I have this persistent volume claim resource:

import {
  PersistentVolumeClaim,
  type PersistentVolumeClaimArgs,
} from '@pulumi/kubernetes/core/v1/persistentVolumeClaim';
import { output } from '@pulumi/pulumi/output';
import type { CustomResourceOptions } from '@pulumi/pulumi/resource';

export class SaasPersistentVolumeClaim extends PersistentVolumeClaim {

  constructor(name: string, args: SaasPersistentVolumeClaimArgs, opts?: CustomResourceOptions) {

    super(
      name,
      {
        metadata: output(args).apply((args) => ({
          namespace: args.metadata?.namespace,
          labels: args.metadata?.labels,
          annotations: {
            '<http://pulumi.com/patchForce|pulumi.com/patchForce>': 'true',
            '<http://pulumi.com/skipAwait|pulumi.com/skipAwait>': 'true'
          },
          name: name,
        })),
        spec: {
          accessModes: ['ReadWriteOnce'],
          resources: { requests: { storage: `${args.newSize}Gi` } },
          storageClassName: 'performance-regular',
        },
      },
      {
        ...opts,
        ignoreChanges: ['metadata', 'spec.accessModes', 'spec.storageClassName'],
      },
    );
  }
}

Here, I deploy the pvc with 10Gi, and I want to upgrade it to have 20Gi. When I: • include the

metadata

field in the

ignoreChanges

array, the upgrade fails with

metadata.managedFields must be nil

• remove the

metadata

field from the

ignoreChanges

array, the upgrade is successful I think the

'<http://pulumi.com/patchForce|pulumi.com/patchForce>': 'true'

is getting ignored during the update because it's in the

ignoreChanges

array. I want to keep the

metadata

field in the

ignoreChanges

array. Theoretically, the

'<http://pulumi.com/patchForce|pulumi.com/patchForce>': 'true'

should be used during the operation, just changes to it should be ignored. This looks like a bug with pulumi, or is there something I'm missing? Versions:

"@pulumi/kubernetes": "4.5.5",
    "@pulumi/pulumi": "3.94.2",

Any help is appreciated!

Why do you want to ignore those fields?

Because changing the metadata.name recreates a PVC. To not lose data, we ignore the whole metadata field because we don't know updates to which other metadata fields cause recreations.

Changing the name does indeed recreate a PVC, but that's the intended behaviour. I'd suggest using

preview

to verify change sets, as ignoring the entire metadata could have unintended side effects as you've noticed.

That's not a feasible option since the up flow is automated using the automation API, so we do not preview the changes. But anyway, ignoring just the

metadata.name

has been enough to ensure that the PVC is not deleted on changes. Thank you for your input!