Define size for /dev/shm on container engine

Question

I'm running Chrome with xvfb on Debian 8. It works until I open a tab and try to load content. The process dies silently...

Fortunately, I have gotten it to run smoothly on my local docker using docker run --shm-size=1G.

There is a known bug in Chrome that causes it to crash when /dev/shm is too small.

I am deploying to Container engine, and inspecting the OS specs. The host OS has a solid 7G mounted to /dev/shm, but the actual container is only allocated 64M. Chrome crashes.

How can I set the size of /dev/shm when using kubectl to deploy to container engine?

Solution

Mounting an emptyDir to /dev/shm and setting the medium to Memory did the trick!

spec:
  volumes:
  - name: dshm
    emptyDir:
      medium: Memory
  containers:
  - image: gcr.io/project/image
    volumeMounts:
      - mountPath: /dev/shm
        name: dshm

Comments

[posit labs]

Possible duplicate of How to increase shm size of a kubernetes container (--shm-size equivalent of docker)

[speedplane]

I just posted a similar question, relating to app engine flexible instead of container engine. stackoverflow.com/questions/46414587/…

[Nutritioustim]

I'm having a similar issue, but with docker swarm. I don't see spec: or containers: entries in the docker-compose reference. How does this map to the volumes spec?

[posit labs]

Please re-read the question. I already use --shm-size to run locally. This is a question about kubectl configuration

[posit labs]

I recommend posting a new question that is specific to docker swarm

[welch]

Thank you!!! One truly cannot configure options to the docker launch in the kube config (and the devs are consistently obtuse about not understanding the need for this). Your config has our docker-ized selenium test no longer randomly crashing.

[WeiChing 林煒清]

but where to constrains the size of emptyDir?

[Mitar]

Size limit is being worked on here: github.com/kubernetes/kubernetes/pull/63641

[bu11d0zer]

Be aware that sizeLimit on the emptyDir will result in pod evictions if the limit is exceeded. This is much worse behavior than memory-hungry apps crashing.

[EnriqueH]

It worked for me on a Kubernetes (AWS EKS) cluster. Thanks SO MUCH