Ceph » rgw

Hi Nathan,

The process on L is indeed (or should I say "remove_indeed") has a very different flow. It looks to me that this behavior was most likely a regression relative to L, though.

On L, the logic is unrolled in RGWLC::bucket_lc_process, into two large cases: !versioned bucket, and else (the case has a misleading comment "/* bucket versioning is enabled or suspended */"). So if there can be a delete marker, it's only in the else case, and, in the else case, remove_expired_obj can be reached only if we dont take the branch (my l.555):

if (obj_iter->is_delete_marker()) {
              if ((obj_iter + 1)==objs.end()) {
                if (is_truncated) {
                  //deal with it in next round because we can't judge whether this marker is the only version
                  list_op.get_next_marker() = obj_iter->key;
                  break;
                }
              } else if (obj_iter->key.name.compare((obj_iter + 1)->key.name) == 0) {   //*obj_iter is delete marker and isn't the only version, do nothing.
                continue;
              }
              skip_expiration = prefix_iter->second.dm_expiration;
              remove_indeed = true;   //we should remove the delete marker if it's the only version
            } else {
              remove_indeed = false;
            }

Which looks like the same logic from the upstream PR l. 729?

}
    if (o.is_delete_marker()) {
      if (oc.ol.next_has_same_name()) {
        return false;
      } else {
        *exp_time = real_clock::now();
        return true;
      }
    }

Matt