Operate NVIDIA DOCA Management Service daemons via gNMI and gNOI protocols.
npx skills add https://github.com/NVIDIA/skills --skill doca-dms
Please help me install this Agent Skill. Skill: doca-dms Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-dms Command: npx skills add https://github.com/NVIDIA/skills --skill doca-dms
Deploying and operating NVIDIA DOCA Management Service (DMS) to manage BlueField DPUs and ConnectX SmartNICs over gRPC is error-prone: operators must choose the right deployment shape, authentication mode, and protocol surface, and missteps with destructive gNOI operations can take production devices offline. This Skill guides agents through the documented DMS workflows so configuration, operation, and debugging follow the official public guide.
Ask your agent to help you bring up the DMS daemon on a BlueField host and issue a gNMI Get against a documented interface path.
High-intent search queries and answers about installing and using this skill.
Deploy the dmsd daemon on the appropriate shape (host, BlueField Arm, or Kubernetes pod), choose an authentication mode, then issue gNMI Get/Set requests for configuration or gNOI operations for system tasks. The public DMS guide is the authoritative source for flags and paths.
DMS documents four modes: localhost-only (development only, never expose externally), PAM, credentials, and mTLS. Read the Security Best Practices section of the public DMS guide before committing, and configure the -allowed_users list as the gRPC client authorization boundary.
Yes, DMS implements gNMI Subscribe with STREAM mode (SAMPLE intervals of 1s-60s) and ONCE mode; POLL and Aggregation modes are unimplemented. For a productized telemetry-aggregation surface, use the separate DOCA Telemetry Service instead.
The -allowed_users flag (default root) is the gRPC client authorization boundary enforced by dmsd across all authentication modes. The dmsgroup Unix group is a separate gate used only by the privileged dmspe backend helper, not the boundary gRPC clients see.
Debug in layers: first check transport (daemon running, listener, TLS), then authentication/authorization, then whether the path or operation is supported, then the backend tool layer (e.g. mlxconfig errors), and finally state persistency. Read frontend dmsd logs before backend dmspe logs.
Do not use it for installing DOCA itself, building DOCA library applications, or turnkey telemetry aggregation. Route those to doca-setup, doca-programming-guide with the matching library skill, or the DOCA Telemetry Service respectively.
air-gappedAutomate NVIDIA datacenter GPU bring-up on Ubuntu 24.04 with health validation.
armoin2018Automate ESP32/ESP8266 device lifecycle management across USB and network connections.
mietheOrchestrate validated software releases across version files, API specs, SDKs, changelogs, and GitHub.
stolyarchukGuide C++ engineers in designing and optimizing DOCA Flow pipelines on NVIDIA SmartNICs.
