Concepts
This reference design is built on Infrahub. If you come from an AVD-and-files background, these are the platform terms used throughout the docs. Each links to the upstream Infrahub documentation for full detail.
Branch​
A named, isolated line of change over the whole data model — like a Git branch, but for the graph. Every change set in this reference design happens on a branch: the generator chain and the service portal both create and operate on branches, so parallel work does not interfere. See Infrahub branches.
Proposed change​
A request to merge a branch into main, with a full diff of the data and the rendered artifacts it affects. Reviewers see exactly what a change does — new devices, reallocated addresses, changed EOS config — before it reaches production. This is the review gate for every fabric change. See proposed changes.
Generator​
A Python routine that reads high-level design intent and creates the many detailed objects it implies — devices, interfaces, links, and address/number allocations. In this reference design, FabricGenerator, PodGenerator, and RackGenerator expand a fabric design into the full technical model, and the AVD generators assemble per-device host_vars and structured configuration. Generators are idempotent (see below). See the Generators reference and Infrahub generators.
Transform​
A routine that converts stored data into an output document, saved as an artifact. Here, transforms render EOS CLI configuration, fabric and per-device Markdown documentation, the cabling-plan CSV, the ANTA test catalog, and the ContainerLab topology — with PyAVD running inside Infrahub's workers. See the Transforms reference.
Check​
A Python routine that Infrahub runs during proposed-change validation. It writes no artifact — it reports pass, information, or error, and an error blocks the merge. This reference design ships one: cv-config-validation, which deploys the rendered EOS configs into a CloudVision workspace and reports CloudVision's build result back to the proposed change. See the Checks reference and CloudVision Validation.
Artifact​
A generated, downloadable output stored in the object store and versioned with the data that produced it — a rendered EOS config, a documentation file, a cabling CSV. An artifact regenerates when its underlying data or its transform changes, so it never drifts from the source of truth. The AvdArtifact node tracks each device's host_vars and structured config with checksums. See AvdArtifact & File Storage.
Resource pool​
A managed range — IP prefixes, IP addresses, BGP ASNs, or node IDs — that hands out unique values on demand. Pools in this reference design are branch-aware, so two engineers working on different branches never allocate the same prefix, ASN, or node ID. See Infrahub resource manager.
Checksum-based idempotency​
Each generator records a checksum of the objects it depends on. On re-run, if nothing relevant changed, the generator skips its work; if only part of the design changed, it regenerates only the affected objects. This makes re-running after a partial failure safe, and limits each day-two change to the objects it actually affects. See Regenerate a Fabric.