István Fajth found if we delete file with Hadoop shell, namely hadoop fs -rm, without -skipTrash option, the operation would fail in OFS due to the client is renaming the file to /user/<username>/.Trash/ because renaming across different buckets is not allowed in Ozone. (Unless the file happens to be under /user/<username>/, apparently.)
We could override getTrashRoot() in BasicOzoneFileSystem to a dir under the same bucket to mitigate the problem. Thanks Uma Maheswara Rao G for the suggestion.
This raises one more problem though: We need to implement trash clean up on OM. Opened HDDS-3575 for this.