mirror of
https://github.com/edera-dev/krata.git
synced 2025-08-02 21:00:55 +00:00
added a technical overview and added a link in the readme (#474)
This commit is contained in:
parent
53052b24a0
commit
699c2e09ef
@ -11,6 +11,7 @@ krata is an implementation of a Xen control-plane in Rust.
|
||||
- [Frequently Asked Questions](FAQ.md)
|
||||
- [Code of Conduct](CODE_OF_CONDUCT.md)
|
||||
- [Security Policy](SECURITY.md)
|
||||
- [Edera Technical Overview](technical-overview.md)
|
||||
|
||||
## Introduction
|
||||
|
||||
@ -23,3 +24,4 @@ It provides the base layer upon which Edera Protect zones are built on: a secure
|
||||
|--------------|------------------|-------------------------|
|
||||
| x86_64 | 100% Completed | None, Intel VT-x, AMD-V |
|
||||
| aarch64 | 10% Completed | AArch64 virtualization |
|
||||
|
||||
|
86
technical-overview.md
Normal file
86
technical-overview.md
Normal file
@ -0,0 +1,86 @@
|
||||
# Edera Technical Overview
|
||||
|
||||
## What is Edera?
|
||||
|
||||
Edera is a secure-by-default, cloud-native platform built on a reimagined, memory-safe type-1 hypervisor. It unlocks hard multitenancy and strong container isolation—without the performance hit.
|
||||
|
||||
Unlike traditional container runtimes that share a single Linux kernel across containers, Edera runs each container in a lightweight virtual machine (called a **zone**), with its own dedicated Linux kernel. This eliminates the kernel as a shared attack surface.
|
||||
|
||||
And because Edera doesn’t rely on nested virtualization, it runs wherever containers do—across public clouds, on-prem, and edge environments.
|
||||
|
||||
## How Edera Works
|
||||
|
||||
At its core, Edera uses a custom hypervisor based on Xen, with key components rewritten in Rust for safety, performance, and maintainability. Edera introduces the concept of **zones**—independent, fast-booting virtual machines that serve as security boundaries for container workloads.
|
||||
|
||||
Each zone runs its own Linux kernel and minimal init system. The kernel and other system components are delivered via OCI images, keeping things composable, cacheable, and consistent.
|
||||
|
||||
Zones are paravirtualized using the Xen PV protocol. This keeps them lightweight and fast—no hardware virtualization required. But when hardware support is available (e.g., on x86 with VT-x), Edera uses it to get near bare-metal performance.
|
||||
|
||||
## How Edera Runs & Secures Containers
|
||||
|
||||
Edera allows you to compose your infrastructure the same way you compose workloads: using OCI images.
|
||||
|
||||
Each zone consumes a small number of OCI images:
|
||||
- A **kernel image** that provides the zone kernel.
|
||||
- One or more **system extension images** that provide init systems, utilities, and kernel modules.
|
||||
- Optionally, **driver zones**—zones that provide shared services (like networking) to other zones.
|
||||
|
||||
Inside each zone, container workloads run via a minimal OCI runtime called **Styrolite**, written in Rust. Unlike traditional setups (like Kata Containers, which layer containerd and runc as external processes), Styrolite is embedded inside the zone itself.
|
||||
|
||||
### Key Benefits of This Design
|
||||
- No external container runtime processes
|
||||
- Zone init system directly manages containers
|
||||
- Minimal attack surface, optimized for secure execution
|
||||
|
||||
This tightly integrated design avoids the complexity, latency, and exposure introduced by conventional container runtimes. It keeps the execution path short, verifiable, and secure-by-design.
|
||||
|
||||
## Zones as Security Boundaries
|
||||
|
||||
In Kubernetes, Edera runs pods inside **zones**—isolated virtual machines that eliminate risks like container escape, privilege escalation, and lateral movement.
|
||||
|
||||
Each zone boots its own kernel, pulled via OCI, and runs a single pod by default. You can also configure zones to run a replica set, a namespace, or a set of trusted workloads together.
|
||||
|
||||
To use Edera, apply the `RuntimeClass`:
|
||||
|
||||
```yaml
|
||||
apiVersion: node.k8s.io/v1
|
||||
kind: RuntimeClass
|
||||
metadata:
|
||||
name: edera
|
||||
handler: edera
|
||||
```
|
||||
|
||||
Then annotate your pod:
|
||||
|
||||
```yaml
|
||||
apiVersion: v1
|
||||
kind: Pod
|
||||
metadata:
|
||||
name: edera-protect-pod
|
||||
spec:
|
||||
runtimeClassName: edera
|
||||
```
|
||||
|
||||
This causes the pod to be scheduled to a node running Edera’s hypervisor. The pod is transparently launched inside its own VM zone—no image changes, no config rewrites, and no extra work from developers.
|
||||
|
||||
## What Exactly Is an Edera Zone?
|
||||
|
||||
An Edera zone is a minimal VM built from OCI-delivered components. At launch time, the Edera daemon unpacks:
|
||||
|
||||
### Kernel Image
|
||||
Located under `/kernel` in the OCI image:
|
||||
- `image`: the Linux kernel (vmlinuz)
|
||||
- `metadata`: key-value pairs for boot parameters
|
||||
- `addons.squashfs`: includes kernel modules in `/modules`
|
||||
- `config.gz`: the kernel configuration file
|
||||
|
||||
### Initramfs Contents
|
||||
Packaged in a CPIO archive, typically mounted from:
|
||||
`usr/lib/edera/protect/zone/initrd`
|
||||
|
||||
The initramfs includes:
|
||||
- `/init`: static Rust binary that initializes the zone
|
||||
- `/bin/styrolite`: embedded container runtime
|
||||
- `/bin/zone`: control plane for managing containers and services via IDM (inter-domain messaging)
|
||||
|
||||
This structure lets Edera launch zones rapidly, with well-defined boundaries and no dependency on the host OS kernel. Everything the workload touches is defined, versioned, and validated.
|
Loading…
Reference in New Issue
Block a user