Rekognition.md

Background

This is an interactive portion of the workshop. The end state is that we will create a database trigger that, when a document is inserted with an attribute called URL that it will call the AWS Rekcognition API to describe what the JPEG is a picture of.

Here will do this with the AWS object within Stitch.

The cluster should already be deployed so follow that tutorial first.

Using Stitch to Make REST Requests

Enable outbound REST calls

• On the main Atlas landing page, click "Stitch" under services

• Press the green "Create New Application" button in the top right
• Give the application a name. Here I chose "webapi" and leave everything else default. Click the green "Create" button. This may take several minutes to initially configure.

• On the left, choose "Services" under "Control"
• Press the green "Add a Service" in the top right

• In the "Add a Service" page, choose the "AWS" service, give it a name (here I chose aws), complete filling in the form with your AWS Access Key ID, create a new secret and store it in Secret Access Key Value, then press the green "Add Service" button

• Add a rule by giving it a name, here rek, then click the green "Add Rule"
• Click on rek
• In the drop down choose the API of rekognition and the action of DetectLabels then save

• Add another service. Press the green "Add a Service" in the top right

• In the "Add a Service" page, choose the "HTTP" service, give it a name (here I chose hook), then press the green "Add Service" button

• We now have a service which knows how to issue HTTP requests.
• Next we want to create a trigger that, when a document gets inserted that has a URL attribute, it will call AWS Rekognition to identify what it is a picture of, then update the document.

Enable triggers

• Press the "Triggers" button then the "Add a Trigger" green button

• We will fill out the form accordingly such that:
• The type is Database Trigger
• We can leave the default name if you want
• Make sure it is enabled
• We can leave Event Ordering on or off, doesn't matter here
• Choose your cluster (there should only be one)
• Create a new database called rek
• Create a new collection called rek
• Operation type is Insert only, leave all others unchecked

• Choose to get the full document
• Create a new function, here I am calling it rekTrigger_oninsert

• put the following code in the text box for the trigger. Comments are inline

// add the async declaration so that way we simplify javascript promises
exports = async function(changeEvent) {
  // if we checked the box above, we can have the full document when the trigger runs
  var doc = changeEvent.fullDocument;
  
  // only run if the document has a url attribute
  if(doc.hasOwnProperty('url')){
    // handler to db
    var conn = context.services.get("mongodb-atlas").db("rek").collection("rek");
    // we created these two services earlier
    const aws = context.services.get('aws');
    const httpService = context.services.get("hook");
    
    // aws rekognition expects a bit stream, not the path to the image
    // so do an http get to get the image first
    var uri = doc.url
    var args = {"url":uri, "headers": {"Content-Type":["image/jpeg"]}};
    var req = httpService.get(args);
    var img = await req;
    
    // we now have the bits making up the image. pass them to aws rek
    try {
        var awsreq = aws.rekognition().DetectLabels({"Image": {"Bytes":img.body}});
        var res = await awsreq;
        // update  that mongodb document with the response
        // here we set the rek attribute to what the API returned
        conn.updateOne({_id:doc._id},{$set: {rek: res}});
      } catch (error) {
        console.log(JSON.stringify(error));
      }
    
  }
};

• save the trigger
• Insert a document into the database with a URL attribute which is a full path to a JPEG
• Notice the trigger runs and updates the document with the AWS Rekognition results

Extra Credit:

MongoDB Atlas and Stitch have extra features up its sleeve to improve this section. Here are some examples you may be interested in furthering this application:

• Try using the S3 API within stitch to query or put files there
• Make a web page using MongoDB Stitch QueryAnywhere to show these records
• Upload that webpage to Stitch Hosting