On Mon, Dec 25, 2017 at 11:52:36AM +0800, QR wrote: > Is anyone know the reason that ERR_BUCKET_EXISTS is modified to zero? > Thanks. This comes down to arguing about AWS S3 CreateBucket behavior if the bucket already existed and was owned by you (plus which region it is in vs where the request was made to). AWS S3 us-east-1: returns 200 OK AWS S3 other regions: return 409 Conflict: BucketAlreadyOwnedByYou RGW implements the 200-OK behavior, which is why the return is modified to zero (ret==0 -> HTTP 200 OK). I created issue 17398 a year ago to discuss it: http://tracker.ceph.com/issues/17398 I see recently that issue 22279 was created for the same thing, with an attached PR, but the same discussion. http://tracker.ceph.com/issues/22279 https://github.com/ceph/ceph/pull/19249 Yehuda's opinion is that the 200-OK behavior is better, because CreateBucket should be an idempotent operation. My opinion is that the 409-BucketAlreadyOwnedByYou behavior is better, because the CreateBucket annoyingly is actually two operations: (CreateBucket, SetBucketACL). If the bucket does already exist, but the ACLs you pass in bucket creation are NOT the same as the existing bucket ACLs, what should happen? - Skip creation, skip ACLs? 200-OK response is dangerously misleading here. - Skip creation, replace ACLs? This is going to overwrite the existing ACLs, rather than modify them. - Do nothing and issue a warning about ACL mismatch? Annoying in that we're diverging more from AWS behavior. AWS S3 went to 409-BucketAlreadyOwnedByYou for a reason. They've never stated exactly why, but I believe their choice makes sense. -- Robin Hugh Johnson Gentoo Linux: Dev, Infra Lead, Foundation Asst. Treasurer E-Mail : robbat2@xxxxxxxxxx GnuPG FP : 11ACBA4F 4778E3F6 E4EDF38E B27B944E 34884E85 GnuPG FP : 7D0B3CEB E9B85B1F 825BCECF EE05E6F6 A48F6136
Attachment:
signature.asc
Description: Digital signature
