EVE: a Secure API for the Edge That Delights App Developers

### Overview Presented at the Linux Foundation Open Source Summit by Kathy Jory of Zededa, this talk introduces Project EVE (Edge Virtualization Engine), an open-source initiative under LF Edge. The project aims to bring the convenience, multi-tenancy, and security advantages of cloud-native computing down to diverse, physical edge environments. By establishing an open, secure-by-design operating system layer (EVE-OS) paired with centralized orchestration controllers, Project EVE allows developers to deploy applications without worrying about underlying hardware maintenance, while ensuring IT operations retains control and prevents costly physical maintenance visits ("truck rolls").

### Key Topics - **Cloud Native vs. Edge Computing Realities**: In public clouds, underlying operating systems and hardware configurations are abstracted away from developers. At the edge, deployments span diverse hardware (predominantly x86 alongside growing ARM and emerging RISC-V architectures), varied base OSs, package conflicts, and harsh remote operational realities. - **EVE Architecture & Security Model**: EVE-OS functions as an open edge virtualization engine that does not operate in isolation. It relies on outbound communication to a controller (such as an open-source EVE controller or Zededa's commercial ZedCloud) using TLS and certificates. Because the edge node pushes data outward (e.g., every five minutes) to phone home, there is no need to expose incoming ports or punch holes in firewalls. - **Multi-Tenant Workload Support**: EVE-OS supports running standard containers (leveraging Docker registries and container-d) and full virtual machines (e.g., Ubuntu, Windows 10, or lightweight Kubernetes clusters like k3s) side by side with separate network policies and dependency isolation. - **Dual Partition Reliability**: EVE-OS requires dual partitions (~500 MB partition size) for safe operating system upgrades with automated fallback mechanisms, ensuring devices are never bricked remotely. - **Under the Hood**: EVE-OS is built upon Alpine Linux (leveraging its small footprint, musl libc, and BusyBox), LinuxKit, and containerd.

### Technical Details & Demonstration During a live demonstration, Jory walks through the complete lifecycle of provisioning, onboarding, workload deployment, and decommissioning: - **Bare-Metal Provisioning**: Using Equinix Metal, an on-demand x86 bare-metal server (Supermicro / packet t1.small) is spun up via custom iPXE scripts (fewer than 20 lines) that download kernel, rootfs, and initialization packages directly from GitHub releases. - **Onboarding Process**: In the controller console, the node is registered using its public IP address as the serial identifier, an onboarding key from the EVE image, and predefined hardware profiles. Once booted, EVE-OS automatically establishes a secure outbound TLS connection to register as online. - **Application Deployment**: An open-source LF Edge project container called Fledge (pre-configured for an OSDU oil and gas drilling demo) is retrieved from a Docker registry and deployed to the node. Port forwarding rules (mapping internal port 80 to 8082) allow accessing the container's web dashboard and viewing ingested sensor data translated into OPC-UA format. - **Teardown & Isolation**: The speaker demonstrates that compromised or unneeded applications can be instantly terminated and deleted directly from the controller without losing control of the physical node, before ultimately tearing down the bare-metal server instance.

### Takeaways Project EVE decouples edge infrastructure management from application workloads. Devops and software developers can treat edge hardware as an extension of their cloud environments, deploying containers or VMs from registries without managing host OS dependencies. Concurrently, IT administrators maintain root ownership and security over the physical edge hardware through secure outbound API orchestration, ensuring robust isolation and avoiding operational disruptions even if individual application workloads become compromised.