question

TJJ-2213 avatar image
0 Votes"
TJJ-2213 asked timoklimmer answered

[Azure DSVM] GPU not usable in pre-installed python kernels and file permission(read-only) problems in jupyterhub environment

I got two questions.

Steps to reproduce Q1:
1. create a "Data Science Virtual Machine- Ubuntu 18.04" resource, area US northcentral, authentication by password, other options setting to default.
2. After the VM was created, enter https://VMIP:8000 and login.
3. create a file testGPU.ipynb
4. switch to different kernels and test GPU availability and see results below:
98481-image.png
98482-image.png
98483-image.png
98472-image.png
5. in command line, run python, and then run the same codes above, it indicated that GPU is available. (e.g. Num GPUs: 1)
6. in command line, run conda activate py38_tensorflow, then python, and run codes above, it indicated that GPU is available. (e.g. Num GPUs: 1)

Problems Q1: GPU not usable in pre-installed python kernels in jupyterhub environment
Expect: GPU is available in tensorflow in jupyterhub.



Steps to reproduce Q2:
1. create an azure storage account and create an azure file share resource
2. put some files in the Azure file share space(e.g. test.txt)
3. mount an Azure file share drive by the command recommended in Azure portal, but add arguments gid=100(assign usergroup=users to all mounted files)
4. add a user USERID to a group by sudo usermod -a -G users USERID
5. create a symbolic link by ln -s /mnt/mounted_file_share ~/notebooks/fileshare
6. reboot the machine
7. sudo chmod 777 test.txt
8. vi test.txt and it does not say it is read-only.
9. open test.txt in jupyterhub and try to edit and save but finding out that save failed.
10. vi test.txt again and find out it is readonly now. r-xr-xr-x 1 root users 5981 May 21 04:22 test.txt

Problems Q2: file permissions are always read-only in jupyterhub environment for those files not created by jupyterhub.


















azure-data-science-vm
image.png (80.7 KiB)
image.png (80.4 KiB)
image.png (82.6 KiB)
image.png (64.6 KiB)
· 3
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

@TJJ-2213 What is the size that is selected for the DSVM that is used?

0 Votes 0 ·

It's Standard_NC6_Promo

0 Votes 0 ·

Hi, I am from the DSVM team. We are currently looking into this. I will come back asap.

0 Votes 0 ·
timoklimmer avatar image
1 Vote"
timoklimmer answered timoklimmer edited

The GPU problem can be fixed by installing TensorFlow 2.5.0 which came out just recently. To fix, run !pip install --upgrade tensorflow-gpu in a Jupyter cell and restart the kernel. We will fix this in the next releases, either next week or end of June/beginning of August. I haven't had the time to look into Q2 but will do as well and come back again.

@TJJ-2213 fyi

5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

timoklimmer avatar image
1 Vote"
timoklimmer answered

I have taken a look at the second issue meanwhile. When you run the command from Azure Portal, there is no need to add something like gid=100 or add the new user to the users group. The script mounts the file share with 777 permissions already. I tried and could write to the same file with two different users.

@TJJ-2213 fyi

5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.