This is an example application deployed on Google App Engine which involves other Google Cloud functionalities in runtime Python 3.7. Until the project is running, the app may be accessed at https://project-ii-gae.appspot.com/.

An image may be uploaded after logging into a valid google account, after which the user receives an email to the specified email with text detected in the image.

The following technologies and skills are used:

• Developing an application running on Google App Engine in Python 3.7 runtime
• Using Google Source Repositories as source control system
• Using Google Cloud Ruild for building container images and deployment automation
• Using Google managed database technologies as data persistency, i.e. Cloud Datastore and Cloud Storage
• Using Google Cloud Functions to act in response to events
• Orchestrating Continuous Integration system to build application artefacts
• Preparing proper deployment instrumentation

The user flow defined for the project is as follows:

1. As a user I can navigate to a website where I can log in using GCP account credentials (users without valid GCP credentials cannot log into the system).
2. After successful login I can upload an image in JPG or PNG format and this image is stored in Cloud Storage bucket (see Note 2)
3. I have a Cloud Function GCF#1 with a Cloud Storage trigger that discovers that an image was uploaded to Storage bucket#1. Cloud Function scales down the picture and stores the result in Cloud Storage bucket#2.
4. There is another Cloud Function (GCF#2) that processes images from bucket#2. GCF#2 discovers transformed images stored in bucket#2 and puts the message in Pub/Sub channel that is a trigger for GCF#3. \
5. GCF#3 makes a call to Vision API and discover text visible in the image. Result of Vision API call is stored in the Datastore.
6. Finally, GCF#3 should send an email with the link to the image to GCP user and. Email should inform the user about successful or not successful operation of image transformation. Email should contain signed URLs to original and transformed image. Email should also contain text discovered for this specific image by Vision API
7. I can press 'Logout' button to log out of the service. Once I press 'Logout' I'm navigated back to login/initial page where there is a "Login" button I can use to log back into the system.

The implementation of the above user flow takes certain liberties, but those are explained below.

Within the repository one may find the setup.sh shell script. More specific assumptions may be found within the comments of the files, but in general - the script may be used in an empty project to build the project out of the box in a Google Cloud environment. Please note, that the script does not set up a Git Repository and Cloud Build. These are optional and must be created manually.

The script may be run with setup.sh [PROJECT_ID] [CLIENT_ID],

where the required fields are empty project ID and a Client ID, which may be created after configuring the application consent form (ie. first create a project, then setup the consent form and finally run the script).

If one chooses to set up the project manually, the script file may serve as a collection of instructions.

The following overview sections are set in order of the project requirements table.

Based on the specifications described in the section The Project Overview, presented in the diagram below is a generalized techincal flow of the application. The steps are descirbed in the next subsection.

Once Google App Engine deploys the application, the following frontend may be accessed by the user.

The user must log in to be able to upload an image. After signing in with their Google account through OAuth, the upload form becomes available. An image may be uploaded and optionally the recipient email may be changed. After upload, the appropriate response is generated (error/success). Sign out is available or the user may repeat the process.

The page is serviced through Flask (Python package). A session is created for the logged in user to authorize uploads. The form is intercepted upon submit and then posted by AJAX. This is relatively unstable, but should pass for a small sized project.

If an error appears in file uploading, the user will be redirected to an error page with the appropriate error code.

The app is always serviced at https thanks to Google's built-in redirect system found in the app.yaml setup file.

The backend for this upload (Flask) first validates the data for correctness. Then the digital digest is calculated (to check if image exists in the database). Afterwards appropriate steps are taken to upload the image to the first bucket with a timestamp-based id. Datastore entity is initialized.

The first GCF is triggerd upon an image upload to bucket-1. It downloads the image and rescales it to a width of 512 pixels while respecting aspect ratio.

The resulting image is uploaded to bucket-2. The related entity in Datastore is updated with both image URLs.

The second GCF is triggered upon an image upload to bucket-2, ie. after GCF1 runs. The sole existence of this function is to publish a message to the pub/sub topic 'rescaled-images'. The message posts an attribute with the filename of the uploaded file.

This third and final GCF is the bread and butter of the backend functionality. It is triggered upon a message publication to the 'rescaled-images' topic. It stores the related image into memory and uses Google's Vision API to detect visible text in the image. This text is stored in the related entity Datastore.

A simple SMTP server is set-up with a dedicated Google/Gmail account to send the results to the uploader/recipient email.

Customer-Managed Encryption is available through Google's KMS. A keyring/key is created for the project and permissions are granted to the App Engine service account. Due to this, the service account will now always encrypt files uploaded to bucket-1 are encrypted with this symmetric key.

A cloudbuild.yaml file is present in the repository. If the project is integrated with a Google Cloud Source Repository, every git push will run cloudbuild.yaml if set up in the Cloud Build section. The file runs unit tests on GCF, Flask and then integration tests on both components. If these succeed, the newest versions of the GCF and Google App Engine are deployed.

Very basic unit tests (Python/unittest) are prepared for the GCFs, Flask and integration of both. Their existence is rather just a use for presenting how Google Build may run these tests on git push.

Below is the list of unit tests.

• GCF
  • GCF1 - test if image was uploaded to second bucket (test014_image_exists_in_bucket_2)
  • GCF1 - test if image was rescaled properly (test015_rescale_success)
  • (Disabled) GCF2 - test if message was properly published (test110_published_trigger)
    • Disabled due to broken Google pub/sub API. The API randomly fails to pull freshest messages
• Flask
  • test if correct image file is successfully uploaded (test010_positive_form_post)
  • test if negative image file is unsuccessfully uploaded (test011_negative_form_post)
  • test if oversized image is unsuccessfully uploaded (test012_oversize_form_post)
  • test if undersized image is unsuccessfully uploaded (test013_undersize_form_post)
  • test if user not logged in results in unsuccessful upload (test014_not_logged_in)
  • test if invalid email format results in unsuccessful upload (test015_invalid_email)
  • test if the image has been uploaded to bucket-1 (test020_image_stored_bucket1)
• Integration
  • test if the datastore entity has populated fields (test010_datastore_entries_exist)
  • test if text has been detected in image (test015_vision_text)
  • test if the storage blob files are public (test021_images_public)

Google App Engine is yet another solution pulled from the never-ending catalogue of Google products, specifically created for producing applications. The idea is simple: backend, frontend, side-end, whatever-end - the solution will host it and strive to run it perfectly. Unfortunately, that is rarely the case. The Google platform often riddles the user with false errors or allows itself a casual self-entitled prank by disabling random components of the Google Cloud platform every now and then. Alas, the terrible contrast between the marketing material and reality is a subject for another time.

General

App Engine

Functions

Git Deploy

gcloud

Setup for python37

HTTPS always for appspot apps

relative paths in app engine

read Flask uploads from memory

templates in App Engine Flask

uploading files

google sign-in

js functionality

html form interception for authentication

hashing

reading Flask FileStorage into Pillow

datastore python ref

google storage blob python ref

google storage buckets python ref

image manipulations with PIL and google storage

blobs saved as strings of bytes

maintain aspect ratio

pubsub ref

pubsub synchron. pull

encoding as bytes

text detection

relative paths to container names (pubsup topics)

sendgrid