rgw: add the remove-x-delete feature to cancel swift object expiration #13621

Jing-Scott · 2017-02-24T07:38:27Z

In openstack swift, it also support the feature to cancel the object expiration, which could be found at last point in https://docs.openstack.org/user-guide/cli-swift-set-object-expiration.html. we can remove the object expiration by set 'X-Remove-Delete-At:'.

This patch also could fix the bug that when we set the object expiration and then upload the same object to the container again. The previous object expiration also works, which is not compatible with the openstack swift.

Fixes: http://tracker.ceph.com/issues/19074
Signed-off-by: Jing Wenjun jingwenjun@cmss.chinamobile.com

Jing-Scott · 2017-02-27T00:54:18Z

Hi, @rzarzynski
Could you help me to take a look? Thanks!

rzarzynski · 2017-02-28T16:45:07Z

src/rgw/rgw_op.h

 {
-  if (real_clock::is_zero(delete_at)) {


It looks we started treating0 in delete_at as special value saying "expiration hint revoked". Does the expirer part is OK with that?

Yeah. @rzarzynski : In this change, if we want to set the object expiration time. The var set_delete_at will be true and the delete_at updated relatively.

In int RGWObjectExpirer::garbage_single_object(objexp_hint_entry& hint) it will call delete_obj()
In int RGWRados::Object::Delete::delete_obj(), you will find these code:

if (!real_clock::is_zero(params.expiration_time)) { bufferlist bl; real_time delete_at; if (state->get_attr(RGW_ATTR_DELETE_AT, bl)) { try { bufferlist::iterator iter = bl.begin(); ::decode(delete_at, iter); } catch (buffer::error& err) { ldout(store->ctx(), 0) << "ERROR: couldn't decode RGW_ATTR_DELETE_AT" << dendl; return -EIO; } if (params.expiration_time != delete_at) { return -ERR_PRECONDITION_FAILED; } } }

In if (params.expiration_time != delete_at) { :
if the expiration time of object is updated, the delete_at is the current expiration time in bufferlist and the params,expiration_time is the non-current expiration time in the list of cls_timeindex_entry.

So, that means if the two expriation time is not consistent, the object will not be deleted.
If we intend to cancel the object expiration time, the delete_at will be 0 and the object will never be deleted because of if (!real_clock::is_zero(params.expiration_time)) { in upper code.

Thanks for the very detailed explanation! Conceptually everything looks good.

Jing-Scott · 2017-03-01T06:06:07Z

src/rgw/rgw_op.h

 {
-  if (real_clock::is_zero(delete_at)) {
+  map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_DELETE_AT);
+  if (!set_delete_at && iter != attrs.end()) {


@rzarzynski in this change, the delete_at will be encoded only in two situations:

the object was upload in first time

the object expiration time was updated(including remove the expiration time)

What do you think about switching to boost::optional<ceph::real_time> instead of introducing theset_delete_at flag? Optional seems to be intended for handling cases exactly like this one.

Jing-Scott · 2017-03-06T15:49:06Z

@rzarzynski Hi, thanks for your suggestion! I've updated them with boost::optional<ceph::real_time> delete_at instead of ceph::real_time delete_at.

rzarzynski · 2017-03-08T04:02:10Z

src/rgw/rgw_file.cc

@@ -1145,7 +1145,7 @@ namespace rgw {
      emplace_attr(RGW_ATTR_SLO_UINDICATOR, std::move(slo_userindicator_bl));
    }

-    op_ret = processor->complete(s->obj_size, etag, &mtime, real_time(), attrs, delete_at,
+    op_ret = processor->complete(s->obj_size, etag, &mtime, real_time(), attrs, *delete_at,


It looks we might be dereferencing an empty boost::optional here. If that's true, following check should help:

delete_at ? *delete_at : 0

rzarzynski · 2017-03-08T04:02:41Z

src/rgw/rgw_op.cc

@@ -3343,7 +3343,7 @@ void RGWPostObj::execute()
    emplace_attr(RGW_ATTR_COMPRESSION, std::move(tmp));
  }

-  op_ret = processor.complete(s->obj_size, etag, NULL, real_time(), attrs, delete_at);
+  op_ret = processor.complete(s->obj_size, etag, NULL, real_time(), attrs, *delete_at);


The same here.

rzarzynski · 2017-03-08T04:03:49Z

src/rgw/rgw_op.h

-  if (real_clock::is_zero(delete_at)) {
-    return;
+  map<string, bufferlist>::iterator iter = attrs.find(RGW_ATTR_DELETE_AT);
+  if (delete_at == boost::none) {


This is much more readable now. Thanks!

Hmm, I wonder whether we really need to have the RGW_ATTR_DELETE_AT set for all objects (including those created through the S3 API). Maybe the check solely for boost::none would be enough? What do you think?

@rzarzynski ,I've updated them. I considered the http://tracker.ceph.com/issues/19074, so I set RGW_ATTR_DELETE_AT to all objects before. But now there is another solution instead which is described below.

In openstack swift, it also support the feature to cancel the object expiration, which could be found at last point in https://docs.openstack.org/user-guide/cli-swift-set-object-expiration.html. we can remove the object expiration by set 'X-Remove-Delete-At:'. This patch also could fix the bug that when we set the object expiration and then upload the same object to the container again. The previous object expiration also works, which is not compatible with the openstack swift. Fixes: http://tracker.ceph.com/issues/19074 Signed-off-by: Jing Wenjun <jingwenjun@cmss.chinamobile.com>

Jing-Scott · 2017-03-09T14:39:59Z

src/rgw/rgw_rados.cc

@@ -8328,6 +8328,8 @@ int RGWRados::Object::Delete::delete_obj()
      if (params.expiration_time != delete_at) {
        return -ERR_PRECONDITION_FAILED;
      }
+    } else {


Hi, @rzarzynski ,this change will fix http://tracker.ceph.com/issues/19074 .

OK, RGWRados::delete_obj takes expiration_time which is by default ceph::real_time():

/** Delete an object.*/ int delete_obj(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_owner, const rgw_obj& src_obj, int versioning_status, uint16_t bilog_flags = 0, const ceph::real_time& expiration_time = ceph::real_time());

cbodley · 2017-03-24T15:09:30Z

passed teuthology: http://pulpito.ceph.com/cbodley-2017-03-23_12:12:34-rgw-wip-cbodley-testing---basic-mira

(waiting on @rzarzynski review)

rzarzynski

Looks good. Thanks, @Jing-Scott!

rzarzynski · 2017-04-03T13:15:51Z

src/rgw/rgw_file.cc

-    op_ret = processor->complete(s->obj_size, etag, &mtime, real_time(), attrs, delete_at,
-				 if_match, if_nomatch);
+    op_ret = processor->complete(s->obj_size, etag, &mtime, real_time(), attrs,
+                                 (delete_at ? *delete_at : real_time()), if_match, if_nomatch);


If delete_at is empty, then we supply an instance of real_time. ceph::real_time is an alias over std::chrono::time_point<ceph::time_detail::real_clock>. Its default-constructed instance represents logical zero --real_clock::is_zero(ceph::real_time()) == true. OK.

rzarzynski · 2017-04-03T13:17:42Z

src/rgw/rgw_rados.cc

@@ -8328,6 +8328,8 @@ int RGWRados::Object::Delete::delete_obj()
      if (params.expiration_time != delete_at) {
        return -ERR_PRECONDITION_FAILED;
      }
+    } else {


OK, RGWRados::delete_obj takes expiration_time which is by default ceph::real_time():

/** Delete an object.*/ int delete_obj(RGWObjectCtx& obj_ctx, const RGWBucketInfo& bucket_owner, const rgw_obj& src_obj, int versioning_status, uint16_t bilog_flags = 0, const ceph::real_time& expiration_time = ceph::real_time());

Jing-Scott force-pushed the add-remov-x-delete-feature branch from f6d6c6c to 25ad980 Compare February 24, 2017 13:57

liewegas added feature rgw labels Feb 26, 2017

rzarzynski self-assigned this Feb 27, 2017

rzarzynski reviewed Feb 28, 2017

View reviewed changes

Jing-Scott commented Mar 1, 2017

View reviewed changes

Jing-Scott force-pushed the add-remov-x-delete-feature branch from 25ad980 to 8c6753e Compare March 6, 2017 15:40

Jing-Scott force-pushed the add-remov-x-delete-feature branch from 8c6753e to 2a9da73 Compare March 6, 2017 16:11

rzarzynski reviewed Mar 8, 2017

View reviewed changes

Jing-Scott force-pushed the add-remov-x-delete-feature branch 6 times, most recently from 1347007 to e0ade73 Compare March 8, 2017 12:56

Jing-Scott force-pushed the add-remov-x-delete-feature branch from e0ade73 to 230429e Compare March 9, 2017 13:31

Jing-Scott commented Mar 9, 2017

View reviewed changes

cbodley added the wip-cbodley-testing label Mar 23, 2017

cbodley removed the wip-cbodley-testing label Mar 24, 2017

rzarzynski approved these changes Apr 3, 2017

View reviewed changes

rzarzynski merged commit 51dd349 into ceph:master Apr 3, 2017

Jing-Scott deleted the add-remov-x-delete-feature branch April 5, 2017 07:29

Jing-Scott restored the add-remov-x-delete-feature branch June 8, 2017 09:32

Jing-Scott deleted the add-remov-x-delete-feature branch June 8, 2017 09:56

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

rgw: add the remove-x-delete feature to cancel swift object expiration #13621

rgw: add the remove-x-delete feature to cancel swift object expiration #13621

Jing-Scott commented Feb 24, 2017 •

edited

Jing-Scott commented Feb 27, 2017

rzarzynski Feb 28, 2017

Jing-Scott Mar 1, 2017

rzarzynski Mar 2, 2017

Jing-Scott Mar 1, 2017

rzarzynski Mar 2, 2017

Jing-Scott commented Mar 6, 2017

rzarzynski Mar 8, 2017

rzarzynski Mar 8, 2017

rzarzynski Mar 8, 2017

rzarzynski Mar 8, 2017

Jing-Scott Mar 9, 2017 •

edited

Jing-Scott Mar 9, 2017

rzarzynski Apr 3, 2017

cbodley commented Mar 24, 2017

rzarzynski left a comment

rzarzynski Apr 3, 2017

rzarzynski Apr 3, 2017

rgw: add the remove-x-delete feature to cancel swift object expiration #13621

rgw: add the remove-x-delete feature to cancel swift object expiration #13621

Conversation

Jing-Scott commented Feb 24, 2017 • edited

Jing-Scott commented Feb 27, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Jing-Scott commented Mar 6, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Jing-Scott Mar 9, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cbodley commented Mar 24, 2017

rzarzynski left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Jing-Scott commented Feb 24, 2017 •

edited

Jing-Scott Mar 9, 2017 •

edited