Debian Working System for Docker

NOTE: Github is no longer the home for this project; see the new home on Salsa.

This image is part of the docker-debian-base image set.

This is a simple set of images that transform the standard Docker Debian environment into one that provides more traditional full Unix APIs (including syslog, zombie process collection, etc.)

Despite this, they are all very small, both in terms of disk and RAM usage.

You can find a description of the motivation for these images on my blog.

This is loosely based on the concepts, but not the code, in the phusion baseimage-docker. You can look at that link for additional discussion on the motivations.

You can find the source and documentation at the Salsa page and automatic builds are available from my Docker hub page. The builds are auto-generated from Salsa CI and run at least weekly.

OUDATED: For stretch and jessie, this image uses sysvinit instead of systemd, not because of any particular opinion on the merits of them, but rather because sysvinit does not require any kind of privileged Docker or cgroups access.

For buster and bullseye, systemd contains the necessary support for running in an unprivileged Docker container and, as it doesn't require the hacks that sysvinit does, is used there. The systemd and sysvinit images provide an identical set of features and installed software, which target the standard Linux API.

Here are the images I provide from this repository:

• jgoerzen/debian-base-minimal - a minimalistic base for you.
  • Provides working sysvinit/systemd, syslogd, cron, anacron, at, and logrotate.
  • syslogd is configured to output to the docker log system by default.
• jgoerzen/debian-base-standard - adds some utilities. Containes everything above, plus:
  • Utilities: less, nano, vim-tiny, man-db (for viewing manpages), net-tools, wget, curl, pwgen, zip, unzip
  • Email: exim4-daemon-light, mailx
  • Network: netcat-openbsd, socat, openssl, ssh, telnet (client)
• jgoerzen/debian-base-security - A great way to keep things updated. Contains everything above, plus:
  • automated security patches using unattended-upgrades and needrestart
  • debian-security-support
  • At container initialization, runs the unattended-upgrade code path to ensure that the system is up-to-date before services are exposed to the Internet. This addresses an issue wherein security patches may hit security.debian.org before Docker images are refreshed, a fairly common issue with the Docker infrastructure. This behavior can be suppressed with DEBBASE_NO_STARTUP_APT (see below).
• jgoerzen/debian-base-vnc - For systems that need X. debian-base-security, plus:
  • tightvncserver, xfonts-base, lwm, xterm, xdotool, xvnc4viewer
• jgoerzen/debian-base-apache - A web server - debian-base-security, plus:
  • apache2 plus utilities: ssl-cert
  • LetsEncrypt options: certbot, acme-tiny
• jgoerzen/debian-base-apache-php - debian-base-apache, plus:
  • libapache2-mod-php (mod-php5 on jessie)

Memory usage at boot (stretch):

• jgoerzen/debian-base-minimal: 6MB
• jgoerzen/debian-base-standard: 11MB
• jgoerzen/debian-base-security: 11MB

Docker Tags

These tags are autobuilt:

• latest: whatever is stable (currently bullseye, systemd)
• bullseye: Debian bullseye (systemd)
• buster: Debian buster (systemd)
• stretch: Debian stretch (sysvinit) - no longer supported, may be removed at any time
• jessie: Debian jessie (sysvinit) - no longer supported, may be removed at any time
• sid: Debian sid (not tested; systemd)

Install

You can install with:

docker pull jgoerzen/debian-base-whatever

Your Dockerfile should use CMD to run /usr/local/bin/boot-debian-base.

When running, use -t to enable the logging to docker logs

Container Invocation

A container should be started using these commands, among others. See also the section on environment variables, below.

Container Invocation, systemd containers (buster/bullseye/sid)

Here's how you invoke for systemd (buster/bullseye) on a system running an older systemd on the host, with cgroups v1:

docker run -td --stop-signal=SIGRTMIN+3 \
  --tmpfs /run:size=100M --tmpfs /run/lock:size=100M \
  -v /sys/fs/cgroup:/sys/fs/cgroup:ro \
  --name=name jgoerzen/debian-base-whatever

For a host running bullseye, or a newer cgroups and systemd, you have to use this:

docker run -td --stop-signal=SIGRTMIN+3 \
  --tmpfs /run:size=100M --tmpfs /run/lock:size=100M \
  -v /sys/fs/cgroup:/sys/fs/cgroup:rw --cgroupns=host \
  --name=name jgoerzen/debian-base-whatever

Note that the buster image has not been tested under these situations, and since bullseye is now stable, it is the recommended image for all modern deployments.

The /run and /run/lock tmpfs are required by systemd. The 100M sets a maximum size, not a default allocation, and serves to limit the amount of RAM an errant process could cause the system to consume, down from a default limit of 16G.

Note that these images, contrary to many others out there, do NOT require --privileged.

For more information about the systemd/cgroups situation, consult these links

• systemd/systemd#19245
• containers/podman#5153
• moby/moby#42275
• https://serverfault.com/questions/1053187/systemd-fails-to-run-in-a-docker-container-when-using-cgroupv2-cgroupns-priva/1054414#1054414
• http://docs.podman.io/en/latest/markdown/podman-run.1.html#cgroupns-mode

Container Invocation, sysvinit containers (jessie/stretch)

docker run -td --stop-signal=SIGPWR --name=name jgoerzen/debian-base-whatever

Environment Variables

This environment variable is available for your use:

• DEBBASE_SYSLOG defaults to stdout, which redirects all syslog activity to the Docker infrastructure. If you instead set it to internal, it will use the default Debian configuration of logging to /var/log within the container. The configuration is applied at container start time by adjusting the /etc/syslog.conf symlink to point to either syslog.conf.internal or syslog.conf.stdout. syslog.conf.internal is the default from the system. dpkg-divert is used to force all packages' attempts to write to /etc/syslog.conf to instead write to /etc/syslog.conf.internal.
• DEBBASE_TIMEZONE, if set, will configure the /etc/timezone and /etc/localtime files in the container to the appropriate timezone. Set this to the desired timezone; for instance, America/Denver.
• DEBBASE_SSH defaults to disabled. If you set to enabled, then the SSH server will be run.
• DEBBASE_NO_STARTUP_APT defaults to empty. If set, it will cause images based on debian-base-security to skip the apt job run at container startup.

Container initialization

Executables or scripts may be placed in /usr/local/preinit, which will be executed at container start time by run-parts prior to starting init. These can therefore perform container startup steps. A script which needs to only run once can delete itself after a successful run to prevent a future execution.

Orderly Shutdown

The --stop-signal clause in the "Container Invocation" section above helps achieve an orderly shutdown.

If you start without --stop-signal, you can instead use these steps:

# jessie or stretch use this line:
docker kill -s SIGPWR container
# bullseye, buster or sid use this one:
docker kill -s SIGRTMIN+3 container

# Either way, then proceed with:
sleep 10
docker kill container

Within the container, you can call telinit 1 (jessie/stretch) or poweroff (bullseye/buster/sid) to cause the container to shutdown.

Advanted topic: Orderly Shutdown Mechanics

By default, docker stop sends the SIGTERM (and, later, SIGKILL) signal to PID 1 (init) iniside a container. Neither sysvinit nor systemd act upon this signal in a useful way. This will shut down a container, but it will not give your shutdown scripts the chance to run gracefully. In many situations, this is fine, but it may not be so in all.

A workaround is, howerver, readily available, without modifying init. These images are configured to perform a graceful shutdown upon receiving SIGPWR (jessie/stretch) or SIGRTMIN+3 (bullseye/buster/sid).

The process for this with sysvinit is... interesting, since we are unable to directly kill PID 1 inside a docker container. First, init calls /etc/init.d/powerfail. The powerfail script I install simply tells init to go to single-user mode. This causes it to perform an orderly shutdown of the daemons, and when it is done, it invokes /sbin/sulogin. On an ordinary system, this prompts for the root password for single-user mode. In this environment, we instead symlink /sbin/init to /bin/true, then tell init to re-exec itself. This causes PID 1 to finally exit.

With sysvinit, one of the preinit scripts makes sure that /sbin/init properly links to /sbin/init.real at boot time.

With systemd in bullseye/buster/sid, no special code for all this is needed; systemd handles it internally with no fuss.

Configuration

Although the standard and security images run the SMTP and SSH servers, they do not expose these to the Internet by default. Both require site-specific configuration before they are actually useful.

Because the SMTP service is used inside containers, but the SSH service generally is not, the SSH service is disabled by default.

Enabling or Disabling Services

You can enable or disable services using commands like this (jessie/stretch):

update-rc.d ssh disable 
update-rc.d ssh enable

Or this (bullseye/buster/sid):

systemctl disable ssh
systemctl enable ssh

(Note, that in the case of ssh, the environment variable will cause commands like this to be executed automatically on each container start.)

Email

email is the main thing you'd need to configure. In the running system, dpkg-reconfigure -plow exim4-config will let you do this.

SSH

SSH host keys will be generated upon first run of a container, if they do not already exist. This implies every instantiation of a container containing SSH will have a new random host key. If you want to override this, you can of course supply your own files in /etc/ssh or make it a volume.

Advanced topic: programs that depend on disabled scripts (stretch/jessie only)

This section pertains only to stretch/jessie; systemd in bullseye/buster/sid does not have these issues.

There are a number of scripts in /etc/init.d that are normally part of a Debian system initialization, but fail in a Docker environment. They do things like set up swap space, mount filesystems, etc. Docker images typically leave those scripts in place, but they are never called because Docker systems typically don't run a real init like these images do.

Although calling the scripts produces nothing worse than harmless errors, I have disabled those scripts in these images in order to avoid putting useless error messages in people's log files. In some very rare circumstances, this may cause installation of additional packages to fail due to boot script dependency ordering not working right. (Again, this is very rare).

I saw this happen once where a package had a long chain of dependencies that wound up pulling in cgmanager, which died in postinst complaining that its init script required mountkernfs. I worked around this in my Dockerfile like this:

update-rc.d mountkernfs.sh defaults
apt-get -y --no-install-recommends offending-package
update-rc.d -f cgmanager remove
update-rc.d -f mountkernfs.sh remove

Also, I have blocked systemd from accidentally being installed on the system. There are a few packages that pull in systemd shims and so forth, so if you get errors about systemd not installing, try adding rm /etc/apt/preferences.d/systemd to your Dockerfile.

Advanced Topic: Adding these enhancements to other images

Sometimes, it is desirable to not have to rebuild an image entirely. These images are also designed to make it easy to add the functionality to other images. You can do this by using the support for multiple FROM lines in a Dockerfile. For instance, here's a simple one I worked up:

FROM jgoerzen/debian-base-security:jessie AS debian-addons

FROM homeassistant/home-assistant:0.63.1

COPY --from=debian-addons /usr/local/preinit/ /usr/local/preinit/
COPY --from=debian-addons /usr/local/bin/ /usr/local/bin/
COPY --from=debian-addons /usr/local/debian-base-setup/ /usr/local/debian-base-setup/

RUN run-parts --exit-on-error --verbose /usr/local/debian-base-setup
CMD ["/usr/local/bin/boot-debian-base"]

It happens that home-assistant is based on a Python image which, in turn, is based on Debian jessie. There are just those four lines that are needed: copying the /usr/local/preinit, bin, and debian-base-setup directories, and then the run-parts call. This effectively adds all the features of debian-base-security to the home-assistant image.

This works because each image that is part of the chain leading up to security (minimal, standard, and security) performs all of its activity from scripts it drops -- and leaves -- in /usr/local/debian-base-setup. Those scripts need nothing other than the files in the three directories referenced above. By adding those three directories and calling the scripts, it is easy to add these features to other images.

Source

This is prepared by John Goerzen jgoerzen@complete.org and the source can be found at https://salsa.debian.org/jgoerzen/docker-debian-base

See Also

Some references to additional information:

• systemd's contianer interface documentation
• Article on running systemd in a container. Highlights some of the reasons to do so: providing a standard Linux API, reaping zombie processes, handling of logging, not having to re-implement init, etc. All of these have already been implemented in these images with sysvinit and continue with systemd.
• serverfault thread

Copyright

Docker scripts, etc. are Copyright (c) 2017-2019 John Goerzen All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
3. Neither the name of the University nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Additional software copyrights as noted.