Allow passing files not at beginning to 'ResumableUpload'

[retnikt]

Environment details

Ubuntu, Python 3.7, Google Cloud Storage 1.23.0

Steps to reproduce

1. Create a Blob
2. Use blob.upload_from_file(fp) where fp is not at the beginning of the stream (i.e. it has been sought)

Code example

from google.cloud import storage
client = storage.Client()
bucket = client.bucket("somebucket")
blob = bucket.blob("path/to/blob.txt")
fp = io.BytesIO(b'\0' * 100)
fp.seek(40)
blob.upload_from_file(fp)

Traceback

Traceback (most recent call last):
  ...
  File "/.../google/cloud/storage/blob.py", line 1262, in upload_from_file
    created_json = self._do_upload(
  File "/.../google/cloud/storage/blob.py", line 1172, in _do_upload
    response = self._do_resumable_upload(
  File ".../google/cloud/storage/blob.py", line 1110, in _do_resumable_upload
    upload, transport = self._initiate_resumable_upload(
  File "/.../google/cloud/storage/blob.py", line 1059, in _initiate_resumable_upload
    upload.initiate(
  File "/.../google/resumable_media/requests/upload.py", line 338, in initiate
    method, url, payload, headers = self._prepare_initiate_request(
  File "/.../google/resumable_media/_upload.py", line 415, in _prepare_initiate_request
    raise ValueError(u"Stream must be at beginning.")
ValueError: Stream must be at beginning.

Description

I cannot see any reason why there is the check if stream.tell() != 0. It seems to serve only to frustrate users trying to upload from custom unseekable file-like objects (e.g. googleapis/google-cloud-python#7282), or only parts of files. Can this not be removed, or moved to somewhere else more relevant in the codebase if it serves a purpose only thence.

[retnikt]

whoa, ten-thousandth issue!

[shunghsiyu]

I second this, it would be nice if we don't need wrapper around unseekable file-like object to make it work with upload_from_file.

The if stream.tell() != 0 check code seems to belong to googleapis/google-resumable-media-python though.

[vperezb]

Hi all, I arrived here because I was struggling to upload an image transformed using PIL to do a reduction in size. Was trying in the same way you did to upload an object from "bytes" type.

I hope this is the way I should participate in an issue like this one, by providing a fix that worked for me in my particular case. And would have helped me in the first place when I reached the issue.

From the issue snippet provided by @retnikt :

from google.cloud import storage
client = storage.Client()
bucket = client.bucket("somebucket")
blob = bucket.blob("path/to/blob.txt")
fp = io.BytesIO(b'\0' * 100)
fp.seek(40)
blob.upload_from_file(fp)

I've just changed the last line using the upload_from_string method instead of the upload_from_file and the get_value() method from io.Bytes object.

import io
from google.cloud import storage
client = storage.Client()
bucket = client.bucket("somebucket")
blob = bucket.blob("path/to/blob.txt")
fp = io.BytesIO(b'\0' * 100)
fp.seek(40)
blob.upload_from_string(fp.getvalue())

Hope it helps other people reaching this issue for the same reason I did, apologies if is not the way to do it.

[retnikt]

@vperezb the issue with this is, if the file-like object is a stream rather than in memory, e.g. network or disk, then using a string requires loading all the data into memory.

[tseaver]

google.resumable_media requires:

• that the stream be seekable (see AttributeError: 'mod_wsgi.Input' object has no attribute 'tell' #123)
• that the stream be at the start.

It is possible that some combination of stream-not-at-start, a passed total_bytes might still work, but this issue needs to be traced in that repository.

[Thirunayan22]

Hi all, I arrived here because I was struggling to upload an image transformed using PIL to do a reduction in size. Was trying in the same way you did to upload an object from "bytes" type.

I hope this is the way I should participate in an issue like this one, by providing a fix that worked for me in my particular case. And would have helped me in the first place when I reached the issue.

From the issue snippet provided by @retnikt :

from google.cloud import storage
client = storage.Client()
bucket = client.bucket("somebucket")
blob = bucket.blob("path/to/blob.txt")
fp = io.BytesIO(b'\0' * 100)
fp.seek(40)
blob.upload_from_file(fp)

I've just changed the last line using the upload_from_string method instead of the upload_from_file and the get_value() method from io.Bytes object.

import io
from google.cloud import storage
client = storage.Client()
bucket = client.bucket("somebucket")
blob = bucket.blob("path/to/blob.txt")
fp = io.BytesIO(b'\0' * 100)
fp.seek(40)
blob.upload_from_string(fp.getvalue())

Hope it helps other people reaching this issue for the same reason I did, apologies if is not the way to do it.

Thank you, worked for me