| folder | description |
|---|---|
| android | the android client |
| web | the web client |
| server | the backend server |
| mongo-k8s-sidecar | the opensource project used for mongo-replication |
| node_modules | the dependencies resolved by npm (created by npm) |
Run the deploy_script.sh file to start the deployment of backend server and the
web application. Please note that you may need to add execute permission for the
deploy_script.sh file. Run sudo chmod +x deploy_script.sh to assign the
permissions run ./deploy_script.sh to execute the script.
The deploy_script.sh creates a cluster named "mean-cluster". The cluster contains 2 nodes of machine type "n1-standard-2". The opensource project https://github.com/leportlabs/mongo-k8s-sidecar.git is used to perform mongoDB replication. Currently, 3 pods/replicas of mongoDB are created which are handled by 3 respective services. In case a mongo replica fails the respective service will autometically create a replacement replica. The web backend and web app are deployed under the same container, which is replicated over 2 replicas. After the server deployment is complete, the deploy_script.sh invokes another script called "create_test_user.sh" which is responsible to create a demo user in the database. The demo user has the following credentials: username = "username" and password = "password".
Overall, we have 5 pods in total: 2 for web-app/backend and 3 for mongoDB
Finally, deploy_script.sh also runs android_deploy.sh script to build Android
application and deploy it on device. android_deploy.sh script can be run
separately from deploy_script.sh. Run sudo chmod 777 android_deploy.sh to
assign the permissions and run ./android_deploy.sh to execute the script.
Demo user: username: "username" password: "password"
The web client uses a JavaScript port of the Tesseract library called TesseractJS which is developed by MIT [1]. The scan is using training data for the English language.
The backend uses the same library as the web client, TesseractJS. There is an issue with the library in case of NodeJS backends and JPEG images, and the scan would be extremely slow. Thus the backend converts all JPEG images into the PNG format before scanning. The backend returns processing times with the scan results for each image.
Scanning at the backend has been adapted to synchronous serial processing from asynchronous simultaneous processing to get more accurate per-image processing time results. Otherwise the later images would show inaccurately long processing times. This may lead to marginally longer total scan time but ensures that the per-image results are sensible.
[1] http://tesseract.projectnaptha.com/
All traffic is forced through HTTPS instead of HTTP, and self-signed certificates are used for this (so the web client should be expected to ask for a security exception).
The backend authenticates the necessary endpoints by using JSON Web Tokens. A token is generated when a user logs in (or registers which is also an immediate login) and the client needs to include that token in all authenticated requests.
For normally registered new users, hashes of the passwords are stored in the database (generated with the NodeJS crypto module [1]). On login, the backend verifies matching hashes and provices the client an authentication token.
[1] https://nodejs.org/api/crypto.html
Facebook login is implemented by first using the Facebook Login API to make a request from the client to Facebook to login the user in their service, and if Facebook provides successfully an access token, the token is forwarded with another request to the backend server which then verifies that token with Facebook by using the Passport library [1] with a token strategy [2] designed for the purpose of verifying the access token.
If the token is successfully verified, the backend server creates a new user account if that user has never logged in. Once a user account exists, an authentication token is generated as for a normal login with a User schema method, and the token is sent to the client as if they logged in without an external Identity Provider.
[1] http://passportjs.org/ [2] https://github.com/drudge/passport-facebook-token
The database is a MongoDB database handled by the backend with the Mongoose library [1]. User and scan data including references to filenames are stored in normal Mongoose schemas, but files are stored using MongoDB GridFS. When a request for a remote scan is made, the backend temporarily stores the images on disk, makes a thumbnail of 200px width of the images, and then stores both the originals and the thumbnails to the database with GridFS using streams [3].
A scan is performed using the locally temporarily stored images, and references for filenames for the originals and thumbnails are stored into the Scan objects so that they can be retrieved later on. After the local files are no longer needed and everything is in the database, the local files are deleted.
[1] http://mongoosejs.com/ [2] https://docs.mongodb.com/v3.2/core/gridfs/ [3] https://www.npmjs.com/package/gridfs-stream
The backend runs a Cron job every day at 00:00 Finnish time to remove from the database those original images that are over 7 days old. After that only thumbnails can be shown in the clients for the related scan results. The NPM module cron is used for this [1].
[1] https://www.npmjs.com/package/cron
The web client is implemented as a single page application using AngularJS. It performs local OCR in the device but requires an internet connection for proper functionality since it is a web page served by the backend. TesseractJS also uses a CDN to store some of its internal JavaScript files, so an internet connection is required for that reason as well, but all actual scan processing is performed locally.
Note that due to the HTML5 API and file handling in file inputs in mobile devices,
the input designed to take in files from the file system / existing library does
not immediately launch the library since iOS and Android want to display their
own popup for the file source. However, the other input does lead directly to
the device camera since an attribute has been implemented in the HTML5 API for
that purpose (the capture attribute).
Android client is implemented using following libraries: Retrofit and Okhttp for remote API access, Gson for json serialization and deserialization, facebook-android-sdk for facebook authentication, Slider and Glide for image viewing and tess-two for OCR. Android app covers all requirements described in assignment. Challenge implementation is described below.
The authentication related endpoints return a token to be used with all authenticated requests:
sample request data: {"username":"username", "password":"password"},
content type is Content-Type: application/json
sample request data: {"username":"username", "password":"password"},
content type is Content-Type: application/json
sample request data: {"access_token": "sometokenfromfacebook"},
content type is Content-Type: application/json
The following endpoints are authenticated with the token provided by the login or register endpoints:
The request is files from e.g. a form with but always with content type Content-Type: multi-part/formdata
Returns the scan history for the logged in user
Returns from the MongoDB GridFS database the specified file
Deletes from the MongoDB GridFS database the specified file
Please add the hash key to the facebook application in developers.facebook.com. This hash can be checked when you try login using facebook for the first time.
Android application checks the availability of network connection and reacts accordingly by offering the Local operating mode only. It also stores OCR history and thumbnails in cache (but not the original images). Without network connection the original images on OCR details screen will be empty. Application recognize network status only after restart.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent