Get non-ASCII filename from S3 notification event in Lambda

Question

The key field in an AWS S3 notification event, which denotes the filename, is URL escaped.

This is evident when the filename contains spaces or non-ASCII characters.

For example, I have upload the following filename to S3:

my file řěąλλυ.txt

The notification is received as:

{ 
  "Records": [
    "s3": {
        "object": {
            "key": u"my+file+%C5%99%C4%9B%C4%85%CE%BB%CE%BB%CF%85.txt"
        }
    }
  ]
}

I've tried to decode using:

key = urllib.unquote_plus(event['Records'][0]['s3']['object']['key']).decode('utf-8')

but that yields:

my file ÅÄÄλλÏ.txt

Of course, when I then try to get the file from S3 using Boto, I get a 404 error.

Answer 1

tl;dr

You need to convert the URL encoded Unicode string to a bytes str before un-urlparsing it and decoding as UTF-8.

For example, for an S3 object with the filename: my file řěąλλυ.txt:

>>> utf8_urlencoded_key = event['Records'][0]['s3']['object']['key'].encode('utf-8')
# encodes the Unicode string to utf-8 encoded [byte] string. The key shouldn't contain any non-ASCII at this point, but UTF-8 will be safer.
'my+file+%C5%99%C4%9B%C4%85%CE%BB%CE%BB%CF%85.txt'

>>> key_utf8 = urllib.unquote_plus(utf8_urlencoded_key)
# the previous url-escaped UTF-8 are now converted to UTF-8 bytes
# If you passed a Unicode object to unquote_plus, you'd have got a 
# Unicode with UTF-8 encoded bytes!
'my file \xc5\x99\xc4\x9b\xc4\x85\xce\xbb\xce\xbb\xcf\x85.txt'

# Decodes key_utf-8 to a Unicode string
>>> key = key_utf8.decode('utf-8')
u'my file \u0159\u011b\u0105\u03bb\u03bb\u03c5.txt'
# Note the u prefix. The utf-8 bytes have been decoded to Unicode points.

>>> type(key)
<type 'unicode'>

>>> print(key)
my file řěąλλυ.txt

Background

AWS have commited the cardinal sin of changing the default encoding - https://anonbadger.wordpress.com/2015/06/16/why-sys-setdefaultencoding-will-break-code/

The error you should've got from your decode() is:

UnicodeEncodeError: 'ascii' codec can't encode characters in position 8-19: ordinal not in range(128)

The value of key is a Unicode. In Python 2.x you could decode a Unicode, even though it doesn't make sense. In Python 2.x to decode a Unicode, Python first tries to encode it to a [byte] str first before decoding it using the given encoding. In Python 2.x the default encoding should be ASCII, which of course can't contain the characters used.

Had you got the proper UnicodeEncodeError from Python, you may have found suitable answers. On Python 3, you wouldn't have been able to call .decode() at all.