Reversibility Policy for the AI Managed Container product

Objective

This document describes the reversibility policy for the AI Managed Container product covering the OVHcloud AI Deploy and AI Training offers.

This policy aims to implement the general reversibility principles and our compliance with the SWIPO IaaS Code of Conduct for cloud providers.

List of features

Features of the product line fall into three categories:

Core features for which we guarantee migration capacity.
OVHcloud implementations that require adaptation to a new migration environment.
Specific features that cannot be guaranteed for migration as they are related to the OVHcloud environment or involve custom developments.

1. Core features

Function	Description	Available formats	Migration model	Available documentation
OCI container deployment and management	AI models deployment or workloads via standard Docker/OCI images, orchestrated on a native Kubernetes cluster or compatible API	OCI, Docker, YAML, JSON	Inbound: direct import of OCI/Docker images via API, CLI (Docker, Kubectl, Helm, etc.), or via CI/CD pipeline. Outbound: Export of OCI/Docker images, YAML/JSON manifests and configurations which can be used on any Kubernetes cluster or OCI platform.	Build & use a custom Docker image
Standard API usage	Manage workloads, services, configurations and monitoring via the Kubernetes API (kubectl, Helm, etc.)	YAML, JSON, OCI	Inbound: Direct deployment of manifests, charts, AI models via standard API. Outbound: Export of manifests, charts and configurations via kubectl that can be used on any compatible Kubernetes cluster.	Build & use a custom Docker image
OCI/Docker Image Registry	Using public/private registries for container images and AI models	OCI, Docker	Inbound: Pull/extract images from any compatible registry. Outbound: Push/export images to any other compatible registry.	Portfolio of AI apps and models
Using API, CLI and Infrastructure as Code	Deployment, management and automation via the OVHcloud API, CLI or Infrastructure as Code (Terraform)	YAML, JSON, HCL	Inbound: Configurations, scripts and templates. Outbound: Export scripts, templates and configurations that are reusable on other target environments	Features, Capabilities and Limits

2. OVHcloud Implementations

Function	Description	Available formats	Migration model	Available documentation
Access and secret management	Kubernetes RBAC model and secret management or use of OVHcloud solutions	YAML, JSON	Inbound: Adapt roles, secrets and permissions to the environment Outbound: Export configurations, adaptation required according to the target environment	AI Training - Getting started
Managed auto-scaling and load balancing	Horizontal and vertical auto-scaling, integrated OVHcloud load balancing	YAML, JSON	Inbound: Adaptation of auto-scaling and load balancing parameters in the environment Outbound: Export configurations, adaptation to auto-scaling and load balancing in the target environment	AI Deploy - Scaling strategies
Integrated monitoring and logging	Integrated supervision (Prometheus, Grafana, Loki, etc.) via OVHcloud managed services	Prometheus metrics, logs, JSON	Inbound: Adapt dashboards and agents Outbound: Export metrics, logs and reconfigure the new environment	AI Deploy - Features, Capabilities and Limits AI Training - Features, capabilities and limitations
Managed Persistent Storage	OVHcloud Block storage, Object Storage integration, specific storage classes	PVC, YAML, JSON	Inbound: Adapt PersistentVolumeClaims to the environment. Outbound: Export data via snapshot. Export, adapt volumes to the target (e.g. migration via CSI driver or rsync)	AI Deploy - Features, Capabilities and Limits

3. Specific features

Function	Description	Available formats	Migration model	Available documentation
Infrastructure as a Code	Automated deployment via Terraform modules specific to OVHcloud	NA	Inbound: Adapt scripts Outbound: Rewrite Terraform configurations required	Terraform
Anti-DDoS protection	Anti-DDoS is a set of tools and mechanisms designed to absorb denial of service attacks. It includes traffic analysis, "clean-up" via a specialized network, and mitigation using VAC technology developed by OVHcloud.	N/A	Inbound: The anti-DDoS system is part of our infrastructure and is enabled by default. No action is required. Outbound: Order and configure an anti-DDoS solution from the new provider.	Anti-DDoS OVHcloud

List of architectures

The product is based on managed Kubernetes clusters, with full orchestration of AI workloads (deployment, scaling, monitoring). Architectures support horizontal and vertical auto-scaling, integrated load balancing, high availability, GPU/CPU integration for training and inference and access to persistent storage services (Block storage, Object Storage). CI/CD integration, multi-model management, and access governance are native.

Partner services

The OVHcloud partners concerned are listed in the OVHcloud partners directory under the "Data center expansion and Migration" keywords.

OVHcloud also has a dedicated service: OVHcloud Professional Services.

Cost and fees

Billing is on a pay-per-use basis and stops when you no longer consume resources. No cancellation fees apply: deleting the service will stop the billing immediately. Any associated OVHcloud credits cannot be transferred. Once the service is cancelled, OVHcloud releases its resources, making data retrieval impossible. Clients are responsible for exporting their configurations, manifests, and images before termination, as all resources are released upon deletion.

Data retention after contract termination

After service deletion or contract termination, OVHcloud permanently deletes all images, models, logs, metrics and configurations stored in its environment. Automatic backups and logs are also deleted. It is therefore imperative to export all the necessary data before deletion, as you cannot restore it after these actions.

Esta página foi útil?