fix: purge any cephfs storage classes installed by ops.manifest #38
Conversation
addyess
force-pushed
the
bug/lp2098004/purge-cephfs-sc-without-knowning-names
branch
from
de4acaf to
dbf9ac3
Compare
| if cast(SafeManifest, self.manifests).purgeable: | ||
| # If we are purging, we may not be able to create any storage classes | ||
| # Just return a fake storage class to satisfy delete_manifests method | ||
| # which will look up all storage classes installed by this app/manifest | ||
| return [StorageClass.from_dict(dict(metadata={}, provisioner=self.PROVISIONER))] |
There was a problem hiding this comment.
There's something I don't understand in this dunder method. Why do we need to create an empty sc when we are purging? Could there be any leftover sc, or is this just an edge case?
There was a problem hiding this comment.
Manifest objects can support Manipulations. These are enacted by calling the Manipulation.
The types of manipulations are defined in the project
- Subtraction
- Patch
- Addition
In this case. The CephStorageClass manipulation is an Addition. It is a set of objects added to the existing static manifests in the upstream yaml files. ops.manifest will call this manipulation to create the lightkube resource definition for the storage class.
Normally, for INSTALLING resources, this is straight forward to create these additions. ops.manifest knows the resources it should add. But in the purging situation, it's a bit different. We want to make sure that ops.manifest can delete all the objects this manifest may have added.
What's worse, is we don't REALLY know the name of the storage classes when we're purging. The state information necessary to generate the list of storage class names is gone (it was the cephfs filesystem listings).
But ops.manifest would have labelled each kubernetes resource with some metadata labels to indicate which manifest placed it into the cluster. In this case, if any storage class is created, it will have two labels associated with it:
- juju.io/application: ceph-csi-app-name # from the running application
- juju.io/manifest: cephfs # from this moduleWhat this __call__ is returning is a single list of StorageClasses with no names.
when ops.manifest builds its list of resources, it will query the cluster looking for all the StorageClasses with these matching labels and then delete those resources if they were found
addyess
deleted the
bug/lp2098004/purge-cephfs-sc-without-knowning-names
branch
fix: Purge cephfs storage class purging by purging only those SCs labelled by this app and manifest
Addresses LP#2098004
Depends on