Github user NicoK commented on the issue:
https://github.com/apache/flink/pull/5601
Different FS implementations try to circumvent the problem (or not), e.g.
Presto's implementation implements the retries itself whenever it cannot find
an object, while Hadoop's implementation did not (may depend on the version and
may be subject to change).
Also I don't know what the guarantees say about someone else reading your
write - that would be the actual real usecase: artefact uploading by the
client, downloading the artefacts via yarn on the worker nodes.
---
