AutoGineer
Generated banner for Doca Bare Metal Deployment
Generated logo for Doca Bare Metal Deployment

Doca Bare Metal Deployment

NVIDIAOfficialCreated by NVIDIA

  • Skill
  • Workflow
  • Cloud and DevOps
  • Apache-2.0
  • Medium risk
Download ZIP

From the author

Running a DOCA-linked binary directly on the hardware, with no container or kubelet: on host x86 with a BlueField NIC over PCIe, or bare-metal on BlueField Arm. Use this skill for launching and supervising it (direct, tmux, systemd), and when it "exits 1 right after launch", "systemd keeps restarting it", reports "no matching device on the BF", or the BlueField itself is unhealthy: "bfb-install exited 0 but DPU is dead", "ping 192.168.100.2 works but ssh fails", "host PFs aren't showing netdevs". Refuse service containers under kubelet (doca-container-deployment). Covers PCI/NUMA/CPU/IRQ binding, co-tenant isolation (cgroup-v2/netns/numactl), a seven-layer error taxonomy, platform lifecycle (BFB install, RShim/TMFIFO, host PF rebind, post-BFB recovery) and a six-state BlueField lifecycle classifier. Destructive firmware burn / mlxconfig set requires explicit confirmation via doca-hardware-safety; library APIs, env prep, and build use other skills.
— the description by NVIDIA

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

This skill provides a structured approach for running DOCA-linked application binaries directly on host x86 systems with BlueField NICs or on BlueField Arm cores. It covers the full lifecycle of bare-metal deployment, including launch modes, hardware resource binding, per-tenant isolation, and troubleshooting. It is designed for operators who need to manage binaries outside of containerized environments, focusing on stability, performance, and correct hardware interaction.

What it does

  • Guides the configuration and launch of DOCA-linked binaries on bare-metal hardware.
  • Provides methods for process supervision using direct execution, tmux, or systemd.
  • Details hardware-resource binding for PCIe, NUMA, CPU, and IRQ affinity.
  • Implements per-tenant isolation using cgroup-v2, network namespaces, and numactl.
  • Offers a seven-layer error taxonomy for diagnosing launch and runtime failures.
  • Manages BlueField platform lifecycle including BFB installation and recovery.

When to use it

  • When deploying a DOCA-linked binary directly on hardware without containers or Kubernetes.
  • When diagnosing launch failures, immediate exits, or device-access issues on BlueField.
  • When setting up process isolation for multiple co-tenant DOCA applications.
  • When determining the appropriate supervision strategy for long-running DOCA services.
  • When recovering a BlueField DPU from an unhealthy state or failed BFB installation.

Advantages

  • Ensures optimal performance through precise hardware and NUMA binding.
  • Prevents resource contention between multiple DOCA processes on the same hardware.
  • Provides a systematic, layered approach to debugging complex hardware-level failures.
  • Standardizes deployment procedures across different host and DPU environments.
  • Promotes safe operational practices for high-stakes bare-metal environments.

Usage examples

“My DOCA-linked binary exits with status 1 immediately after launch. How do I debug this?”
The agent will guide you through the seven-layer error taxonomy to isolate whether the failure is due to device binding, resource constraints, or environment configuration.
“Should I run my DOCA application in tmux or as a systemd unit?”
The agent will provide a decision framework based on your requirements for persistence, logging, and automatic recovery after reboots.
“How do I pin my DOCA process to a specific NUMA node to improve performance?”
The agent will provide the correct commands for CPU and NUMA binding using numactl and taskset to ensure the process aligns with the BlueField PCIe topology.
“My BlueField DPU is unresponsive after a BFB install. What are the recovery steps?”
The agent will walk you through the BlueField lifecycle classifier and the post-BFB recovery sequence to restore the DPU to a healthy state.

Install

  1. Download the skill as a ZIP:

    https://autogineer.com/agent-skills/nvidia/doca-bare-metal-deployment/download

  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip doca-bare-metal-deployment.zip -d .agents/skills

Documentation by NVIDIA

DOCA bare-metal deployment

Examples

  • "The binary exits 1 right after launch." — this is in scope. Treat it as a launch / device-bind / isolation failure before rewriting application code.

  • "systemd keeps restarting my DOCA app." — a restart loop on hardware is high-stakes. Debug the first exit before enabling Restart=always.

Instructions

Follow Loading order below. Open the tasks companion for launch / supervise / recover, and the capabilities companion for the runtime contract. Do not open every companion file.

Where to start: This skill is the bundle's home for operating a DOCA-linked application binary directly on hardware — no container, no kubelet, no static-pod manifest. It is the parallel of doca-container-deployment/SKILL.md for the non-container path. If the user has a DOCA-linked binary they built (per the canonical workflow in doca-programming-guide/SKILL.md) and they want to know how to actually run it on the host or on the BlueField Arm cores correctly, open the tasks companion and start at ## configure. If the question is what shape does the bare-metal runtime even have and what is the deployment contract, start at the capabilities companion. If the user is not yet sure whether their target system shape is the container path or the bare-metal path, route the recognition step to doca-setup/SKILL.md first; only return here once bare-metal is the confirmed shape.

Audience

This skill serves external DOCA developers and operators who have a DOCA-linked application binary they built and want to run it directly on hardware — i.e., people who already have:

  • a DOCA-linked application binary they built per doca-programming-guide ## build,

  • a real BlueField NIC and a host that talks to it (the host x86 path — DOCA host install on the host talks to the BlueField NIC over PCIe), OR a BlueField with a console or SSH to the Arm side (the BlueField Arm bare-metal path — DOCA installed on the DPU Arm cores; the binary runs there directly), and

  • a desire to RUN that binary directly on the hardware, not inside a kubelet-standalone-managed container.

It is not for:

  • kernel-driver developers contributing to mlx5_* or the BlueField OS,

  • DOCA library contributors (library changes are outside the scope of a bare-metal deployment),

  • full-Kubernetes-cluster operators managing a fleet of BlueFields (the bundle covers doca-container-deployment/SKILL.md for the single-host kubelet-standalone shape; fleet/production-scale deployment is fleet-orchestration scope — route to the orchestration entry-point in doca-public-knowledge-map ## Deploying DOCA services at scale (DPF / Network Operator / Launch Kit), not hand-rolled static-pod loops),

  • fresh-laptop-no-hardware users with no DOCA install yet — those belong on doca-setup ## no-install.

The skill teaches the agent the bare-metal-deployment procedure and the rules for quoting documented commands from the public DOCA Programming Guide and the public BlueField / DPU User Manual via doca-public-knowledge-map/SKILL.md; it does not invent flag names, PCI BDFs, NUMA numbers, devlink paths, representor strings, or systemd Restart= mode names from memory.

When to load this skill

Load this skill when the user is doing hands-on bare-metal deployment of a DOCA-linked application binary on either of the two supported host modes (host x86 or BlueField Arm), or asking a cross-cutting bare-metal question that is not specific to one library's API. Concretely:

  • Launching a DOCA-linked binary for the first time on a host with a BlueField NIC in a PCIe slot, with DOCA installed on the host.

  • Launching a DOCA-linked binary on the BlueField Arm cores directly (BlueField Arm bare-metal mode), with DOCA installed on the Arm side per the BlueField OS image.

  • Deciding which launch mode to use (direct foreground for interactive debug; tmux/screen for long-running with manual reattach; systemd-supervised for restart-after-reboot, journald-integrated logs, and Restart= policy).

  • Binding the DOCA process to the right PCIe function, the right representor, the right NUMA node, and the right CPU set — and pinning IRQs to match — without inventing the addresses or the flag names.

  • Setting up per-tenant isolation (cgroup-v2 cpu / memory / io controllers, network namespaces for multi-tenant deployments, numactl / taskset for CPU + NUMA binding) so multiple DOCA processes co-tenant on the same BlueField without crushing each other.

  • Diagnosing a bare-metal launch that is misbehaving — won't start, starts and exits immediately, runs but can't find the device, attaches to the device but the workload errors, OOMs or is signal-killed, is in a restart loop under a supervisor, or is being interfered with by a co-tenant.

  • Cross-cutting questions: "should I run this in tmux or as a systemd unit", "what is the smoke-before-bulk loop for a binary on bare metal", "my binary works in a container on the BlueField but not when I run it directly on the Arm — what changed".

Do not load this skill for the container-path equivalent (those questions go to doca-container-deployment/SKILL.md); for full-Kubernetes-cluster operations (out of scope per the bundle's non-goals); for library-API questions (route to the matching libs/<library> skill); for env-preparation questions including hugepages, IOMMU, pkg-config, and devlink mode flips (use doca-setup/SKILL.md); for any hardware-state-changing operation including mlxconfig writes and BFB reflashes (route to doca-hardware-safety/SKILL.md for the cross-cutting meta-policy); or for cross-library programming questions (use doca-programming-guide/SKILL.md).

What this skill provides

This is a thin loader. Substantive material lives in two companion files:

  • the capabilities companion — the bare-metal deployment runtime contract for a DOCA-linked binary: the two host modes (host x86 vs BlueField Arm bare-metal), the three launch modes (direct, tmux/screen, systemd-supervised), the hardware-resource-binding surface (PF / VF / representor enumeration; NUMA topology discovery; CPU pinning rationale; IRQ affinity rules), the per-tenant isolation surface (cgroup-v2 cpu / memory / io, network namespaces, numactl / taskset), the restart and recovery semantics (documented systemd Restart= modes vs crash-and-investigate vs supervisor-driven restart), the bare-metal-specific version overlay on the four-way version match owned by doca-version/SKILL.md, the cross-cutting error taxonomy (seven layers, walked in order), the observability surface (stdout/stderr discipline by launch mode; device-state introspection via devlink / sysfs / mlxconfig query; per-tenant resource visibility), and the safety policy (overlay on doca-hardware-safety/SKILL.md: smoke-before-bulk for binaries; failed bare-metal process is HIGH-STAKES; do not invent PCI addresses, NUMA numbers, representor names, devlink paths, or systemd Restart= mode names; confirm tenant-isolation primitives BEFORE the workload starts).

  • the tasks companion — step-by-step workflows for the in-scope bare-metal verbs: configure, build, modify, run (with an explicit ### isolation sub-anchor covering cgroup-v2 / namespaces / numactl per-tenant primitives), test, debug, bluefield-lifecycle (the BFB-install → RShim/TMFIFO → post-BFB-recovery operational sequencing ladder, with the six-state bluefield-state-classifier sub-anchor), the Command appendix (documented commands the agent may quote, each cross-linked to its public-doc source — no invented commands), and the Deferred task verbs block routing container-path / cluster / library-API / env-prep / hardware-state-change / cross-library questions out to their owning skills. (The change-application discipline for any mutating burn invoked from ## bluefield-lifecycle is still meta-policy owned by doca-hardware-safety/SKILL.md, loaded alongside.)

The skill assumes a host or BlueField target where:

  • DOCA is already installed and healthy (per doca-setup ## test),

  • the user has a DOCA-linked application binary they built (per doca-programming-guide ## build),

  • the user has the host-OS permissions to enumerate devices, reserve hugepages, write systemd units (if they choose that launch mode), and bind processes to NUMA nodes.

It does not cover installing DOCA — that path goes through doca-setup/SKILL.md — and it does not cover building the binary — that path goes through doca-programming-guide/SKILL.md.

Loading order

  1. Read this skill first to confirm the user's question is in scope (bare-metal launch of a DOCA-linked binary on host x86 or BlueField Arm; NOT the container path, NOT a full cluster, NOT a library-API question).

  2. For the runtime contract (two host modes, three launch modes, hardware-binding surface, per-tenant isolation, version overlay, seven-layer error taxonomy, observability surface, bare-metal safety overlay), see the capabilities companion.

  3. For step-by-step workflows — configure, build (routing stub), modify (routing stub), run (with ### isolation sub-anchor), test, debug, bluefield-lifecycle (BFB install + RShim/TMFIFO + post-BFB recovery + the six-state bluefield-state-classifier), plus the Command appendix and the Deferred task verbs block — see the tasks companion.

Example questions this skill answers well

What this skill deliberately does not ship

Related skills

Files

references/details.md

doca-bare-metal-deployment — reference detail

Moved out of SKILL.md to keep the loader under the per-file size budget. This is supporting detail, not routing logic.

Example questions this skill answers well

The CLASSES of bare-metal-deployment questions this skill is built to answer, each with one worked example. The class is the load-bearing piece; the worked example is one instance.

  • "I have a DOCA-linked binary I built. What does it actually take to run it correctly on real hardware — not inside a container?" — worked example: "I built a doca-flow application on my host with a BlueField-3 in the PCIe slot; how do I launch it the right way?". Answered by the pattern overview + launch-mode table in CAPABILITIES.md ## Pattern overview

  • "I want to run my binary on the BlueField Arm cores themselves, not on the x86 host. Is that the same workflow or a different one?" — worked example: "the BlueField OS image has DOCA installed on the Arm side; I'd like to run my DOCA app directly on the DPU, talking to its local NIC". Answered by the two-host-modes contract in CAPABILITIES.md ## Capabilities and modes

  • "Should I just ./my-doca-app & it, run it in tmux, or wire a systemd unit?" — worked example: "I want this binary to come back automatically after a host reboot, but I also want to be able to attach to it and see what it is doing right now". Answered by the three-launch-modes decision table in CAPABILITIES.md ## Capabilities and modes

  • "How do I bind my DOCA process to the right PCIe function and the right NUMA node so it doesn't trip over itself?" — worked example: "the BlueField is on NUMA node 1; my app is being scheduled on cores from node 0 and performance is terrible". Answered by the hardware-binding rules in CAPABILITIES.md ## Capabilities and modes

    ## run (cgroup-v2 / namespaces / numactl per-tenant primitives).

  • "My binary won't start; or it starts but exits immediately; or it starts but can't see the device. How do I diagnose this without guessing?" — worked example: "my doca-flow binary exits with status 1 within a second of launch; I have no idea which layer broke". Answered by the seven-layer error taxonomy in CAPABILITIES.md ## Error taxonomy

  • "systemd put my DOCA binary in a Restart=always loop because it keeps crashing. Should I let it keep restarting, or is that exactly the wrong thing?" — worked example: "the unit is auto-restarting my binary every five seconds and the device is reporting odd errors; should I just bump the restart limit?". Answered by the restart-loop-is-HIGH-STAKES rule in CAPABILITIES.md ## Safety policy

  • "Two of my colleagues are running DOCA processes on the same BlueField. How do I make sure their workload doesn't crush mine?" — worked example: "I want one DOCA-Flow process per representor, one DOCA-RDMA process for the storage path, all on the same BlueField, without cross-tenant interference". Answered by the per-tenant isolation rules in CAPABILITIES.md ## Capabilities and modes

  • "My host is fine and the BlueField was working last week, but after a BFB push it never came back. bfb-install exited 0, but I cannot ssh to the BF, ping 192.168.100.2 works but feels wrong, and ip link doesn't show any BlueField netdev on the host any more." — worked example: "DOCA 3.3 host upgrade is fine; BFB install on the BlueField reported Ubuntu installation completed then INFO[MISC]: NIC firmware update failed, but bfb-install still exited 0; now the DPU never reaches DPU is ready, host PFs are present in lspci -d 15b3: but ip link doesn't list their netdevs." Answered by the BlueField lifecycle anchor in TASKS.md ## bluefield-lifecycle (the bfb-install partial-failure recognition + the 192.168.100.2 host-loopback ip route get gotcha + the host PF rebind sequence + the post-BFB four-way version-match re-close) and the six-state classifier in ### bluefield-state-classifier.

What this skill deliberately does not ship

This skill is agent guidance, not a templates / sample-binaries / sample-units bundle. To keep the boundary clean, it deliberately does not contain — and pull requests should not add:

  • Pre-baked binaries. No DOCA application binary, no sample ELF, no statically-linked test program is shipped with this skill. The canonical artifact is the user's own DOCA-linked binary, built per doca-programming-guide ## build (doca-programming-guide/TASKS.md#build).

  • Sample systemd units, sample numactl invocations, sample taskset invocations, or any other ready-to-copy launch recipe. Bare-metal launch is deployment-specific (per-host PCI BDF, per-host NUMA topology, per-tenant CPU set, per-site systemd policy) and the safe answer for an external operator is to derive the launch recipe from the public DOCA Programming Guide and the public BlueField / DPU User Manual against their own target. The agent's job is to prescribe the procedure and quote the documented command shapes from doca-public-knowledge-map/SKILL.md, not to ship a .service file or a numactl --cpunodebind=... line the user might run unmodified.

  • PCI addresses, NUMA node numbers, representor names, devlink paths, hugepage allocation amounts, or systemd Restart= mode names invented from generic Linux knowledge. The public DOCA Programming Guide, the public BlueField / DPU User Manual, the Linux man pages (numactl(8), taskset(1), systemd.service(5), systemd.unit(5)), and --help on the installed tool are the authoritative sources. Inventing a 0000:01:00.0 or a Restart=on-failure-with-burst-cap from memory is the load-bearing first-run failure for this skill.

  • A samples/, templates/, units/, or reference/ subtree of any kind. A mock or incomplete artifact in this skill's tree, even one labeled "reference", is misleading: operators will read it as production-ready.

Related skills

  • doca-container-deployment/SKILL.md — the SIBLING path. Two parallel deployment shapes in this bundle: containers (that skill) vs bare metal (this one). The recognition step that picks between them lives in doca-setup/SKILL.md. Once the shape is bare metal, the agent stays here; if it is container, the agent routes there.

  • doca-setup/SKILL.md — env preparation (install verification, hugepages mount and reservation, IOMMU posture, devlink mode, pkg-config path, representor visibility, kernel module load state). This skill assumes its preconditions are satisfied at the bare-metal target. The recognition step that decides container-vs-bare-metal is in doca-setup per the bundle convention; load doca-setup in parallel when the user's situation is ambiguous.

  • doca-hardware-safety/SKILL.md — the cross-cutting meta-policy for any change touching DPU / NIC hardware state. This skill's ## Safety policy overlays that meta-policy with bare-metal-specific rules (smoke-before-bulk-for-binaries, restart-loop-is-HIGH-STAKES, do-not-invent-PCI-addresses-or-NUMA-numbers-from-memory) and does not redefine the meta-policy itself. When the change the agent is about to recommend writes mlxconfig, burns firmware, reflashes the BFB, flips the BlueField mode, or changes a kernel boot parameter, the agent leaves this skill for doca-hardware-safety and only returns once the hardware-state change is complete.

  • doca-debug/SKILL.md — the cross-cutting layered debug ladder (install / version / build / link / runtime / program / driver). Bare-metal-deployment-specific debug (process didn't start, started and exited, couldn't find the device, OOM / signal, restart loop, co-tenant noise) layers on top of the cross-cutting ladder; this skill's ## debug cross-links into doca-debug for the broader context.

  • doca-programming-guide/SKILL.md — canonical DOCA build / modify / first-app patterns and the cross-library DOCA_ERROR_* taxonomy. This skill assumes the user already has a built binary; questions about building the binary or interpreting library-specific errors route there.

  • doca-version/SKILL.md — the four-way version match rule (host package ↔ binary build ↔ BlueField firmware ↔ DOCA-version policy). This skill's ## Version compatibility cross-links the body of the rule there and adds only the bare-metal-specific overlay (the binary's link-time pkg-config doca-* version must match the runtime LD_LIBRARY_PATH'd install).

  • doca-public-knowledge-map/SKILL.md — the routing table to the public DOCA Programming Guide, the public BlueField / DPU User Manual, the public Installation Guide, and the NGC catalog. This skill does not duplicate URLs; it points at the map and adds the bare-metal-deployment overlay.

  • doca-structured-tools-contract/SKILL.md — the bundle's structured-tools precedence rule (detect / prefer / fall back / report). The ## Command appendix in TASKS.md honors this contract — the agent probes for the matching structured helper first (doca-env --json, doca-capability-snapshot, version-matrix.json) and falls back to the documented manual commands when the probe fails.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Node.js

The author says: No DOCA install required to read this skill (it is an overlay loaded against any DOCA artifact skill); the validation steps within this skill require a live DOCA install at /opt/mellanox/doca on a host or BlueField with a built DOCA-linked binary.

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by
NVIDIA
Publisher
NVIDIA (Official)
Source
NVIDIA/skills · skills/doca-bare-metal-deployment · commit fce3cca
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE-APACHE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by NVIDIA or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download doca-bare-metal-deployment.zip