This repository contains code to create containers that test the isolation of Kubernetes or other container orchestrators. Typically, the containers in this repository excessively consume resources, which may affect other containers or processes running on the same host.
These containers can be used to demonstrate shared resource issues and mitigations.
The following containers are available in this repository:
The cpuload container consumes all the CPU that it sees. It checks for the available cores and spawns a process for each core consuming CPU cycles.
The memoryeater container consumes all memory resources. This will lead to the container typically being OOM killed by the orchestrator.
The forkbomb container runs a script that forks infinitely. This creates new processes, potentially affecting other workload by using up system resources (e.g. PIDs).
The filedescriptors container opens as many file descriptors as possible. The file /etc/hosts is used to create the file descriptors. Typically, file descriptors are shared system-wide and not namespaced.
The consume-inodes container creates many small files in its working directory, trying to consume all available inodes. When the underlying filesystem is shared between multiple containers, this can lead to the kubelet having "DiskPressure"
The entropy container consumes randomness by repeatedly querying /dev/random. This will deplete the entropy pool for the kernel. Typically, entropy is system-wide and is not namespaced.
The logspam container writes a lot of data to stdout. Depending on your configuration, this can overwhelm your logging stack or logging infrastructure.
The containers in this repository are available on quay.io.
To run these containers locally:
podman run quay.io/simonkrenger/forkbomb
podman run quay.io/simonkrenger/cpuload
podman run quay.io/simonkrenger/memoryeater
podman run quay.io/simonkrenger/filedescriptors
podman run quay.io/simonkrenger/consume-inodes
podman run quay.io/simonkrenger/entropy
podman run quay.io/simonkrenger/logspam
To run these containers on Kubernetes as a pod (example):
apiVersion: v1
kind: Pod
metadata:
name: simonkrenger-cpuload
spec:
containers:
- name: cpuload
image: quay.io/simonkrenger/cpuload:latest