Someone asked on IRC about running ntpd in a container on Atomic, so I've put together a small example. We'll start with a very simple Dockerfile:

FROM alpine
RUN apk update
RUN apk add openntpd
ENTRYPOINT ["ntpd"]

I'm using the alpine image as my starting point because it's very small, which makes this whole process go a little faster. I'm installing the openntpd package, which provides the ntpd binary.

By setting an ENTRYPOINT here, the ntpd binary will be started by default, and any arguments passed to docker run after the image name will be passed to ntpd.

We need to first build the image:

# docker build -t larsks/ntpd .

And then we can try to run it:

# docker run larsks/ntpd -h
ntpd: unrecognized option: h
usage: ntpd [-dnSsv] [-f file] [-p file]

That confirms that we can run the command. Now we need to provide it with a configuration file. I looked briefly at the ntpd.conf man page, and I think we can get away with just providing the name of an ntp server. I created /etc/ntpd.conf on my atomic host with the following content:

servers pool.ntp.org

And then I tried running the container like this:

docker run -v /etc/ntpd.conf:/etc/ntpd.conf larsks/ntpd -d -f /etc/ntpd.conf

The -v in the above command line makes /etc/ntpd.conf on the host available as /etc/ntpd.conf inside the container. This gets me:

ntpd: can't set priority: Permission denied
reset adjtime failed: Operation not permitted
adjtimex (2) failed: Operation not permitted
adjtimex adjusted frequency by 0.000000ppm
fatal: privsep dir /var/empty could not be opened: No such file or directory
Lost child: child exited
dispatch_imsg in main: pipe closed
Terminating

The first few errors ("Permission denied") mean that we need to pass --privileged on the docker run command line. Normally, Docker runs containers with limited capabilities, but because an ntp service needs to be able to set the time in the kernel it won't run in that limited environment.

The "fatal: privsep dir /var/empty could not be opened" suggests we need an empty directory at /var/empty. With the above two facts in mind, I tried:

docker run --privileged -v /var/empty \
  -v /etc/ntpd.conf:/etc/ntpd.conf larsks/ntpd -d -f /etc/ntpd.conf -s

The -s on the end means "Try to set the time immediately at startup." This results in:

adjtimex adjusted frequency by 0.000000ppm
ntp engine ready
reply from 104.131.53.252: offset -3.543963 delay 0.018517, next query 5s
set local clock to Fri Oct  9 18:03:41 UTC 2015 (offset -3.543963s)
reply from 198.23.200.19: negative delay -7.039390s, next query 3190s
reply from 107.170.224.8: negative delay -6.983865s, next query 3139s
reply from 209.118.204.201: negative delay -6.982698s, next query 3216s
reply from 104.131.53.252: offset 3.523820 delay 0.018231, next query 8s

So that seems to work correctly. To make this service persistent, I can add -d to start the container in the background, and --restart=always to make Docker responsible for restarting it if it fails:

docker run --privileged -v /var/empty \
  --restart=always -d \
  -v /etc/ntpd.conf:/etc/ntpd.conf larsks/ntpd -d -f /etc/ntpd.conf -s

And my Atomic host has an ntp service keeping the time in sync.


Heat-kubernetes Demo with Autoscaling

Next week is the Red Hat Summit in Boston, and I'll be taking part in a Project Atomic presentation in which I will discuss various (well, two) options for deploying Atomic into an OpenStack environment, focusing on my heat-kubernetes templates.

As part of that presentation, I've put together a short ...

read more

Fedora Atomic, OpenStack, and Kubernetes (oh my)

Mon 24 November 2014 by Lars Kellogg-Stedman Tags openstack kubernetes fedora atomic

While experimenting with Fedora Atomic, I was looking for an elegant way to automatically deploy Atomic into an OpenStack environment and then automatically schedule some Docker containers on the Atomic host. This post describes my solution.

read more