gian/ceph-mirror
1
0
Fork 0
mirror of https://github.com/ceph/ceph synced 2026-08-02 07:03:18 +00:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

rgw: copy_obj_data() record uncompressed size for a compressed source

Browse Source 
When copying an object and a compressor/encryption processor is configured
for the destination, the bucket-index accounted_size was set from `ofs`. If
the source is compressed but the destination is not (re-)compressed (e.g.
compression is off, or rgw_compression_type=random rolled no-compression),
then `ofs` is the compressed size, not the logical size.

In multisite this diverges from the peer zone, which recomputes its own
accounted_size via fetch_remote_obj() (already corrected in #63306).
ListObjectVersions then reports different sizes per zone.

Pass along the source's logical size (astate->accounted_size, derived from the
object's own attrs and correct for plain/compressed/encrypted) into
copy_obj_data() and use it when the destination is not compressed. The
rewrite_obj() and transition_obj() callers pass 0 and keep the prior
ofs-based behaviour. This is the copy_obj analog of the fetch_remote_obj
fix in #63306.

No on-disk/encoding change; corrects newly copied objects only (existing
index entries are unaffected) and is safe across a mixed-version upgrade.

Fixes: https://tracker.ceph.com/issues/75716
Fixes: https://tracker.ceph.com/issues/75715
Signed-off-by: Oguzhan Ozmen <oozmen@bloomberg.net>
This commit is contained in:
Oguzhan Ozmen 2026-07-07 12:27:14 +00:00
parent c11c7144fc
commit 5addb80aba
2 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
src/rgw/driver/rados/rgw_rados.cc
View File

@ -5304,7 +5304,8 @@ int RGWRados::copy_obj(RGWObjectCtx& src_obj_ctx,
attrs.erase(RGW_ATTR_CRYPT_PART_NUMS);
attrs.erase(RGW_ATTR_CRYPT_PREFETCH_ALIGN);
return copy_obj_data(dest_obj_ctx, owner, dest_bucket_info, dest_placement, read_op, obj_size - 1, dest_obj,
mtime, real_time(), attrs, olh_epoch, delete_at, petag, dp_factory, dpp, y);
mtime, real_time(), attrs, olh_epoch, delete_at, petag, dp_factory, dpp, y,
astate->accounted_size);
}
/* This has been in for 2 years, so we can safely assume amanifest is not NULL */
@ -5475,6 +5476,7 @@ int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx,
rgw::sal::DataProcessorFactory *dp_factory,
const DoutPrefixProvider *dpp,
optional_yield y,
uint64_t src_accounted_size,
bool log_op)
{
string tag;
@ -5543,8 +5545,13 @@ int RGWRados::copy_obj_data(RGWObjectCtx& obj_ctx,
ldpp_dout(dpp, 0) << "ERROR: failed to read compression info" << dendl;
return ret;
}
// pass original size if compressed
accounted_size = compressed ? cs_info.orig_size : ofs;
// The bucket index accounted_size must be the object's logical (uncompressed)
// size. When the destination is compressed we take it from the compression info.
// Otherwise `ofs` is the number of RAW bytes read from the source: but if the source
// itself is compressed, the ofs is the *compressed* size, not the logical
// size. So, prefer the source's known logical size.
accounted_size = compressed ? cs_info.orig_size
: (src_accounted_size ? src_accounted_size : ofs);
}
const req_context rctx{dpp, y, nullptr};
@ -5650,6 +5657,7 @@ int RGWRados::transition_obj(RGWObjectCtx& obj_ctx,
nullptr, /* dp_factory */
dpp,
y,
0, /* src_accounted_size: unknown here, keep ofs-based size */
log_op);
if (ret < 0) {
return ret;

1
src/rgw/driver/rados/rgw_rados.h
View File

@ -1273,6 +1273,7 @@ public:
rgw::sal::DataProcessorFactory *dp_factory,
const DoutPrefixProvider *dpp,
optional_yield y,
uint64_t src_accounted_size = 0,
bool log_op = true);
int transition_obj(RGWObjectCtx& obj_ctx,