[Support] D34DC3N73R - Netdata GLIBC (GPU Enabled)

[D34DC3N73R]

11 hours ago, veri745 said:

 

I reinstalled and used those default mappings.  There does not appear to be any write activity to any of those mapped directories.

 

The config directory remained complete empty.  I copied over the data from my "override" folder, but that doesn't seem to make any difference.

 

I would just go ahead and use the official netdata docker, but the template for that image is completely empty, too

 

/mnt/user/appdata/netdata/config will be populated on first run with the following folders & files:


From here, you should use the netdata edit-config script to make changes to any conf. 

For instance, to make changes to python.d.conf
 

root@Tower:~# docker exec -it netdata-glibc bash
unraid:/# cd /etc/netdata
unraid:/etc/netdata# ./edit-config python.d.conf

 

That will open up the default /usr/lib/netdata/conf.d/python.d.conf in a vi editor where you can make any changes. That fill will be saved to /etc/netdata as an override. This is the same functionality as the default netdata image.

 

At this point, the only difference between this image and the default image is the glibc library required for nvidia GPUs and nvidia-smi being enabled by default in python.d.conf

I do have a template I'm testing for the default netdata image now that it's compatible. If you want to test it out, see this post.

[veri745]

25 minutes ago, D34DC3N73R said:

 

/mnt/user/appdata/netdata/config will be populated on first run with the following folders & files:

 

The only files created are `.container-hostname` and `.opt-out-from-anonymous-statistics`

 

I did get a bunch of folders and a `netdata.conf` when I installed the `stable` version, but not with `latest`

[D34DC3N73R]

11 minutes ago, veri745 said:

 

The only files created are `.container-hostname` and `.opt-out-from-anonymous-statistics`

 

I did get a bunch of folders and a `netdata.conf` when I installed the `stable` version, but not with `latest`

 

I moved my entire netdata appdata folder to netdata.bak and started from scratch, but I'm not able to reproduce that with the latest or stable images and the current template from CA. Do you have auto-updates enabled? Is it possible you're using old images? Can you confirm it's running netdata v1.39.0?

 

[veri745]

10 minutes ago, D34DC3N73R said:

 

Do you have auto-updates enabled? Is it possible you're using old images? Can you confirm it's running netdata v1.39.0?

 

Auto-update is enabled, but I just did a fresh install today.  The netdata UI says "v1.39.0-23-nightly"

[D34DC3N73R]

11 minutes ago, veri745 said:

Auto-update is enabled, but I just did a fresh install today.  The netdata UI says "v1.39.0-23-nightly"

 

That seems correct. Did you make any edits to the template or run it as is?

[zoggy]

 

netdata changed a bit of stuff recently (their icon are svg), so fallout is that unraid netdata template's icon is no longer there.
Since unraid doesnt support loading svg, the easy fix for icon is just to point to old image from old version when it was still there:

https://raw.githubusercontent.com/netdata/netdata/v1.40/web/gui/dashboard/images/ms-icon-70x70.png

 

[mihcox]

Is there any direct guide on how to convert a docker container on your image on unraid to the new netdata/netdata one while preserving the ability to view my nvidia card?

[D34DC3N73R]

2 hours ago, mihcox said:

Is there any direct guide on how to convert a docker container on your image on unraid to the new netdata/netdata one while preserving the ability to view my nvidia card?

 

I haven't written it all out yet, but I do have a template that works. It need some documentation updates since the python nvidia-smi has been deprecated. If you'd like to test it out, you can follow these steps.

 

Stop and delete the existing container. (configuration, not data)

 

download the XML template

wget -O /boot/config/plugins/dockerMan/templates-user/my-netdata-gpu.xml https://raw.githubusercontent.com/D34DC3N73R/unraid-templates/netdata-image/netdata-glibc.xml

 
In the unraid docker page, scroll to the bottom and click 'add container'. Select netdata-gpu from the template dropdown. The standard paths should be good for most new installs. The data and paths are compatible with the previous version of my image. However, my image shipped with python nvidia-smi enabled by default. Since that is now deprecated, you should disable it and enable the go.d implementation. 

 

To do that, you should start the new container and then use the built in .edit-config 

 

If you're using an existing configuration, you should edit python.d.conf

docker exec -it netdata bash
cd /etc/netdata
./edit-config python.d.conf

You'll need to add a # in front of nvidia-smi to disable the python nvidia-smi module. Save and quit vi

 

If you're using a new configuration or you've disabled the python nvidia-smi, then you can edit go.d.conf

docker exec -it netdata bash
cd /etc/netdata
./edit-config go.d.conf

 

That will open the default go.d.conf file using vi, and save changes in your mounted directory /mnt/user/appdata/netdata/config. You'll need to delete the # and change 'no' to 'yes'. 

#  nvidia_smi: no

so it becomes 

  nvidia_smi: yes

The two spaces before nvidia-smi are important.

 

When you're done editing, save, exit the container cli, and restart the container. 

 

If you have questions or run into problems please let me know. I'll be pushing the new template soon after I have time to run some more tests.

 

Edit: Tip: If you'd like to use an alternate editor in place of vi, you can add the nano package for instance to the NETDATA_EXTRA_APK_PACKAGES environment variable (default `gcompat`). add a space between package names `gcompat nano`. Then you'll be able to call the edit config script using the new editor

./edit-config --editor nano python.d.conf

 

Edited July 28, 2023 by D34DC3N73R
add editor tip

[mihcox]

Ran through this without any noticable issue, but still not seeing it in the dashboard after rebooting Netdata. I can run nvidia-smi from inside the container and get results, but nothing on the dashboard.

 

 

Edit: after waiting about 5 minutes, its now showing in the dashboard!

Edited July 29, 2023 by mihcox

[D34DC3N73R]

Excellent. Glad it's working for you. I believe there is still more work to be done on the go.d plugin, and honestly it seemed a bit premature to deprecate the python.d version, but hopefully it'll catch up soon. Let me know if you run into any issues. Thanks for trying out the new version of the template.

[Cyklone]

On 7/28/2023 at 10:50 PM, D34DC3N73R said:

Excellent. Glad it's working for you. I believe there is still more work to be done on the go.d plugin, and honestly it seemed a bit premature to deprecate the python.d version, but hopefully it'll catch up soon. Let me know if you run into any issues. Thanks for trying out the new version of the template.

Thank you for this!

It also worked for me.  Could not get the standard netdata to report nvidia stats.

However, I seem to remember that more stats were available: pci_bandwidth, encoder/decoder utilisation, processes, etc.  Are these no longer available?

Cheers!

[D34DC3N73R]

On 10/7/2023 at 3:12 PM, Cyklone said:

Thank you for this!

It also worked for me.  Could not get the standard netdata to report nvidia stats.

However, I seem to remember that more stats were available: pci_bandwidth, encoder/decoder utilisation, processes, etc.  Are these no longer available?

Cheers!

 

It doesn't look like everything from the python plugin has been added to the go plugin yet. I would hope they're coming, but the change was probably necessary for the new interface. These are what show up for me