rgw: fix issue 16435 #10193
rgw: fix issue 16435 #10193
Conversation
Fixes: http://tracker.ceph.com/issues/16435 Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
If the head object contains data, need to set it in the list of objs (in addition to being pointed at by head_obj). Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
adjust the manifest head object: - when reading manifest, set the head object to the object we read the manifest from (and its size). Some manifests are broken and point at a different object - when copying multipart object, set the manifest head object to point at the new head object Signed-off-by: Yehuda Sadeh <yehuda@redhat.com>
|
Passes my eyeball check, but will it sanely read the copied manifest without issue? Not sure until I can test it next week. |
|
lgtm, but needs testing |
|
@robbat2 @oritwas I tested the specific functionality manually: objects created in dumpling, both regular and multipart, of various sizes (< 1k, 512k, 1M, 4M, 40M). Made sure that after copying this object and removal of source the target is readable and contains correct data. Also made sure that manifest looks correct. I stumbled on a second issue on the hammer branch I was running that copy of the multipart dumpling objects did not succeed. I assume this doesn't happen in @robbat2's environment, so it could be a regression in hammer. The last commit handles that case. Left to do: run this in teuthology, see that it doesn't break current stuff. |
Fix multiple issues related to handling of the old explicit objs manifest.
cc: @robbat2