Ceph » rgw

The archive zone is creating multiple copies of the same object, one for each source zone configured.

For eg.,

1. aws --no-verify-ssl --endpoint-url http://localhost:8000 s3 cp ../../scripts/sample_hello.txt s3://bucket1/test2
   upload: ../../scripts/sample_hello.txt to s3://bucket1/test2
2. aws s3api --no-verify-ssl --endpoint http://localhost:8002 list-object-versions --bucket bucket1 {
   "Versions": [ {
   "ETag": "\"ed076287532e86365e841e92bfc50d8c\"",
   "Size": 12,
   "StorageClass": "STANDARD",
   "Key": "test2",
   "VersionId": "LhGwcb-wEgEdTdcXQdlASTmSZMmQ7vz",
   "IsLatest": true,
   "LastModified": "2023-03-26T13:53:30.014Z",
   "Owner": {
   "DisplayName": "cosbench_user",
   "ID": "cosbench"
   }
   }, {
   "ETag": "\"ed076287532e86365e841e92bfc50d8c\"",
   "Size": 12,
   "StorageClass": "STANDARD",
   "Key": "test2",
   "VersionId": "Z3.arxptMW6iZq1bKaGrgH9czWPKrOf",
   "IsLatest": false,
   "LastModified": "2023-03-26T13:53:30.014Z",
   "Owner": {
   "DisplayName": "cosbench_user",
   "ID": "cosbench"
   }
   }
   ]
   } #

From Casey >>>
in normal sync, the destination zone sends its current object mtime in the If-Modified-Since header to avoid duplicating sync from multiple source zones

it looks like that logic is broken by RGWArchiveDataSyncModule::sync_object(), which generates a random version id for the destination object; so RGWObjFetchCR would try to look up the mtime for that version and fail, so wouldn't be able to send a If-Modified-Since header
<<<