Upgrade from v0.8.1 to v1.0.0 [DRAFT]

Knowing before the upgrade

1. We only support upgrading from v0.8.1 to v1.0.0. For other versions, please upgrade to v0.8.1 first.
2. We only support offline upgrade from v0.8.1 to v1.0.0 due to an Instance Manager change.

Prepare for the upgrade.

1. Check to make sure the parameters in the default StorageClass weren't changed if you're installed using Helm Chart/Rancher Catalog App. Change default StorageClass's parameter might result in Chart upgrade failure. You can copy the default StorageClass's configuration to create another StorageClass if you want to reconfigure the parameters in the StorageClass.

   The current default StorageClass is with the following parameters:

parameters:
  numberOfReplicas: <user specified replica count, by default it's 3>
  staleReplicaTimeout: "30"
  fromBackup: ""
  baseImage: ""

2. Shutdown your workload following the instruction here
3. If you still have any volumes using pre-v0.7.0 CSI driver name io.rancher.longhorn, follow the instruction here to convert your old PVs.

Upgrade

1. Perform the manager upgrade according to the manager upgrade instruction
2. Perform the engine upgrade according to the offline engine upgrade instruction, but don't scale back the workload just yet.
3. We recommend updating the Guaranteed Engine CPU to 0.25. This step will restart all the Instance Managers on the node, so any attached volume will be detached.
   1. Please make sure you have at least 2 VCPUs per node before updating this setting to 0.25. See the setting reference for details.
4. Scale back the workload. Check if everything works well.
5. We also recommend updating the Replica Node Soft Anti-affinity setting to false.
   1. Please make sure you have more nodes than the default replica count before updating this setting.