Error 502 (Bad Gateway) when accessing Jupyter Notebook from Google Deep Learning VM

Question

I have recently quoted a Deep Learning VM from Google Cloud. DLVM provides a link to the jupyter notebook of our VM, but when I paste the link to the browser, it gave us Error 502 (Bad Gateway).

The VM consists of 100 GB of memory, 4 vCPU (26 GB Memory), and 2 Tesla T4 GPUS. I tried to stop and restart the instance several times but it did not work.

The link to Jupyter Notebook is provided by this command given by the "Get started" page:

gcloud compute instances describe --project <project-id> --zone <zone-id> <instance-id> | grep googleusercontent.com | grep datalab

The command successfully gave me a link.

However, when I paste the link to the browser, it gave me an error as followings:

Error 502 (Bad Gateway)!! 502. That’s an error. That’s all we know.

Answer 1

Restarting the Jupyterhub server will resolve this error

Answer 2

It is hard to debug the problem based on just this information.

Some things to try:

• try opening the link in the incognito mode (after clearing the cookies) and sing in with Google Account
• make sure that your user has Project Editor permissions on the GCP project
• try to create notebooks via our new UI: https://console.cloud.google.com/mlengine/notebooks/instances

let me know if any of this helps, if not please ping me directly I can assist with debugging.

Answer 3

I had the same problem and here's what I found helps. If you deploy the DLVM with specs that are same as or above the default, i.e

Machine type : 2 vCPUs with 13GB memory and 1 GPU

it won't crash. Any config less than this, gives me the same error as you. I can't get anymore info or logs on the error nor do I know if lesser configs are actually below the minimum system reqs for the DLVM but this will definitely let you use the DLVM seamlessly.