Skip to content

Latest commit

 

History

History
112 lines (78 loc) · 3.4 KB

02-machine-console-access.md

File metadata and controls

112 lines (78 loc) · 3.4 KB
title iep-number creation-date status authors reviewers
Machine Console Access
2
2022-12-05
implementable|implemented
@adracus
@gehoern
@afritzler
@Gchbg

IEP-02: Machine Console Access

Table of Contents

Summary

A user of the ironcoreshould be able to access the serial console of their machine to access / debug / run imperative commands on it. For this, an endpoint + client-side tooling has to be created as well as the server-side machinery.

Motivation

Users should be able to access / debug / run imperative commands on their machines. This gives immediate feedback on the state of the machine and is a required feature for our minimum viable product.

Goals

  • Define an endpoint + client side tooling for machine console access
  • Define a server-side interface machine pool providers have to implement in order to support console access to their machines.

Non-Goals

  • Due to the imperative nature of consoles, no declarative interface to consoles should be defined.
  • Have consoles as a building piece of other parts of the ironcore.

Proposal

User-facing API

The compute.ironcore.dev/Machine resource is extended with an exec subresource. When connecting to that subresource, a websocket connection to the backing machine console should be opened. Supported HTTP methods for the exec call are POST and GET (in order to be able to do this from a browser as well).

Example call to the Kubernetes API server hosting the aggregated API:

GET https://<address>/apis/compute.ironcore.dev/v1alpha1/namespaces/<namespace>/machines/<machine-name>/exec

Server-Side API

Once the server receives such a request, it gets the Machine and looks up the MachinePool the Machine is running on. If the Machine does not exist or is not scheduled onto a MachinePool, an error is returned.

After identifying the responsible MachinePool, it is retrieved and its .status.addresses field is inspected for an address to call. The .status.addresses field does not exist yet and has to be updated in the ironcore. It is the responsibility of the MachinePool implementor to report its endpoints in the status.

For reference on the address type, see the Kubernetes node address type which will be used as reference for designing the address type.

Example manifest:

apiVersion: compute.ironcore.dev/v1alpha1
kind: MachinePool
metadata:
  name: my-machine-pool
spec:
  providerID: my://machine-pool
status:
  addresses:
    - address: 10.250.0.38
      type: InternalIP
    - address: my-machine-pool-host
      type: ExternalDNS

Once an address has been identified, the ironcore API server calls the endpoint of the MachinePool provider with an exec request for the Machine. The resulting websocket connection is proxied through the ironcore API server to the user.

GET https://<machine-pool-adddress>/apis/compute.ironcore.dev/namespaces/<namespace>/machines/<machine>/exec

Caution: This proposal does not include anything on authentication mechanisms yet. Implementors can already implement the endpoint but authentication will be added in the future.

Alternatives