NVIDIA UFM Enterprise User Manual v6.23.1

Unmanaged Switch Dump (USD) Plugin

Changes and New Features

Plugin Version

Feature

1.1.0-2

Start a background task which collects the abnormal System Dump of all/specified unmanaged switches automatically


Bug Fixes

Plugin Version

Bug Fix

1.1.0-2

N/A


The NVIDIA Unmanaged Switch Dump (USD) plugin operates as a server that uses the "resourcedump" command to retrieve system collections from unmanaged NVIDIA switches, compiles the data into files, and transfers these files to the same remote destination used for data collection from managed switches.

  • The unmanaged switch must be running firmware version 31.2014.2100 (production) or later.

  • UFM version must be LTS 6.19.1.1 or higher.

Deploying the Plugin with UFM (SA or HA)

  1. Install the latest version of UFM.

  2. Start UFM using the command: /etc/init.d/ufmd start.

  3. Pull the plugin image from the specified repository.

  4. Enable the plugin by running /opt/ufm/scripts/manage_ufm_plugins.py with the options -p usd -t <version>, or alternatively, enable it through the UFM UI by navigating to Settings → Plugin Management, right-clicking on usd, selecting Add, choosing the desired version, and clicking Add.

  5. Verify the plugin is running by using docker ps or by checking the UI under Settings → Plugin Management to confirm it has been added and enabled.

    image-2025-5-8_8-58-58-1-version-1-modificationdate-1762688264393-api-v2.png

  6. If the default port for usd is in use, update the usd_httpd_proxy.conf configuration file and restart the plugin.

Deploying the Plugin on UFM-SDN Appliance

  1. Log in as the admin user and enter configuration mode:

    Copy
    Copied!
                

    enable config terminal

  2. Verify that UFM is running by executing:

    Copy
    Copied!
                

    show ufm status

    If UFM is not running, start it with:

    Copy
    Copied!
                

    ufm start

  3. Ensure Docker is active by disabling any shutdown state:

    Copy
    Copied!
                

    no docker shutdown

  4. Load the latest USD plugin Docker image:

    • If using an HA setup, repeat this step on the standby node.

    • If the appliance is connected to the internet:

      Copy
      Copied!
                  

      docker pull mellanox/ufm-plugin-usd

    • If the appliance is offline, use another internet-connected machine to save the image:

      Copy
      Copied!
                  

      docker save mellanox/ufm-plugin-usd:latest | gzip > ufm-plugin-usd.tar.gz

      Transfer the file to an SCP-accessible location, then on the appliance:

      Copy
      Copied!
                  

      image fetch scp://user@hostname/path-to-file/ufm-plugin-usd.tar.gz docker load ufm-plugin-usd.tar.gz

  5. Enable and start the plugin:

    Copy
    Copied!
                

    ufm plugin usd add

  6. Verify the plugin is running:

    Copy
    Copied!
                

    show ufm plugin

Collecting a System Dump via the UI

To collect a system dump from an unmanaged switch using the UI:

  1. Right-click on the target unmanaged switch.

    image-2025-5-8_9-3-9-1-version-1-modificationdate-1762688264643-api-v2.png

  2. Select Collect System Dump from the context menu.

While the system dump is being generated, a progress message will be displayed.

image-2025-5-8_9-3-20-1-version-1-modificationdate-1762688264900-api-v2.png

Once the process completes successfully, a confirmation message will appear, including the dump file path in the format:

<protocol>://<server>/<path>

Collecting System Dump Automatically

Besides of manually starting the collection, the plugin will regularly check for unmanaged switches and collect their system dumps in the background. This helps ensure important diagnostic data is gathered automatically, improving user experience and reducing manual work. The collection interval and supported switch models can be configured in the poller-config section of plugin settings.

Configurations

Logs Configurations

Parameter

Description

Default Value

logs_file_name

Full path of the log file .

/log/usd.log

logs_level

Sets the logging level for the plugin (e.g., INFO, DEBUG).

INFO

log_file_max_size

Maximum size of the log file before it rotates, measured in bytes

10485760

log_file_backup_count

Number of backup log files to retain after rotation.

5


Poller Configurations

Parameter

Description

Default Value

polling_first_delay_secs

Delay before the first polling cycle starts (in seconds). This allows the USD plugin and its web service to initialize properly.

5

polling_interval_secs

Interval between consecutive polling cycles (in seconds). A value of zero or negative disables the polling task.

300

polling_duration_secs

Total duration for which polling runs (in seconds). Zero means polling continues indefinitely until the plugin stops.

0

polling_matched_models

Comma-separated list of unmanaged switch models to poll. Leave blank to poll all unmanaged switch models.

MQM9790


Dump Command Configurations

Parameter

Description

Default Value

dump_cmd_timeout_secs

The timeout duration for the MFT (Mellanox Firmware Tools) command, measured in seconds. The maximum value is 900 seconds. If the value is left blank, set to 0, or assigned a negative number, the timeout will automatically default to 300 seconds.

300

If the operation fails, an error message will be shown.

In cases where the failure is related to the remote storage location, a specific error message indicating this issue will be displayed.

image-2025-5-8_9-3-42-1-version-1-modificationdate-1762688265503-api-v2.png

© Copyright 2025, NVIDIA. Last updated on Nov 20, 2025