Doca Programming Guide

DOCA programming guide

Examples

• "How do I write my first DOCA program?" — copy a shipped sample and follow the modify workflow, then the canonical pkg-config + meson build. Docker / NGC can compile and link; runtime needs a real BlueField or ConnectX.

• "What's the lifecycle every DOCA object follows?" — create → configure → start → connect → use → stop → destroy. Configuration finishes before start.

Instructions

Follow Loading order below. Open the tasks companion for how-to and the capabilities companion for coverage. Do not open every companion file.

Where to start: Read ## Audience to confirm the user is consuming DOCA, not contributing to it. Then jump to the H2 that matches the verb (## modify for first-app derivation, ## build for the canonical build pattern, ## test for the test loop, ## debug for the program-class debug ladder). Docker / NGC is the build half, not the runtime half. The NGC container can compile and link. Flow pipe creation, RDMA QP ops, and any call that needs a doca_dev on a port need a real BlueField or ConnectX. Do not promise the app will work entirely inside docker.

Example questions this skill answers well

These are the CLASSES of program-class questions the skill is built to answer, each with one worked example. Library-specific overlays (Flow / DMS / Caps / …) live in the matching library skill; this skill answers the library-agnostic shape.

• "How do I write my first DOCA program for <any library>?" — worked example: "I have docker; help me build my first DOCA Flow app." Answered by the modify-a-shipped-sample workflow in

  modify plus the canonical build

  pattern in ## build. Say that docker / NGC builds and links, but runtime needs a real NIC or DPU.

• "What's the right build line for any DOCA library?" — worked example: "How do I compile a program that calls doca_rdma_*?" Answered by the pkg-config doca-<library> pattern in

  build (C/C++ Track 1) and the

  FFI/bindings pattern in Track 2.

• "What's the lifecycle every DOCA object follows?" — worked example: "What's the right order of doca_flow_pipe_* calls in my program?" Answered by the Core order create → configure → start → connect → use → stop → destroy in ## Capabilities and modes (cfg-create / init implement configure and finish before start).

• "DOCA_ERROR_* came back — what does it mean and what do I do?" — worked example: "My code got DOCA_ERROR_BAD_STATE." Answered by the cross-library doca_error_get_descr() rule in

  Error taxonomy

  • the program-class debug order in

  debug.

• "My program built and started, but does nothing on the wire." — worked example: "My Flow program runs cleanly but no traffic is matched." Answered by the validate-before-commit rule in

  Safety policy

  and the layered program-class debug ladder in

  debug.

• "What does <language> consumer of DOCA look like (FFI / bindings)?" — worked example: "How do I call DOCA Comch from Rust without writing C?" Answered by Track 2 of

  build (FFI against the public C

  ABI) and the language-neutral lifecycle in

  Capabilities and modes.

• "How should I classify and build all the shipped DOCA samples and applications? What's the difference between a sample and an application?" — worked example: "I tried to build all DOCA apps and some failed — which ones are real regressions vs missing optional stacks?" Answered by the sample-vs-application model and the category / dependency / skip-vs-fail taxonomy in

  sample-and-app-categorization,

  which separates "the SDK is broken" from "the optional GPU / RMAX / MPI stack is not on this BlueField".

If the question is env-class (install / build env / hugepages / devices), route to doca-setup/SKILL.md. If it is library-specific (Flow pipe topology, RDMA QP setup, DMS service deploy), layer the matching library skill on top.

Audience

This skill serves external developers building applications that consume DOCA libraries — i.e., users whose code calls one or more doca_<library>_* symbols (directly in C/C++, or through FFI / bindings from another language). It is programming with DOCA, not programming of DOCA: it is not for NVIDIA developers contributing to DOCA itself, and it does not assume access to the DOCA source tree, internal NVIDIA tooling, or any non-public information. The only inputs it ever points the agent at are the ones any external user has: the public docs at docs.nvidia.com/doca/sdk/, the public catalog at catalog.ngc.nvidia.com, the public GitHub repos under github.com/NVIDIA / github.com/NVIDIA-DOCA, the public developer forum, and the on-disk /opt/mellanox/doca tree that the public DOCA install (or the public NGC DOCA container, nvcr.io/nvidia/doca/doca) puts on the user's host. Where to find and how to install questions are routed elsewhere — see Related skills below.

Language scope. DOCA itself is a C library family; every shipped sample in /opt/mellanox/doca/samples/ and every shipped reference application in /opt/mellanox/doca/applications/ is C. C and C++ consumers are the canonical case for every prescriptive workflow in this skill. Other-language consumers (Rust, Go, Python, …) consume the same *.so libraries through FFI or language-specific bindings against the public C ABI; the skill keeps the lifecycle, capability, error, observability, and safety guidance language-neutral, and routes the language-specific build / FFI work back to the consumer's own toolchain without authoring wrappers.

When to load this skill

Load this skill when the user has DOCA installed and the env-class preconditions are already satisfied (i.e., doca-setup/SKILL.md has produced a clean install where pkg-config doca-<library> resolves, hugepages are mounted, and devices are visible), and is now asking a question about how to actually program against DOCA in a library-agnostic way:

• Understanding what DOCA's pieces are (libraries, apps, services, tools) and which side of the wire they run on.

• The canonical pkg-config + meson build pattern any DOCA application follows, regardless of which library it consumes.

• The universal derive a custom first application from a shipped sample workflow that every library skill extends with library-specific overrides — moved here from doca-setup because it is a programming verb, not an env verb.

• The universal DOCA Core lifecycle (create → configure (including permissions and task set_conf) → start → connect → use → stop → destroy) and how it manifests across libraries. Configuration MUST finish before doca_ctx_start(). Connect is its own step after start; do not write connect/use as one token.

• The general DOCA_ERROR_* pattern, doca_error_t, and doca_error_get_descr() — the cross-library shape, not Flow- or RDMA-specific overlays.

• The validate-before-commit rule and the program-side safety policy that every library skill inherits.

• Programming patterns that apply to consumers in any language (C/C++ directly, FFI / bindings for Rust / Go / Python / …) — the skill keeps the patterns language-neutral and points the agent at the public C ABI as the authoritative surface.

Do not load this skill for:

• "Is my install healthy? Why does pkg-config not find doca-flow? How do I mount hugepages? I don't have DOCA installed yet — can I use a container?" — env-class questions belong in doca-setup/SKILL.md, which owns install verification, env preparation, env-class debugging, and the no-install → NGC container fallback path for any user on macOS, Windows, or Linux without DOCA.

• "What is DOCA? Where is the Flow programming guide? Which package do I install?" — routing / orientation questions belong in doca-public-knowledge-map/SKILL.md.

• "How do I construct a Flow pipe / set up an RDMA queue / use Comch to send a message?" — library-internal API questions belong in the matching library skill (e.g. doca-flow/SKILL.md).

What this skill provides

This is a thin loader. The body keeps only the orientation needed to pick the right next file. The substantive material lives in two companion files:

• the capabilities companion — what every DOCA program looks like in the abstract: the shape of DOCA (host / DPU / switch, libraries / apps / services / tools, build flavor selection rationale), the universal program lifecycle, the unified version-compat rule that applies to any program linking DOCA, the cross-library DOCA_ERROR_* taxonomy, the program-side observability surface (DOCA logging, capability snapshots), and the program-side safety policy that every library skill inherits.

• the tasks companion — step-by-step workflows for the six in-scope programming verbs: configure, build, modify, run, test, debug. The ## modify verb owns the universal derive a custom first app from a sample pattern that every DOCA library skill extends with library-specific overrides; the ## build verb owns the canonical pkg-config doca-<library> build pattern in two language tracks (C/C++ direct, non-C via FFI).

This skill assumes doca-setup/SKILL.md has already produced a clean install. It does not cover env preparation; that is doca-setup's job, including the no-install → NGC container (nvcr.io/nvidia/doca/doca) fallback for users on macOS, Windows, or Linux without DOCA.

What this skill deliberately does not ship

This skill is agent guidance, not a samples or templates bundle. It deliberately does not contain — and pull requests should not add:

• Pre-written DOCA application source code, in any language. This includes C / C++ files, Rust crates, Go packages, Python modules, and wrapper code for any other language. The DOCA API surface evolves between releases and code written from documentation prose cannot be verified without compiling / linking / FFI-loading it against the live library on a real install. The verified DOCA application source code is the shipped C samples on the user's installed system; the agent's job is to route the user to that file and prescribe a minimum-diff modification on it (## modify) — for C/C++ users — or to route non-C users to the public C ABI surface that their bindings will call (## build Track 2), not to author the wrapper.

• Standalone build manifests (meson.build, CMakeLists.txt, Cargo.toml, setup.py, go.mod, …) parked inside the skill. The agent constructs the build manifest in the user's project directory against the user's installed DOCA, where pkg-config --modversion doca-<library> is the source of truth.

• A samples/, bindings/, or reference/ subtree of any kind. A mock or incomplete artifact in this skill's tree, even one labeled "reference", is misleading: users will read it as buildable.

Loading order

1. Read this skill first to confirm the user's question is a programming-class question (not env, not routing, not library-API).

2. For the DOCA shape, the universal lifecycle, the cross-library error taxonomy, the program-side observability surface, and the safety policy that every library skill inherits, see the capabilities companion.

3. For step-by-step programming workflows — configure, build, modify, run, test, debug — see the tasks companion.

If the user is asking for a first app in a specific library, walk through ## modify for the universal copy-and-edit pattern, then hand off to the library skill (e.g. doca-flow/SKILL.md) for the library-specific values to swap.

Related skills

• doca-public-knowledge-map/SKILL.md — public DOCA documentation routing and the on-disk layout of an installed DOCA package. This skill defers all "where is X documented", "where on disk is Y", and "how do I check the installed version" questions to the knowledge-map.

• doca-setup/SKILL.md — env preparation, install verification, env-class debugging, and the I have no install yet procedure with the NGC container (nvcr.io/nvidia/doca/doca) as the universal Stage-1 fallback for any user on macOS, Windows, or Linux without DOCA. This skill assumes doca-setup's preconditions are already satisfied.

• doca-flow/SKILL.md — DOCA Flow on BlueField. Extends this skill's ## modify (universal first-app derivation) with the Flow-specific list of fields to swap.

• doca-container-deployment/SKILL.md — shipping an application that already builds: container manifests, rollout, and service-readiness verification. This skill ends once the program compiles and runs locally; it does not cover deploying it.