Jetson: docker fails to download image nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

I am trying to run the first exercise in the Jetson tutorial.

Jetson setup is complete. Network connection works and verified (ethernet cord is plugged in and I can ping websites like google form command line).

In headless mode, I try to DL the docker image, but it is not responding. Command output is:

sudo docker run --runtime nvidia -it --rm --network host --volume ~/nvdli-data:/nvdli-nano/data --device /dev/video0 nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1
Unable to find image ‘nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1’ locally
docker: Error response from daemon: Get https://nvcr.io/v2/: net/http: request canceled while waiting for connection (Client.Timeout exceeded while awaiting headers).
See ‘docker run --help’.

I am on
4.6+ v2.0.1-r32.6.1 en-US

Can anyone help?

1 Like
2

Please note:
unable to ping nvcr.io directly from jetson nano. but i can ping google.com

~$ ping nvcr.io
PING nvcr.io (54.176.23.67) 56(84) bytes of data.
From jetson (192.168.0.165) icmp_seq=1 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=2 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=3 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=4 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=5 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=6 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=7 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=8 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=9 Destination Host Unreachable
From jetson (192.168.0.165) icmp_seq=10 Destination Host Unreachable

4

Hi,

Just check the nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1 container, we can start the download normally.

$ sudo docker run --runtime nvidia -it --rm --network hos                                                                                                             t --volume ~/nvdli-data:/nvdli-nano/data --device /dev/video0 nvcr.io/nvidia/dli                                                                                                             /dli-nano-ai:v2.0.1-r32.6.1
[sudo] password for nvidia:
Unable to find image 'nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1' locally
v2.0.1-r32.6.1: Pulling from nvidia/dli/dli-nano-ai
b9bb7af7248f: Pulling fs layer
334a570c08f5: Pulling fs layer
c8bd4a89e71c: Download complete
0a8ce4f08307: Pulling fs layer
2e163bfa50fc: Pulling fs layer
8d6d8c86a148: Pulling fs layer
b483bad8509f: Pulling fs layer
8da3c6b29858: Pulling fs layer
...

It seems that there are some issues in the docker setting.
Do you use the default docker that comes with the JetPack installer or install it manually?

Please also check if the following topic helps.

Thanks.

5

I am simply following the instructions from the NVIDIA course “Getting Started with AI on Jetson Nano”

I did not make changes to or install docker, just going through all the instructions one by one. You can see step 7 here.

If there is an additional step that needs to be added, maybe it should be added to the tutorial?

1 Like
6

The issue seems to be the network setup.
When I first setup jetson, I did not have an ethernet plugged in.

This is the network configuration screen you see in headless mode when you first setup the Jetson Nano:

Screen Shot 2021-10-14 at 2.26.43 PM
Screen Shot 2021-10-14 at 2.26.43 PM1144×642 46.8 KB

What is important is that you must:

  1. have ethernet plugged in
  2. select the ethernet device (NOT the dummy device, which is default)

This is not documented anywhere. Once I did this I was able to connect and download the docker image ok, even though, for the record, you can’t ping nvcr.io directly from command line, but the docker command IS able to connect to this domain somehow

2 Likes
8

Hello,

I am also experiencing this issue. However, trying to use the CSI Camera.

From the documentation, I am trying to execute:

sudo docker run --runtime nvidia -it --rm --network host
–volume ~/nvdli-data:/nvdli-nano/data
–volume /tmp/argus_socket:/tmp/argus_socket
–device /dev/video0 \
nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1

And, I am receiving an error message;
docker: invalid reference format.

I setup first boot with wifi and wifi is working, so the download should work.

Any help troubleshooting this is appreciated. I too am trying to follow the course, and with no prior experience with docker, this is a bit of a struggle this early in the course. Thank you.

9

Ok. I figured it out. Here is the fix:

  1. Login to the jetson with GUI.
  2. Enter the front part of the command line item, with the docker image name at end, without any of the middle:

sudo docker run --runtime nvidia -it --rm --network host nvcr.io/nvidia/dli/dli-nano-ai:v2.0.1-r32.6.1

It will fail to find local, then download it from online (for mine over wifi).

  1. Create the script per the course, if you have CSI camera and the 2GB, see the additional line there about memory when you create the script.

  2. Set the cmod permission, Run the script on the jetson. Shutdown the Jupyter notebook. Reboot.

  3. Login via ssh from your normal device. It will now work.

Basically, downloading it local helps, and this is easier to troubleshoot via the actual device first, then do the ssh.

2 Likes