NVLink Plugin

The NVLink plugin enables centralized monitoring and management of multiple NVLink domains through both the UFM UI and REST APIs. At its core is the NMX Aggregator (NMXAGGR), which connects to multiple NVLink domains, gathers data from their NMX Controllers (NMX-C), and consolidates information about monitored components. By default, the plugin includes a built-in NMXAGGR, but it can also be configured to connect to an external NMXAGGR instance—either on the same host or a different system. Communication with NVLink domains is performed via the NMX-C using a gRPC-based API.

1. Download the Plugin Image

   Run the following command to download the NVLink plugin image:

   Copy

   Copied!

               
   
               
   docker pull mellanox/ufm-plugin-nvlink
           

2. Load the Plugin into UFM

   After downloading, you can load the plugin into UFM using one of the following methods:

   • Via UFM UI:

     Navigate to Settings → Plugins Management in the UFM web interface.

   • Via Command Line:

     Execute the following command on the UFM server terminal:

     Copy

     Copied!

                 
     
                 
     /opt/ufm/scripts/manage_ufm_plugins.sh add -p nvlink
             

Container Volume Mapping

The UFM plugin management system creates the following mappings between the plugin docker container file system and the host machine one:

Any file system path mentioned in this document refers to the container's file system, unless stated otherwise.

NVLink Domains Connection Security

The plugin, specifically its NMXAGGR component, interacts with NVLink domains over a gRPC connection. In this setup, the domain controller (NMX-C) acts as the server, while NMXAGGR functions as the client.

NMXAGGR supports three modes of gRPC communication:

1. Insecure – No encryption is used. This is the default mode.

2. Server-side TLS – Communication is encrypted. Only the server needs to present a certificate to the client. This mode is enabled by setting the cacert option (refer to the Configuration section).

3. Mutual TLS (mTLS) – Communication is encrypted, and both the client and server must authenticate each other using certificates. This mode requires setting the cacert, cert, and key options (refer to the Configuration section).

Managed Domains List

All NVLink domains that are managed or monitored by NMXAGGR are recorded in a list stored in the file <data_dir>/domains.txt (see the Configuration section for the definition of data_dir).

This file serves as an alternative method—alongside the Web UI and REST API—for adding or removing managed domains. Each line in the file represents the gRPC endpoint address of a domain controller (NMX-C), including its port number.

Addresses may include hostnames or IP addresses, and both can incorporate numerical ranges to define multiple addresses in one line.

Examples:

• 10.222.16.333:9370

• nv-dmn-01:6666

• 10.222.[16,17,20-28].[330-350]:9346

• nv-dmn-[01-8]:9371

The plugin can be configured by editing the config file /config/nvlink_plugin.conf.

There are two sections in the config file:

nmxaggr

¹ A duration string is a sequence of decimal numbers, each with optional fraction and a unit suffix, such as 300ms, 1.5h or 2h45m . Valid time units are ns, us, ms, s, m, h .

logging

¹ 10 MB

After the plugin is activated, an "NVLink" section becomes available in the dashboards.

NVLink Dashboard View

This view presents an overview of inventory elements—such as domains, switches, and GPUs—along with a filter for their health status.

Users can drill down from overall status indicators to specific elements, and further into the individual ports or links associated with each selected element.

The user can select a specific domain, upon which a list of associated switches and GPUs will be displayed, as illustrated in the example below.

If the selected domain has any health issues, a detailed breakdown of the affected devices will also be presented.

When an unhealthy device is selected, a list of all its ports and links will be displayed.

Additionally, the "Recent Events" notification panel on the right side of the screen is updated with the most recent health status changes of the devices.

Managed Elements View

The Managed Elements view is a tree-tabular display that shows all inventory elements, allowing users to browse through them. It also provides the option to add or remove domains.

Domains View

Add New Domain Model

Click the + icon in the upper dashboard to add a new domain.

Available Actions for the Selected Domain

The following actions are available when you right-click on the selected domain's row.

Switches View

This screen presents a table listing all the switches, including key details.

Available Actions for the Selected Switch

The following actions are available when you right-click on the selected switch's row.

GPUs View

This screen presents a table listing all the GPUs, including key details.

Available Actions for the Selected GPU

The following actions are available when you right-click on the selected GPU's row.

Ports View

This screen presents a table listing all the ports, including key details.

Available Actions for the Selected Port

The following actions are available when you right-click on the selected port's row.

Links View

This screen presents a table listing all the links, including key details.

Available Actions for the Selected Link

The following actions are available when you right-click on the selected link's row.

This section contains short descriptions of available REST API endpoints. The more detailed documentation is available at /ufmRestV2/plugin/nmxaggr/v1/app/swagger endpoint of the running plugin.

App Operations

Managed Domains Operations

Inventory Operations