How to make minikube memory issues more obvious?

Binder BinderHub
1

Hi all,

I just spent some time troubleshooting a misbehaving install of BinderHub on Minikube, which was apparently due to a memory shortage.

The tests wouldn’t pass but I did not get a clue why, and although the BinderHub UI would start and the build would launch, said build would then deterministically crash in the install_miniconda.sh script while creating a conda environment (at line 51).

After fumbling in the dark for quite a while, being a k8s novice, I ended up googling ‘conda crashes silently’ and ending up there: https://github.com/conda/conda/issues/5003 which points at a high memory usage while creating a conda environment.

Deleting the minikube cluster, re-launching minikube start --memory 8192 solved my issue (and the “not_auth” tests now pass, too)

So my question is twofold:

  • Could anyone attempt to reproduce this issue? (I’m running minikube on a nondescript x86-64 ubuntu laptop, and I believe minikube start without any other tuning gives 2000 MB of memory to the underlying VM)

  • Could we update the CONTRIBUTING instructions with memory tuning guidelines and/or hints on how to troubleshoot such issues? (being a k8s novice, I’m not sure how I would get a reason for a container crash or if I could have gotten the answer way quicker…)

Thanks!

1 Like
2

i’d merge a PR that adds a mention of this memory issue and how to fix it to CONTRIBUTING.md.

Thanks for figuring this out!

1 Like
3

Yes, I was able to reproduce this issue on CentOS 7 and macOS 10.15.

1 Like
4

Thanks for the Pull Request! We now have a hint in the contributing guide :slight_smile:

1 Like