Custom Endpoints

The custom endpoint architecture includes two platform-independent components:

The reverse proxy's configuration file routes all incoming requests:

All shipments contain the file rosent-custom-endpoint-installer-<version>.tar.gz containing an installation script and the files for the reverse proxy and application server. The installation script is currently for Linux and macOS only.

You can install the custom endpoints while Rosette Server is running, or start it after you complete the custom endpoints installation.

The reverse proxy installer comes with a default configuration file for forwarding standard Rosette endpoint calls to the installation on localhost.  It also comes with a templated routing section for custom calls that is updated by the install script.

The file is:  <reverse_proxy_install_directory>/application.yml where the reverse_proxy_install_directory was set when Installing.

Add or modify the existing rules by editing the settings block zuul:

For example, to add a new rule to update the proxy to correctly route calls to a custom application at http://localhost:8182/rest/custom/mycode, add the mycode under the zuul: section:

zuul:
    routes:
      mycode:
       path: /rest/mycode/**
       url: http://localhost:8181/rest/custom/mycode

If your custom code is in Java, put your WAR file into the webapps directory of the Tomcat installation directory. To use other languages, see Adding Non-Java Custom Code.

Your Java web application should define a URI to access the code. Create a rule in the proxy definition to adjust the final path to access the custom code.

Each component, including Rosette Server, has its own, independent, Tanuki Java Wrapper application. Tanuki monitors the health of the application and relaunches if the Java application crashes or gets into a bad state.

In addition, the proxy application exposes two endpoints that can be used to access the proxy health:

The custom endpoint application includes an example application, matchSentences , which calculates the relevance of documents and sentences against a list of keywords. It combines calls to the /sentences and /semantics/vector endpoints with custom code. The custom code uses the cosine similarity function to calculate similarity scores between the sentences and the keywords, as well as comparing the scores to the input threshold value.

The input is a list of keywords, a document or the URL to a document, and a threshold.

The response is the cosine similarity scores and a true/false value indicating if the score is above the threshold for each sentence, and for the entire document.

{
    "keywords": [
       "USA",
       "Iran",
       "attack",
       "protest"
    ],    
    "document": "Iran was planning attacks on    four US embassies 
     when its top general was killed, President Donald Trump says. 
     When asked what threat led to last Friday's US drone strike, 
     he told Fox News: \"I can reveal that I believe it probably
     would've been four embassies.\" \"The killing of Gen Qasem Soleimani, 
     a national hero, came after days of protests at the US embassy in Baghdad.",    
   "threshold": 0.25
}

curl -s http://localhost:8182/rest/custom/matchSentences -XPOST \
-H 'Content-Type: application/json; charset=utf-8' -d \
@matchSentences.json 

zuul:
   routes:
     custom:      
       path: /rest/custom/**      
       url: http://localhost:8183/custom-application/custom-app/services