dpfctl sosreport

The dpfctl sosreport command collects SOS reports from host and DPU cluster nodes using Kubernetes Jobs. It creates privileged Jobs on target nodes that run the NVIDIA sosreport tool to collect system diagnostics. Reports can be downloaded to your local machine or written to an NFS share.

Usage

dpfctl sosreport [command] [flags]

Available Commands:

Quick Start

The simplest way to collect SOS reports is to use the collect command, which orchestrates the full workflow (start → wait → download → cleanup) in a single step:

dpfctl sosreport collect --output-dir /tmp/sos-reports

Alternatively, you can run each step manually:

# Start collection Jobs
dpfctl sosreport start

# Check progress (use -w to watch continuously)
dpfctl sosreport status -w

# Download completed reports
dpfctl sosreport download --output-dir /tmp/sos-reports

# Clean up resources
dpfctl sosreport cleanup

Targeting Host or DPU Nodes

By default, SOS reports are collected from all nodes on both host and DPU clusters. Use the --target flag to limit collection to a specific environment:

# Collect from host cluster nodes only
dpfctl sosreport collect --target host --output-dir /tmp/sos-reports

# Collect from DPU cluster nodes only
dpfctl sosreport collect --target dpu --output-dir /tmp/sos-reports

To collect from specific nodes (comma-separated):

dpfctl sosreport start --nodes worker-1,worker-2

To target nodes by label selector:

dpfctl sosreport start --node-selector node-role.kubernetes.io/worker=

Both --nodes and --node-selector can be combined with --target to narrow the scope further.

If you have multiple DPU clusters, use --dpu-cluster to target a specific one:

dpfctl sosreport start --target dpu --dpu-cluster dpu-cplane-tenant1

NFS Output

Instead of downloading reports locally, you can write them directly to an NFS share:

dpfctl sosreport start --nfs-server 10.0.0.1 --nfs-path /exports/sos

A timestamped subdirectory (e.g., sosreport-20260416-120000) is created automatically. To write directly to the specified path without a subdirectory:

dpfctl sosreport start --nfs-server 10.0.0.1 --nfs-path /exports/sos --nfs-no-subdir

If the NFS server uses root_squash, use --nfs-uid to specify a non-root UID for creating directories and writing files on the NFS share:

dpfctl sosreport start --nfs-server 10.0.0.1 --nfs-path /exports/sos --nfs-uid 1000

Archive

Use the --archive flag to create a single .tar.gz archive of all reports, suitable for attaching to a support ticket. This works in both local download and NFS output modes:

# Archive after local download
dpfctl sosreport collect --archive --output-dir /tmp/sos-reports

# Archive on the NFS share
dpfctl sosreport start --nfs-server 10.0.0.1 --nfs-path /exports/sos --archive

Use --archive-only to create the archive and remove the individual report files afterwards:

dpfctl sosreport collect --archive-only --output-dir /tmp/sos-reports

The --archive and --archive-only flags are also available on the download subcommand:

dpfctl sosreport download --archive --output-dir /tmp/sos-reports

Case ID

A case ID is assigned to each collection run and used to label all created resources. By default, a timestamped ID (e.g., dpf-20260416-120000) is generated. You can specify your own:

dpfctl sosreport collect --case-id CASE-12345 --output-dir /tmp/sos-reports

Watching Status

Use dpfctl sosreport status -w to continuously monitor job progress. New lines are appended as job statuses change, similar to kubectl get -w:

dpfctl sosreport status -w

The default refresh interval is 5 seconds. Use -i to change it:

dpfctl sosreport status -w -i 10

The status, download, and cleanup subcommands accept --case-id to filter by a specific collection run:

dpfctl sosreport status --case-id CASE-12345
dpfctl sosreport download --case-id CASE-12345 --output-dir /tmp/sos-reports
dpfctl sosreport cleanup --case-id CASE-12345

Global Flags

These flags apply to all sosreport subcommands:

Last updated: June 24, 2026