eatcore

What even is this?

eatcore is a tool intended to provoke memory allocation failures in Kubernetes clusters. We were having some very strange problems with networking under memory pressure and wanted to be able to reproduce the problems on-demand.

Don't run this in your production Kubernetes cluster without running it somewhere else first and observing the behaviour! If you really must, maybe consider constraining it to some nodes that aren't running anything you care about.

How do I use it?

Included in this repo is an example Kubernetes Deployment definition, and also a Dockerfile.

$ cat deployment.yaml ## understand what it will do first!
$ kubectl create -f deployment.yaml

CLI options

eatcore has three key controls:

• -s 500: allocate+fill 500MiB of memory per cycle

• -i 60: sleep 60 seconds between cycles

• -n 100: quit after 100 cycles, if -x specified

Additionally, if -x is not specified, eatcore will sleep indefinitely after completing its cycles.

If -r is specified, all memory allocated will be filled with data generated by arc4random, rather than merely being filled with zeroes. This should effectively prevent it being compressed or deduplicated by hypervisors or operating system kernels.

License

Copyright 2018 SiteMinder.

This software is released under the terms of the MIT license.