I'm unable to use upgradeMode: savepoint with the Kubernetes Operator because savepoints don't seem to work. When I test it out, the debug logs just report trying and failing to create a savepoint:

2023-10-04 17:36:39,870 DEBUG org.apache.hadoop.fs.azurebfs.AzureBlobFileSystemStore       [] - createDirectory filesystem: savepoints path: <abfss://savepoints>@[].<http://dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87|dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87> permission: rwxrwxrwx umask: ----w--w- isNamespaceEnabled: true
2023-10-04 17:36:39,938 DEBUG org.apache.hadoop.fs.azurebfs.services.AbfsClient            [] - HttpRequest: CreatePath: 201,,,cid=:416aa93c-5295-42b2-9092-f84eefec5a03:0d5829d1-aae7-4cc8-8342-e84c6712d3e8:::MK:0,rid=6af89f01-c01f-007e-35e9-f60ac6000000,sent=0,recv=0,PUT,https://[].<http://dfs.core.windows.net/savepoints/kafka/savepoint-b56166-a33a8d912f87?resource=directory&timeout=90|dfs.core.windows.net/savepoints/kafka/savepoint-b56166-a33a8d912f87?resource=directory&timeout=90>
2023-10-04 17:36:39,945 DEBUG org.apache.hadoop.fs.azurebfs.AzureBlobFileSystem            [] - AzureBlobFileSystem.delete path: <abfss://savepoints>@[].<http://dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87|dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87> recursive: true
2023-10-04 17:36:39,945 DEBUG org.apache.hadoop.fs.azurebfs.AzureBlobFileSystemStore       [] - delete filesystem: savepoints path: <abfss://savepoints>@[].<http://dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87|dfs.core.windows.net/kafka/savepoint-b56166-a33a8d912f87> recursive: true
2023-10-04 17:36:39,945 DEBUG org.apache.flink.runtime.jobmaster.JobMaster                 [] - Stop-with-savepoint transitioned from WaitingForSavepoint to FinalState on savepoint creation failure handling for job b56166df251740c739ae36fb13dc4d26.
2023-10-04 17:36:39,965 WARN  org.apache.flink.runtime.jobmaster.JobMaster                 [] - Stop-with-savepoint transitioned from FinalState to FinalState on execution termination handling for job b56166df251740c739ae36fb13dc4d26 with some executions being in an not-finished state: [CANCELED]
2023-10-04 17:36:39,992 DEBUG org.apache.hadoop.fs.azurebfs.services.AbfsClient            [] - HttpRequest: DeletePath: 200,,,cid=:30fef2e1-eea1-4c90-b444-3c62be848eb4:0d5829d1-aae7-4cc8-8342-e84c6712d3e8:::DL:0,rid=6af89f08-c01f-007e-3ce9-f60ac6000000,sent=0,recv=0,DELETE,https://[].<http://dfs.core.windows.net/savepoints/kafka/savepoint-b56166-a33a8d912f87?timeout=90&recursive=true|dfs.core.windows.net/savepoints/kafka/savepoint-b56166-a33a8d912f87?timeout=90&recursive=true>

Restoring from checkpoint using last-state works fine and it uses the same credentials and storage account.