Simon Krenger

I recently bought an NVIDIA Jetson Nano Developer Kit to fiddle around with things like MicroShift or TensorFlow. The board is typically used with L4T (Linux for Tegra) based on Ubuntu 18.04. Fedora can also be installed, although not all drivers (for example for the GPU) are available yet. So after properly updating the system with the latest packages, when starting a container using the nvidia runtime, I got the following error:

docker run -it --rm --runtime nvidia --network host nvcr.io/nvidia/l4t-ml:r32.6.1-py3
[..]
docker: Error response from daemon: failed to create shim: OCI runtime create failed: container_linux.go:380: starting container process caused: error adding seccomp filter rule for syscall clone3: permission denied: unknown.

For my own container images, I often like to use the Fedora Container Images as the base image. This means I often use the “fedora:32” or “fedora-minimal:32” image when building my own images.

Yesterday, while playing around with an image based on the “fedora-minimal” image that then uses nginx and php-fpm, I came across this curious error:

Invalid date.timezone value 'UTC', we selected the timezone 'UTC' for now

Some time ago, I had a curious case of very slow DNS resolution in a container on OpenShift. The symptoms were as follows:

• In the PHP application in the container, DNS resolution was very slow with a 5 second delay before the lookup was resolved
• In the container itself, DNS resolution for curl was very slow, with a 5 second timeout before the lookup was resolved
• However, using dig in the container itself, DNS resolution was instant
• Also, on the worker node, the DNS resolution was instant (using both dig and curl)

TL;DR: Since glibc 2.10, glibc performs IPv4 and IPv6 lookups in parallel. When IPv6 fails, there is a 5 second timeout in many cases before the lookup is returned. Disable IPv6 DNS lookups by setting “single-request” in “resolv.conf” or disable the IPv6 stack completely.

In the past few months, on all my machines I have replaced Docker with Podman and mostly the transition has been quite smooth. There are still some rough edges here and there, but the overall experience of using Podman has been great!

However, when trying to start a very simple container, one often runs into the following issue:

$ podman run -p80:80 nginx:latest 
Error: error from slirp4netns while setting up port redirection: map[desc:bad request: add_hostfwd: slirp_add_hostfwd failed]

The error message looks very cryptic, but the issue is quite simple: As a regular user, one is typically not allowed to bind ports < 1024. So by trying to bind port 80, you will get the error above.

The fix is trivial, just use a port greater than 1024:

$ podman run -p8080:80 -d nginx:latest 
22d2be2966e9cb77246a8b698f9024de89f4e6d1a0edfe44209bbe4fd27aa8b5
$ curl localhost:8080
[..]
Welcome to nginx!
[..]

If you really need to use a port number lower than 1024, there are multiple ways to configure that:

• Set net.ipv4.ip_unprivileged_port_start=80or similar in your sysctl
• Add the CAP_NET_BIND_SERVICE capability to your process or user

At their core, containers are just Linux processes that are namespaced. This means in practice, many containers still run as processes on the same host machine. While namespacing processes using cgroups creates very good boundaries between processes, the isolation is still not perfect.

So when using NodeSelectors in OpenShift, you’ll also have to set labels on your nodes. You can find more information on labeling nodes in the OpenShift documentation. Here is how you can add or remove a label from a node or pod:

To add a label to a node or pod:

# oc label node node001.krenger.ch mylabel=myvalue
# oc label pod mypod-34-g0f7k mylabel=myvalue

To remove a label (in the example “mylabel”) from a node or pod:

# oc label node node001.krenger.ch mylabel-
# oc label pod mypod-34-g0f7k mylabel-

You can also use oc label -h to see more options for the oc label command.