hadoop get 'No such file or directory'
You may want to attach some prefix to local_file, e.g. str(int(time.time()*1000000)) + local_file
_COPYING_ is a staging file during downloading. If you used the same local_file name to download in a multi-process job, some process finished earlier therefore local_file._COPYING_ was deleted. As a result, other processes that run a little bit slower would find local_file._COPYING_ doesn't exist.