Run this on your TOSCA topology and get a full architectural risk report in 30 seconds
pip install pyyaml pyvis
python tools/tosca_audit.py topologies/grid5000/toulouse.yaml --format htmlA lightweight Python toolchain that treats a TOSCA YAML file as a queryable knowledge base about your computing continuum, detecting scheduling risks, memory constraints, accelerator single points of failure, and ARM 32/64-bit incompatibilities before you deploy a single workload.
Managing heterogeneous computing continuums, cloud servers, edge nodes, FPGA boards, NPU accelerators , is hard. Container images fail silently on wrong architectures. K3s schedulers place pods on 512 MiB nodes that immediately OOM. A single NPU becomes a SPOF for your entire inference pipeline.
TOSCA (Topology and Orchestration Specification for Cloud Applications) can describe all of this in a single YAML file. This toolchain makes that file actionable.
python tools/tosca_audit.py topologies/continuum/clusters_topology.yaml
python tools/tosca_to_html.py topologies/grid5000/luxembourg.yaml
18 rules across 8 categories. Each rule has a trigger condition, detection logic, remediation guidance, and a real-world testbed example.
| Category | Rules | Example findings |
|---|---|---|
| ARCH | 4 | ARMv7 32-bit in K3s cluster, mixed ISA without taints |
| MEM | 3 | RAM < 1 GiB (OOM risk), 512× server/agent imbalance |
| ACCEL | 3 | Single NPU = SPOF, unique FPGA fabrics, PS/PL contention |
| ORCH | 4 | K3s cluster without server node, partial Liqo federation |
| REDUND | 2 | Single compute node per cluster, single ISA in continuum |
| SCALE | 1 | Scaling policy default equals maximum |
| CPU | 2 | Sub-1 GHz frequency, single-core nodes |
| COMPAT | 1 | ARM32 declared as K3s server (unsupported since v1.24) |
All rules operate on a three-tier ISA taxonomy applicable to any cloud-fog-edge continuum:
- Tier 1 — 64-bit server-class (
linux/amd64,linux/arm64): full K3s support, 64-bit address space. Architectures: x86_64, AArch64 (ARMv8-A). - Tier 2 — 32-bit embedded-class (
linux/arm/v7,linux/riscv32): dropped K3s v1.24+ support, 4 GB address space ceiling, in-order pipelines. Architectures: ARMv7-A, MIPS32. - Tier 3 — Accelerator-attached (no OS): FPGA, NPU, GPU — sub-nodes to Tier 1/2 hosts.
Four real HPC sites modelled from the Grid'5000 public hardware pages:
topologies/grid5000/
├── nantes.yaml ← 3 clusters, 74 nodes, 6x Nvidia A100, all x86_64
├── luxembourg.yaml ← 3 clusters, 56 nodes, 36x AMD MI210/MI300X, 100% SSD
├── louvain.yaml ← 1 cluster, 8 nodes, 2×100 Gbps SR-IOV, no GPU
└── toulouse.yaml ← 2 clusters, x86_64 + AArch64 (Jetson AGX Xavier)
This toolchain complements TOSCA Designer
(latest: v0.5.1, Sep 2025),
an open-source module for Modelio 5.4.1
that provides graphical UML-integrated modeling of cloud-fog-edge TOSCA topologies,
with a custom eu.myrtus.* node type hierarchy, policy and constraint editors,
and enriched CSAR export. Developed by Softeam R&D as part of the
MYRTUS Horizon Europe project (Grant No. 101135183).
Complementarity:
| TOSCA Designer | This toolchain | |
|---|---|---|
| Interface | Graphical (Modelio UML) | Command-line / Python |
| Input | Visual diagram → TOSCA YAML | TOSCA YAML directly |
| Focus | Design-time modeling, constraint authoring, CSAR export | Static risk analysis, CI/CD integration |
| Output | .tosca / .csar files |
HTML report, JSON audit |
| Use when | Designing a new topology from scratch | Auditing any existing TOSCA file |
Open Source Experience — Paris, December 2025
TOSCA-Driven Governance of Heterogeneous Computing Continuums:
Detecting Architectural Risks Before They Become Runtime Failures
https://www.opensource-experience.com/fr/programme-2026
Slides and live demo available in /talk.
Deploying applications on a heterogeneous cloud-fog-edge infrastructure requires topological models that account for layer heterogeneity, resource diversity, and inter-layer quality constraints. TOSCA (Topology and Orchestration Specification for Cloud Applications) provides a vendor-independent formalism for describing such topologies, but its potential for automated architectural analysis remains largely unexplored. In this presentation, we introduce a TOSCA-based governance toolchain composed of two complementary components: (1) a formal TOSCA model described in the Modelio modeling tool for heterogeneous compute nodes, and (2) an automated architectural auditor implementing 21 detection rules across 8 categories.
Topology files for your own infrastructure are welcome as pull requests. New audit rules too.