relies on
• a DSL describing how to build an image

• a cache for storing previous builds and have quick iterations
The builder input is a context, i.e. a directory containing:
• a file named Dockerfile which describe how to build the

container
• possibly other files to be used during the build
52 / 83
Build an image
docker build [ -t tag ] path
→ build an image from the context located at path and optionally

tag it as tag
The command:
1. makes a tarball from the content12 of path

2. uploads the tarball to the docker daemon which will:
2.1 execute the content of Dockerfile, committing an
intermediate image after each command
2.2 (if requested) tag the final image as tag
12
unwanted files may be excluded if they match patterns listed in
.dockerignore
53 / 83
Dockerfile example

# base image: last debian release
FROM debian:wheezy
# install the latest upgrades

RUN apt-get update && apt-get -y dist-upgrade
# install nginx
RUN apt-get -y install nginx
# set the default container command

# −> run nginx in the foreground
CMD ["nginx", "-g", "daemon off;"]
# Tell the docker engine that there will be somenthing listening on the tcp port 80
EXPOSE 80

54 / 83
Dockerfile format
https://docs.docker.com/reference/builder/
• comments start with “#”
• commands fit on a single line

(possibly continuated with \)
• first command must be a FROM

(indicates the parent image or scratch to start from scratch)
55 / 83
Builder instructions (1/3)

Instructions affecting the image filesystem
instruction description
FROM image|scratch base image for the build
COPY path dst copy path from the context
into the container at location dst
ADD src dst same as COPY but untar archives
and accepts http urls
RUN command run an arbitrary command inside
the container
Note: commands may be expressed as a list (exec) or a string (shell)

# exec form
RUN ["apt-get", "update"]
# shell form
RUN apt-get update # equivalent to: RUN [”/bin/sh”, ”−c”, ”apt−get update”]

56 / 83

Instructions setting the default container config14
CMD command command run inside the container
ENTRYPOINT command entrypoint13
USER name[:group] user running the command
WORKDIR path working directory
ENV name="value". . . environment variables
STOPSIGNAL signal signal to be sent to terminate the
container(instead of SIGTERM)
HEALTHCHECK CMD command test command to check
if the container works well
EXPOSE port. . . listened TCP/UDP ports
VOLUME path. . . mount-point for external volumes
LABEL name="value". . . arbitrary metadata
13
the ENTRYPOINT is a commmand that wraps the CMD command
14
i.e. the default configuration of containers running this image
57 / 83

Extra instructions
ARG name[=value] build-time variables
ON BUILD instruction instruction run when building
a derived image
• build-time variables are usable anywhere in the Dockerfile

(by variable expansion: $VARNAME) and are tunable at build
time: “docker build --build-arg name=value . . . ”
• instructions prefixed with ONBUILD are not run in this build,

their execution is triggered when building a derived image
58 / 83
Builder cache
Each layer created by the builder is fingerprinted according to:
• the ID of the previous image

• the command and its arguments
• the content of the imported files (for ADD and COPY)
RUN’s side-effects are not fingerprinted
When rebuilding an image docker will reuse a previous image if its

fingerprint is the same
59 / 83
Good practices15 for docker files
• use stable base images (eg. debian:jessie)
• run the app as PID 1 inside the container (to be killable)

→ write CMD ["app", "arg"] instead of CMD app arg
• standardise the config, but allow the admin to override it with

env variables or additional config files
(eg. ENV MYSQL HOST="mysql")
15
see also https://docs.docker.com/engine/userguide/eng-image/dockerfile_best-practices/
60 / 83
Multi-stage build (since v17.05)

#======= Stage 1: build the app from sources =======#
FROM debian:stretch AS builder
# update the package lists an install the build dependencies
RUN apt-get -qqy update
RUN apt-get -qqy install gcc make libacme-dev
# install the sources in /opt/src and build them

COPY . /opt/src
RUN cd /opt/src && ./configure && make
# install the files in a tmp dir and make an archive that we can deploy elsewhere
RUN cd /opt/src && make install DESTDIR=/tmp/dst \
&& cd /tmp/dst && tar czvf /tmp/myapp.tgz .
#======= Stage 2: final image ================#

FROM debian:stretch
# update the package lists and install the runtime dependencies
RUN apt-get -qqy update
RUN apt-get -qqy install libacme1.0
# install the app built in stage 1

COPY --from=builder /tmp/myapp.tgz /tmp/
RUN cd / && tar zxf /tmp/myapp.tgz
CMD ["myapp"]

61 / 83
Part 6.
Security
• host/container isolation
• container/container isolation
• other security considerations
62 / 83
Security strategies
Docker containers are not really sandboxed from the host machine.
They talk with the same kernel. You may want to consider
strategies to reduce the risks of privilege escalation.
Container/Host isolation
• run the container with an ordinary user (docker run -u)

• reduce root privileges (capabilities, seccomp, apparmor)
• configure a user namespace
• run the docker engine inside a VM
Container/Container isolation
• disable intercontainer communications (--icc=false)

• isolate containers in different networks
63 / 83
Running containers as normal user
docker run -u USER ...
should be safe, but...
• setuid executables in the docker image

→ should mount /var/lib/docker with ‘-o nosuid’
• setuid executables in external volumes

→ should mount all data volumes with ‘-o nosuid’
• /etc/passwd in the docker image

→ should use numeric ids: (docker run -u UID:GID)
→ not easily enforcable if the image provider is malicious
64 / 83
Reduced root capabilities

• kernel capabilities supported since docker v1.2
• containers use a default set limited to 14 capabilities16 :
AUDIT WRITE CHOWN NET RAW SETPCAP
DAC OVERRIDE FSETID SETGID KILL
NET BIND SERVICE FOWNER SETUID
SYS CHROOT MKNOD SETFCAP
• add additional capabilities: docker run --cap-add=XXXXX ...
• drop unnecessary capabilities: docker run --cap-drop=XXXXX ...
→ should use --cap-drop=all for most containers

$ docker run --rm -t -i debian
root@04223cbb1334:/# ip addr replace 172.17.0.42/16 dev eth0
RTNETLINK answers: Operation not permitted
root@04223cbb1334:/# exit
$ docker run --rm -t -i --cap-add NET_ADMIN debian

root@9bf2a570a6a6:/# ip addr replace 172.17.0.42/16 dev eth0
root@9bf2a570a6a6:/#

16
over the 38 capabilities defined in the kernel (man 7 capabilities)
65 / 83
Reduced syscall whitelist

seccomp-bpf == fine-grained acces control to kernel syscalls
• enabled by default since docker v1.10
• default built-in profile17 whitelists only harmless syscalls18
• alternative configs:
• disable seccomp (--security-opt=seccomp:unconfined)
• provide a customised profile (derived from the default19 )

$ docker run --rm debian date -s 2016-01-01
date: cannot set date: Operation not permitted
$ docker run --rm --cap-add sys_time debian date -s 2016-01-01
$ docker run --rm --security-opt seccomp:unconfined debian date -s 2016-01-01
$ docker run --rm --cap-add sys_time --security-opt seccomp:unconfined debian date -s 2016-01-01
Fri Jan 1 00:00:00 UTC 2016

17
https://docs.docker.com/engine/security/seccomp/
18
harmful means everything that deals with administration (eg: set time) or debugging (eg: ptrace)
19
https://github.com/moby/moby/blob/master/profiles/seccomp/default.json
66 / 83
User namespaces
since docker v1.10 but not enabled by default
• UIDs/GIDs inside the containers mapped to another range

outside the container
• useful for:
• preventing fs-based attacks (eg: root user inside the container
creates a setuid executable in an external volume)
• isolating docker users from each other (one docker daemon for
each user, with uids remapped to different ranges)
• limits (as of v1.10)

• global config only (daemon scope)
• coarse mapping only (hardcoded range: 0..65535)
67 / 83
Docker is not a sandbox !
Even with capabilities+seccomp+user namespaces enabled, you

may still be vulnerable, because the kernel’s attack surface is big
CVE-2019-5736
runc through 1.0-rc6, as used in Docker before 18.09.2 and other products, allows
attackers to overwrite the host runc binary (and consequently obtain host root access)
by leveraging the ability to execute a command as root within one of these types of
containers: (1) a new container with an attacker-controlled image, or (2) an existing
container, to which the attacker previously had write access, that can be attached
with docker exec. This occurs because of file-descriptor mishandling, related to
/proc/self/exe.
68 / 83
Run the docker engine inside a VM
Hypervisors have a smaller attack surface and are much more

mature that containers. Use a VM if you need good isolation!
• either manually-administrated VMs

• either transparently-launched VMs
• on a per-engine basis (docker daemon inside a VM)

docker machine: https://docs.docker.com/machine/overview/
• on a per-container basis (each container in a separate VM)

kata containers: https://katacontainers.io/
runv: https://github.com/hyperhq/runv
gvisor: https://github.com/google/gvisor
69 / 83
Container/Container isolation
• by default all containers can connect to any other container

(located in the same bridge)
• run the daemon with --icc=false

• all communications filtered by default
• whitelist-based access with --link
(only EXPOSEd ports will be whitelisted)
• attach containers to different networks
• by default RAW sockets are enabled (allows ARP spoofing)20

→ use docker run --cap-drop=NET RAW
20
http://lwn.net/Articles/689453
70 / 83
Other security considerations

• images are immutable
→ need a process to apply automatic security upgrades, e.g:
• apply upgrades & commit a new image
• regenerate the image from the Dockerfile
• docker engine control == root on the host machine

• give access to the docker socket only to trusted users
• avoid docker run --privileged (gives full root access)
• avoid the insecure v1 registry API (for push/pull)

→ run the daemon with --insecure-registry=false --disable-legacy-registry
• beware of symlinks in external volumes

eg. ctr1 binds /data, ctr2 binds /data/subdir, if both are malicious and cooperate, ctr1 replaces
/data/subdirwith a symlink to /, then on restart ctr2 has access the whole host filesystem
→ avoid binding subdirectories, prefer using named volumes
71 / 83
Part 7.
Docker Ecosystem
• infrastructure
• docker machine (provisioning)
• docker swarm (clustering)
• swarm mode (clustering)
• underlying projects (moby, containerd, infrakit, ...)
• container deployment & configuration
• docker compose
• image distribution
• docker distribution (registry)
• docker notary (content trust, image signing)
72 / 83
Docker Machine
abstraction for provisionning and using docker hosts
73 / 83
Docker Swarm
manage a cluster of hosts running docker
Docker Inc. folks are misleading: the name
swarm is actually used for two different products:
• docker swarm (or legacy swarm or just swarm)

• early solution (first released in dec 2014)
• standalone server
• superset of the docker engine API
• requires a an external discovery service (eg. etcd, consul)
• network-agnostic (overlay networks to be configured separately)
• the swarm mode

• embedded within the docker engine (since v1.12 in july 2016)
• turnkey cluster (integrated discovery service, distributed,
network aware, encryption by default)
• API break: introduces the service abstration
74 / 83
Docker Compose
configure and deploy a collection of containers
75 / 83
Part 8.
The Future is Now
• swarm mode (since v1.12)
• plugins (since v1.13)
• experimental features
• Docker EE & time-based releases
• The Orchestration Wars
76 / 83
The Future is Now
• Swarm mode (since v1.12)

• service abstraction
• scaling
• service discovery & load balancing
• rolling updates
• stack deployment (docker-compose) (since v1.13)
• secrets management (since v1.13) + config objects (since
v17.06)
• plugins API for datacenter integration (since v1.13)

• volume plugins (eg: flocker)
• network plugins (eg: contiv)
• authorization plugins
• swarm secrets (since v17.07)
77 / 83
Docker CE & Docker EE

since march 2017
Docker inc’s business strategy:
1. be flexible and interoperable with everybody (especially cloud

providers) so that no competing tool emerges
→ open source engine, plugin API for network, storage, authorization integrations
2. sell Docker EE
docker EE = docker CE + support + off-the-shelves datacenter management

(ldap integration, role-based access-control, security scanning, vulnerability
monitoring)
78 / 83
Time-based release
since march 2017 (docker v17.03.0-ce)
• Docker CE
• open source
• edge version released every month
• stable version released every 3 months
• security upgrades during 4 months
• Docker EE
• proprietary
• stable version released every 3 months
• security upgrades during 1 year
79 / 83
The Orchestration Wars
The Container Wars will actually be the Orchestration Wars
• under the hood the base building blocs (runc, containerd) are
open and the competitors cooperate to keep them standard.
• docker itselfs is still free software, although the company

ulture is shifting towards something more “corporate”
• the real fight will be on orchestration solutions

• managing clouds, service hosting
• swarm has opponents (Mesos, Kubernetes, Openshift, ...) and
is lagging.
80 / 83
Apache Mesos
• predates Docker
• designed for very large clusters
• agnostic to the virtualisation technology

• multiple virtualisation tool may coexist in the same cluster
• two-level management
• hard to configure
81 / 83
Kubernetes (k8s)
• project started in 2014 by a group of google developers
• inspired from Google’s internal orchestration framework
• large scale, very sophisticated, not easy to learn
• now hosted by a fundation and adopted by others that use it

as their orchestration backend
• Openshift
• Docker EE
82 / 83
The Open Container Initiative (OCI)

https://github.com/opencontainers/
A Linux Foundation standard for linux containers:

• v1.0.0 released in July 2017
• runtime-spec (launching containers)
• image-spec (image interoperability)
83 / 83

Docker Tutorial - Mars 2019

Uploaded by

Copyright:

Available Formats

Docker Tutorial - Mars 2019

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Docker Tutorial - Mars 2019

Uploaded by

Copyright:

Available Formats

Intro Containers I/O Images Builder Security Ecosystem Future

Université de Rennes 1 / UMR IRISA

March 28, 2019

2. Managing docker containers

4. Managing docker images

5. Building docker images

7. The ecosystem & the future

What is Docker (1/3)

What is Docker (2/3)

What is Docker (3/3)

• a set of REST APIs

How Docker helps?

• normalisation: same environment (container image) for

• archival (ever tried to reuse old codes)

• requires linux kernel >3.8

Lifecycle of a docker container

Container management commands

Notes about the container lifecycle

• the container filesystem is created in docker create and

• the container configuration is mostly static

• other commands are rather basic

Create a new container Start one or more stopped containers

docker run — Run a container

docker run [ options ] image [ arg0 arg1...]

→ create a container and start it

• the container filesystem is initialised from image image

docker run — Foreground mode vs. Detached mode

• Foreground mode is the default

• With -d, the container is run in detached mode:

docker run — TTY allocation

→ with a tty (-t)

docker run — interactive mode

docker run — override defaults (1/2)

working directory (-w)

docker run — override defaults (2/2)

environment variables (-e)

$ docker run -e FOO=foo -e BAR=bar debian sh -c 'echo $FOO $BAR'

docker run — set the container name

Note: Names must be unique

docker run — autoremove

With --rm the container is automatically removed after exit

Remove all zombie containers

Inspecting the container

Interacting with the container

 docker commit example

docker run — mount external volumes

docker run -v /hostpath:/containerpath[:ro] ...

-v mounts the location /hostpath from the host filesystem at the

• store persistent data outside the container

mount examples (1/2)

Inputs (read-only volume)

mount examples (2/2)

docker run — named volumes

• stored inside /var/lib/docker

docker run — grant access to a device

$ docker run --rm debian sh -c 'mknod /dev/sda b 8 0 && fdisk -l /dev/sda'

$ docker run --rm -v /dev/sda:/dev/sda debian fdisk -l /dev/sda

They can be whitelisted with --device

docker commit example