NVIDIA Sync fails to add new host with error "Invalid Argument Host already exists"

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

I’m hitting an error connecting to my Spark with the NVIDIA Sync on macOS. I’ve looked for the ssh_config ( per DGX Spark / GB10 FAQ ) and the config isn’t yet present. I am able to manually ssh from my terminal to the Spark and login in successfully. Any ideas what is going wrong?

image
image838×532 10.2 KB

2

@rickeerickson on your MacOS under $USER/.ssh and delete the offending host name from known_hosts

3

No joy. I’m still getting the error.

4

In that case quit Sync app and delete all the files in /Users/$USER/Library/Application Support/NVIDIA/Sync/config/

The config files will be recreated. Sample list output:

-rw-rw-rw-@ 1 elsaco staff 32 Oct 24 19:40 first-run.json
-rw-------@ 1 elsaco staff 420 Oct 24 19:40 nvsync.key
-rw-------@ 1 elsaco staff 133 Oct 24 19:40 ssh_config
-rw-rw-rw-@ 1 elsaco staff 2670 Oct 24 19:42 state-store.json

and sample ssh_config:

elsaco@m4max config % cat ssh_config

Host spark
Hostname spark
User tux
Port 22
IdentityFile “/Users/elsaco/Library/Application Support/NVIDIA/Sync/config/nvsync.key”
5

@rickeerickson If you remove any wildcard host matches from you ~/.ssh/config it should work. I had something like this:

Match Host *.lan
    User username

Which caused it to fail completely.

For the Nvidia developers. I ran nvsync-arm64 connect <hostname> and that gave me the following error:

{
“heading”: “Unexpected Error”,
“details”: “problem parsing ssh config file: (5, 7): ssh_config: Match directive parsing is unsupported”
}

Since you are using golang under the hood, my PR is still hanging in the wind in the ssh_config parser: Add support for Match (Fixes #6) by sio2boss · Pull Request #67 · kevinburke/ssh_config · GitHub so you might want to fork from there.

6

I had the same error, following a system-restore image on my DGX-Spark. Removed entry from ~/.ssh the old entry, but Nvidia-Sync was still holding on to the old hostname for some reason.

I tried to delete the old DGX-Spark reference from nvsync menu, but that didn’t remove it from the ~/Library/Application\ Support….\ssh_config file.

I added a new entry by IP address and that worked.