hdfs write fail use AsyncWriter

[俞加平]

def hdfs_copy_stream(src, dst, namenode=None):
    try:
        md5 = hashlib.md5()
        offset = 0
        clt = get_client(src, namenode)
        with clt.read(src, offset=offset, chunk_size=2 ** 16) as reader:
            with clt.write(dst, overwrite=True) as writer:
                for chunk in reader:
                    md5.update(chunk)
                    offset += len(chunk)
                    writer.write(chunk)
        md5_value = md5.hexdigest()
        print('md5 = {}, length = {}'.format(md5_value, offset))
        return RunState.Done, (md5_value, offset)
    except Exception as e:
        print("copy file {} to {} failed: {}".format(src, dst, e))
        return RunState.Error, None

Hi，I implemented a copy method like this.
It works well for large files, but for small files with tens of KB, sometimes the function runs successfully, but the target path of HDFS is indeed a file with a length of 0. I checked it for a long time and found no problem.
When I add time.sleep like this:

offset += len(chunk)
writer.write(chunk)
time.sleep(0.001)

Problem solved.
I really don't know why, so I'm here to ask for your help.

[俞加平]

Also, I used the Requests library for 2.26.0.

[俞加平]

I tried debug and found that Consomer was consuming a generator，But there is retry logic in Requests，when send is retried, there is no data left in the generator.

I'm not sure if that's the reason.