Ceph » rgw

It is possible to create an inconsistent state by following this procedure:

1. create a bucket through an API call to the gateway, handled by boto for example, call this bucket A
2. add some stuff to the bucket through API calls
3. do some other stuff
4. remove the bucket object through a call to rados directly, $rados on the server running the radosgw,
5. create a bucket with the same name as the previously deleted bucket, call this bucket B

At this point attempts to access the newly created bucket, bucket B, will return 404 not found errors through a client
such as boto, however, attempts to place objects in bucket B will return 200 success through a client such
as boto. Further attempts to retrieve information such as the keys contained in bucket B will return 400 bad
request errors. Furthermore attempts to delete bucket B through radosgw-admin will return a -22 error. Additionally,
when using $rados to list the objects in .rgw.buckets any objects added to bucket B have the prefix of the old
bucket, bucket A, that had been deleted through a call to rados and a new directory with the prefix of bucket A will
have also been created in .rgw.buckets. This behavior will cause a 409 conflict error if a different user, from the user
originally linked to bucket A when it was created, even if the original user no longer exists.