Google cloud dataproc cluster created with an environment.yaml with a jupyter resource but environment not available as a jupyter kernel

Question

I have created a new dataproc cluster with a specific environment.yaml. Here is the command that I have used to create that cluster:

gcloud dataproc clusters create dataproc-testing1 
--enable-component-gateway 
--bucket my-test-bucket 
--region us-central1 --zone us-central1-c 
--master-machine-type n1-standard-2 
--master-boot-disk-size 32 
--num-workers 3 
--worker-machine-type n1-standard-2 
--worker-boot-disk-size 32 
--num-secondary-workers 3 
--preemptible-worker-boot-disk-type 
--preemptible-worker-boot-disk-size 32 
--num-preemptible-worker-local-ssds 0 
--image-version 2.0-ubuntu18 
--properties dataproc:conda.env.config.uri=gs://my-test-bucket/environment.yaml 
--optional-components JUPYTER 
--scopes 'https://www.googleapis.com/auth/cloud-platform' 
--project my-project

This successfully creates the cluster.

I have been able to ssh into the master and the executor nodes and they all have an environment pyspark that is created with the environment.yaml that I specified in the cluster creation command above. All the dependencies are there and python version is also 3.9.7.

After SSH into the worker or master nodes and running python --version gives me Python 3.9.7

running conda env list gives me

#
base                     /opt/conda/miniconda3
pyspark               *  /opt/conda/miniconda3/envs/pyspark

Hence, the environment activated is pyspark.

I can deactivate this environment with conda deactivate and then the base environment is activated and there python --version results in Python 3.8.12

So far everything is as I expect.

Now, I ran the jupyter notebook from the web interfaces tab in the cluster console and the problem is:

It has just 'PySpark' (note this is not the same as pyspark), 'Python3', 'spylon-kernel`, 'R' kernels available. 'R' is for R and 'spylon-kernel' is for scala.

I activate 'PySpark' kernel and run

import sys
sys.version

and the output is

'3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:59:51) \n[GCC 9.4.0]'

I activate 'Python 3' kernel and run

import sys
sys.version

and the output is '3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:59:51) \n[GCC 9.4.0]'

In both these kernels none of the packages from environment.yaml are available.

In conclusion, I cannot access the pyspark environment created by environment.yaml.

Can you please help me access the pyspark environment created by environment.yaml?

Answer 1

Dataproc launches Jupyter notebook from the root/base environment and the user specified environment is not leveraged there. This is currently not supported by Dataproc.