Google cloud function with wand stopped working

Question

I have set up 3 Google Cloud Storge buckets and 3 functions (one for each bucket) that will trigger when a PDF file is uploaded to a bucket. Functions convert PDF to png image and do further processing.

When I am trying to create a 4th bucket and similar function, strangely it is not working. Even if I copy one of the existing 3 functions, it is still not working and I am getting this error:

Traceback (most recent call last): File "/env/local/lib/python3.7/site-packages/google/cloud/functions_v1beta2/worker.py", line 333, in run_background_function _function_handler.invoke_user_function(event_object) File "/env/local/lib/python3.7/site-packages/google/cloud/functions_v1beta2/worker.py", line 199, in invoke_user_function return call_user_function(request_or_event) File "/env/local/lib/python3.7/site-packages/google/cloud/functions_v1beta2/worker.py", line 196, in call_user_function event_context.Context(**request_or_event.context)) File "/user_code/main.py", line 27, in pdf_to_img with Image(filename=tmp_pdf, resolution=300) as image: File "/env/local/lib/python3.7/site-packages/wand/image.py", line 2874, in __init__ self.read(filename=filename, resolution=resolution) File "/env/local/lib/python3.7/site-packages/wand/image.py", line 2952, in read self.raise_exception() File "/env/local/lib/python3.7/site-packages/wand/resource.py", line 222, in raise_exception raise e wand.exceptions.PolicyError: not authorized/tmp/tmphm3hiezy' @ error/constitute.c/ReadImage/412`

It is baffling me why same functions are working on existing buckets but not on new one.

UPDATE: Even this is not working (getting "cache resources exhausted" error):

In requirements.txt:

google-cloud-storage
wand

In main.py:

import tempfile

from google.cloud import storage
from wand.image import Image

storage_client = storage.Client()

def pdf_to_img(data, context):
    file_data = data
    pdf = file_data['name']

    if pdf.startswith('v-'):
        return 

    bucket_name = file_data['bucket']

    blob = storage_client.bucket(bucket_name).get_blob(pdf)

    _, tmp_pdf = tempfile.mkstemp()
    _, tmp_png = tempfile.mkstemp()

    tmp_png = tmp_png+".png"

    blob.download_to_filename(tmp_pdf)
    with Image(filename=tmp_pdf) as image:
        image.save(filename=tmp_png)

    print("Image created")
    new_file_name = "v-"+pdf.split('.')[0]+".png"
    blob.bucket.blob(new_file_name).upload_from_filename(tmp_png)

Above code is supposed to just create a copy of image file which is uploaded to bucket.

Answer 1

Because the vulnerability has been fixed in Ghostscript but not updated in ImageMagick, the workaround for converting PDFs to images in Google Cloud Functions is to use this ghostscript wrapper and directly request the PDF conversion to png from Ghostscript (bypassing ImageMagick).

requirements.txt

google-cloud-storage
ghostscript==0.6

main.py

import locale
import tempfile
import ghostscript

from google.cloud import storage

storage_client = storage.Client()

def pdf_to_img(data, context):
    file_data = data
    pdf = file_data['name']

    if pdf.startswith('v-'):
        return 

    bucket_name = file_data['bucket']

    blob = storage_client.bucket(bucket_name).get_blob(pdf)

    _, tmp_pdf = tempfile.mkstemp()
    _, tmp_png = tempfile.mkstemp()

    tmp_png = tmp_png+".png"

    blob.download_to_filename(tmp_pdf)

    # create a temp folder based on temp_local_filename
    # use ghostscript to export the pdf into pages as pngs in the temp dir
    args = [
        "pdf2png", # actual value doesn't matter
        "-dSAFER",
        "-sDEVICE=pngalpha",
        "-o", tmp_png,
        "-r300", tmp_pdf
        ]
    # the above arguments have to be bytes, encode them
    encoding = locale.getpreferredencoding()
    args = [a.encode(encoding) for a in args]
    #run the request through ghostscript
    ghostscript.Ghostscript(*args)

    print("Image created")
    new_file_name = "v-"+pdf.split('.')[0]+".png"
    blob.bucket.blob(new_file_name).upload_from_filename(tmp_png)

Anyway, this gets you around the issue and keeps all the processing in GCF for you. Hope it helps. Your code works for single page PDFs though. My use-case was for multipage pdf conversion, ghostscript code & solution in this question.

Answer 2

This actually seems to be a show stopper for ImageMagick related functionalities using PDF format. Similar code deployed by us on Google App engine via custom docker is failing with the same error on missing authorizations.

I am not sure how to edit the policy.xml file on GAE or GCF but a line there has to be changed to:

<policy domain="coder" rights="read|write" pattern="PDF" />

@Dustin: Do you have a bug link where we can see the progress ?

Update:

I fixed it on my Google app engine container by adding a line in docker image. This directly changes the policy.xml file content after imagemagick gets installed.

RUN sed -i 's/rights="none"/rights="read|write"/g' /etc/ImageMagick-6/policy.xml

Answer 3

This is an upstream bug in Ubuntu, we are working on a workaround for App Engine and Cloud Functions.

Answer 4

While we wait for the issue to be resolved in Ubuntu, I followed @DustinIngram's suggestion and created a virtual machine in Compute Engine with an ImageMagick installation. The downside is that I now have a second API that my API in App Engine has to call, just to generate the images. Having said that, it's working fine for me. This is my setup:

Main API:

When a pdf file is uploaded to Cloud Storage, I call the following:

response = requests.post('http://xx.xxx.xxx.xxx:5000/makeimages', data=data)

Where data is a JSON string with the format {"file_name": file_name}

On the API that is running on the VM, the POST request gets processed as follows:

@app.route('/makeimages', methods=['POST'])
def pdf_to_jpg():
    file_name = request.form['file_name']

    blob = storage_client.bucket(bucket_name).get_blob(file_name)

    _, temp_local_filename = tempfile.mkstemp()
    temp_local_filename_jpeg = temp_local_filename + '.jpg'

    # Download file from bucket.
    blob.download_to_filename(temp_local_filename)
    print('Image ' + file_name + ' was downloaded to ' + temp_local_filename)

    with Image(filename=temp_local_filename, resolution=300) as img:
        pg_num = 0
        image_files = {}
        image_files['pages'] = []

        for img_page in img.sequence:
            img_page_2 = Image(image=img_page)
            img_page_2.format = 'jpeg'
            img_page_2.compression_quality = 70
            img_page_2.save(filename=temp_local_filename_jpeg)

            new_file_name = file_name.replace('.pdf', 'p') + str(pg_num) + '.jpg'
            new_blob = blob.bucket.blob(new_file_name)
            new_blob.upload_from_filename(temp_local_filename_jpeg)
            print('Page ' + str(pg_num) + ' was saved as ' + new_file_name)

            image_files['pages'].append({'page': pg_num, 'file_name': new_file_name})

            pg_num += 1

    try:
        os.remove(temp_local_filename)
    except (ValueError, PermissionError):
        print('Could not delete the temp file!')

    return jsonify(image_files)

This will download the pdf from Cloud Storage, create an image for each page, and save them back to cloud storage. The API will then return a JSON file with the list of image files created.

So, not the most elegant solution, but at least I don't need to convert the files manually.