VRG reconciler workflow

Mon, 01 Jan 0001 00:00:00 +0000

VRG reconciler workflow

This design document covers the VRG CRD and its reconciliation workflow.

Terminology

VRG: VolumeReplicationGroup
VR: VolumeReplication
PVC: PersistentVolumeClaim
PV: PersistentVolume

VRG operator functionality

VRG resource exists to manage a group PVCs that need replication. It is a namespace level resource that ensures the following:

For each existing or dynamically created PVC:
- Ensures PVCs are protected from deletion, prior to replicating them
- Ensures PVCs are bound before enabling replication for them
- Ensures only PVCs that are NOT being deleted are protected, when reconciling them
- Ensures PVs backing the PVCs are retained for the duration that they are protected
- Backs up the PV cluster data to an external S3 store, to enable restoring them on a remote peer cluster
Creates and manages VR resource for PVCs that need to be protected
Manages state changes in VRG by ensuring VRs reach the same desired state, coordinating with PVC state and ensuring:
- PVCs are not in use when being marked Secondary
- PVCs are marked for deletion when being marked Secondary
Manages VRG deletion, by ensuring VRs are in the desired state prior to deletion

VRG CRD

Spec

apiVersion: ramendr.openshift.io/v1alpha1
kind: VolumeReplicationGroup
metadata:
 name: vrg-sample
 generation: <generation>
spec:
 pvcSelector:
 matchLabels:
 <key>: <value>
 volumeReplicationClass: <string>
 replicationState: "Primary"|"Secondary"
 s3Endpoint: <url-string>
 s3SecretName: <string>

Fields:
- PVCSelector
  - Label selector to identify all the PVCs that are belong to this VRG and need to be replicated
- VolumeReplicationClass
  - ReplicationClass for all volumes in this replication group; this value is propagated to created VolumeReplication CRs
- ReplicationState
  - Desired state of all volumes [primary or secondary] in this replication group; this value is propagated to created VolumeReplication CRs
- S3Profiles
  - S3 Profiles to replicate PV cluster data

Status

status:
 protectedPVCs: [
 pvcName: <string>
 conditions: [
 type: ["Available"]
 status: [true|false|unknown]
 observedGeneration: <generation>
 lastTransitionTime: <time>
 reason: "Error"|"Progressing"|"Replicating"
 message: <human readable string>
 ]
 ]
 conditions: [
 type: ["Available"]
 status: [true|false|unknown]
 observedGeneration: <generation>
 lastTransitionTime: <time>
 reason: ["Replicating"|"Progressing"|"Error"]
 message: <human readable string>
 ]

status.conditions:
- type “Available” is to denote availability status of the VRG resource
  - NOTE: status may change back to “Progressing” in case there are new PVCs to deal with for the same CRD generation
- status
  - “Unknown” if VRG is not yet picked up for reconciliation
  - “True” if reconciliation is completed
    - reason is “Replicating”
  - “False” if reconciliation is progressing or has errors
    - reason: Progressing|Error
status.protectedPVCs:
- pvcName: Name of the PVC for which conditions are represented
- conditions:
  - type “Available” is to denote availability status for PVC protection
  - status
    - “True” if reason is “Replicating”
    - “False” if reason is one of “Error” or “Progressing”, to denote any hard errors or that reconciliation is in progress
  - NOTE(s):
    - Ideally if status.conditions.[Available].status is True, the all PVC status should be true, with the reason as “Replicating”
    - An observedGeneration number for when the PVC was protected may not be required, as once it is protected it stays protected. It may prove useful though if any straggling PVCs were protected during a cluster heal or related operations, to detect prior failover PVCs with the current generation

Notes

Status may change for the same observed generation, as PVCs are created
Should there be a per PVC condition that it’s backing PV was backed up? To denote that replication at storage maybe missing, but PV cluster data may exist?

VRG Example

---
apiVersion: ramendr.openshift.io/v1alpha1
kind: VolumeReplicationGroup
metadata:
 name: vrg-sample
spec:
 pvcSelector:
 matchLabels:
 appname: myapp
 volumeReplicationClass: "vr-class-sample"
 replicationState: "Primary"
 s3Profiles:
 s3ProfileEast
 s3ProfileWest

VRG reconciliation

PVC preparation and protection for all states

All reconciliation state changes need to ensure a PVC is in the right state and protected prior to proceeding with the state change. The following steps are taken to ensure the same:

Proposals on Ramen

VRG reconciler workflow

VRG reconciler workflow

Terminology

VRG operator functionality

VRG CRD

Spec

Status

Notes

VRG Example

VRG reconciliation

PVC preparation and protection for all states