https://issues.apache.org/jira/browse/HDFS-9405 adds a background thread to pre-warm EDEK cache.
However this fails and retries continuously if key retrieval fails for one encryption zone. In our usecase, we have temporarily removed keys for certain encryption zones. Currently namenode and kms log is filled up with errors related to background thread retrying warmup for ever .
The pre-warm thread should
- Continue to refresh other encryption zones even if it fails for one
- Should retry only if it fails for all encryption zones, which will be the case when kms is down.