Skip to content

Latest commit

 

History

History
103 lines (72 loc) · 4.21 KB

machine-remediation.md

File metadata and controls

103 lines (72 loc) · 4.21 KB
Raw

Machine Remediation

Table of Contents

Summary

This document explains the implementation of the Machine Remediation Request controller and how it will work together with Machine Health Check controller to restore unhealthy machines.

Motivation

To provide a mechanism to apply remediation strategy on an unhealthy node, that will be different by type and by a provider.

Goals

  1. To provide the interface that will make possible to apply different remediation strategy for each cloud or bare metal providers.
  2. To provide details regarding the progress of the remediation operation.

Non-Goals

  1. To provide a remediation strategy for each cloud or bare metal providers.
  2. To provide the integration with the MachineHealthCheck controller.

Proposal

Implementation Details/Notes/Constraints

It's common that high available environment should have a mechanism to restore unhealthy nodes and when delete and create a new node looks feasible solution for a cloud provider, in case of bare metal hosts with additional infrastructure configuration it looks like too destructive approach. To make possible to specify the remediation strategy for different failure cases and for nodes that run on different providers, this proposal introduces MachineRemediation controller that should monitor all MachineRemediation CR's and apply the remediation strategy according to details specified under the CR. When the MachineRemediation will apply the remediation strategy, one who is initiating it will be the MachineHealthCheck controller. It will monitor nodes, and for each unhealthy node will create MachineRemediation CR, if after the specified timeout the MachineRemediation will not have the state equal to Succeeded, the MachineHealthCheck controller will notify about the remediation failure, and can try to use different type of the remediation strategy.

MachineRemediation spec API

This proposal introduces a new API type: MachineRemediation, that makes possible to create simple MachineRemediation object with the type of the remediation strategy for the specific machine.

Example
apiVersion: machineremediation.kubevirt.io/v1alpha1
kind: MachineRemediation
metadata:
  name: mr-test
  namespace: openshift-machine-api
spec:
  type: reboot
  machineName: test-machine

MachineRemediation status API

MachineRemediation status will show what the state the remediation operation has and the time when the remediation operation started.

status:
  state: Started
  reason: The fencing operation started
  startTime: Thu, 20 Jun 2019 03:38:39 -0400
  endTime: Thu, 20 Jun 2019 03:40:39 -0400

Risks and Mitigations

It can introduce some integration complexity between MachineHealthCheck and MachineRemediation controllers, that will make harder to debug and to test things.

Design Details

Work Items

  • Create new MachineRemediation CRD.
  • Write a new controller that will watch MachineRemediation CR's.
  • Create an operator to deploy the MachineRemediation controller.
  • Create integration with the OLM.

Dependencies

N/A

Test Plan

As basic tests, it will be enough to create MachineRemediation CR's with the reboot type and verify that the machine really rebooted. In the future, we will need to test integration between MachineHealthCheck and MachineRemediation controllers to verify that our final fencing solution works as expected.

Upgrade / Downgrade Strategy

Should be managed by OLM.