skills.rest
NVIDIA
NVIDIAOFFICIAL
Explore all skills by this author →
WebsiteGitHub

doca-flow-tune

Snapshot, analyze, and tune live or captured doca-flow pipelines with doca_flow_tune.

3.2k|370|Updated Feb 25, 2026
Search Tags:#tuning#doca#doca-flow#bluefield#pipeline-analysis#network-optimization#dpdk-alternative
One-click install
npx skills add https://github.com/NVIDIA/skills --skill doca-flow-tune
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: doca-flow-tune
Source: https://github.com/NVIDIA/skills/tree/main/skills/doca-flow-tune
Command: npx skills add https://github.com/NVIDIA/skills --skill doca-flow-tune

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Tuning a running DOCA Flow pipeline on BlueField DPUs or ConnectX NICs requires choosing the right tuning axis, measurement, and scope, then safely applying recommendations without disrupting the live dataplane. This Skill guides agents through invoking doca_flow_tune correctly so optimization decisions are grounded in captured evidence rather than guesswork.

Core Features & Use Cases

  • Snapshot and observe pipelines: Capture pipe, counter, and KPI state from a live doca-flow application over a Unix-domain socket, or analyze a previously captured config offline, producing dumper CSV, analyze JSON, and visualize mermaid outputs.
  • Three-axis tuning decisions: Commit to a tuning axis (rule placement, resource hints/table sizing, hardware-offload mode), a matching measurement (rule-install rate, lookup latency, hardware-counter delta), and an explicit scope before any invocation.
  • Safe recommendation workflow: Apply analyze-step recommendations back into the doca-flow program via minimum-diff code changes, gated by smoke-before-bulk testing, before/after snapshot pairs, and per-recommendation confirmation.
  • Use Case: A performance engineer with a doca-flow-perf baseline notices low rule-install rate on a BlueField-3 service; the Skill walks them through a read-only online snapshot, interprets the analyze recommendation, and proposes a table-sizing change to the Flow program with a rollback plan.

Quick Start

Ask your agent to snapshot the running doca-flow pipeline with doca_flow_tune and recommend a table-sizing change for the pipe with low rule-install rate.

Frequently Asked Questions about doca-flow-tune

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I tune a doca-flow pipeline with doca_flow_tune?▼

Configure a JSON file from the shipped flow_tune_cfg templates naming the scope and mode, run an online read-only snapshot against the live application, then read the analyze JSON and visualize mermaid outputs. Apply any recommendation back into the doca-flow program as a minimum-diff code change with a before/after snapshot pair.

What is the difference between doca-flow-tune and doca-flow-perf?▼

doca-flow-perf measures a defensible baseline number such as throughput or rule-install rate on a static pipeline, while doca_flow_tune observes a live or captured pipeline and recommends parameter changes on top of that baseline. Measure with perf first, then optimize with tune.

Does doca_flow_tune require a running doca-flow application?▼

Online modes require a live doca-flow application reachable over the configured Unix-domain socket, while offline mode consumes a previously captured pipeline-description JSON without contacting the application. The tool observes pipelines; it does not create them.

Why does doca_flow_tune produce empty visualize output or fail to attach?▼

Attach failures usually mean the Flow application is not running, the network.server_uds socket path is wrong, or a namespace boundary blocks access. Empty output can also indicate a version mismatch between the tune binary and the application's linked doca-flow library, so verify the four-way version match first.

Is applying a doca_flow_tune recommendation safe for a live dataplane?▼

Read-only snapshot and analyze modes never touch live Flow state, but applying a recommendation changes the Flow program and can disrupt in-flight traffic. The Skill mandates smoke-before-bulk testing, explicit per-recommendation confirmation, and a captured rollback snapshot before any state-changing step.

Related Skills

Explore More Software Engineering Skills→
NVIDIANVIDIA
14

k8s-network-engineer

Configure and troubleshoot NVIDIA networking deployments on Kubernetes.

Official
Advanced
Hopin-incHopin-inc
3

performance-diff

Compare pre- and post-change performance metrics to detect regressions.

Official
Advanced
stolyarchukstolyarchuk

doca-expert

Guide C++ engineers in designing and optimizing DOCA Flow pipelines on NVIDIA SmartNICs.

Community
Advanced
openshiftopenshift
71

investigate-ci-failure

Diagnose CI/Prow failures for GitHub PRs using gh CLI data.

Official
Advanced