Skip to content

zerobase-io/smart-tracing-api

Repository files navigation

Zerobase API Server

This repository contains the back end for the Zerobase smart tracing platform. Refer to the smart-tracing repo for the front end. We are using Kotlin and Neo4j with cloud storage on Heroku. We are using Dropwizard for the REST framework. Kotlin requires Java and Maven.

Any commands shown in this read-me are written in bash. If you are on Windows or use an alternate shell, like fish, please adjust the commands accordingly.

There are other README files throughout the project. They are located in the source folders that most closely align with the readme contents.

Architecture Notes

  • We are leveraging the type system as much as possible. We don't want stringly typed things and why write our own validations when we can just use what's already there?
    • Inline Types - These are mostly just string-wrapper classes provide some context as they are passed around
    • Unsigned Numbers - There are a whole bunch of things that want numbers no less than 0
  • We are trying to avoid coupling as much as possible. To this end, the rest endpoints don't do anything but execute database operations and then fire events to trigger any additional behaviors. This design allows us to add new behaviors without having to constant touch and bloat the rest resources.
  • Dependency injection is your friend. Learn to love it. It's a huge win on testing, among other things.
  • There are 2 levels of automated testing relevant to this project, and I'm going to use Google's terms: small and medium. Both have their place.
    • Small Tests - These are usually referred to as "unit tests". They narrowly test a specific class or method. Everything is mocked. These tests are great for complex pieces and validating individual behaviors, but has at least 2 main problems:are limited in verifying the app is functioning
      • They don't verify the service is working as it should. You can have extremely robust small test suites that validate each piece and still have a broken app that won't start.
      • They make refactoring the internals painful. Since all interactions are mocked in small tests, if you change any of the things that were mocked, you have to rewire all the tests. Sometimes, that's easy using IDEs and tooling and sometimes it's extremely painful because the refactor isn't that simple.
    • Medium Tests - These are service level tests. Medium tests, unlike small tests, have access to resources on localhost. This allows for a local database to be used. This type of test provides a greater confidence that the app will work when deployed because to run, the app must be able to turn on, wire up all the pieces (like database connections), and produce the correct API outputs based on the inputs. This type of test allows easy refactoring of the internals because it verifies that the API contract is maintained, regardless of internal implementation details.

Development Setup

Kotlin

The backend is written in Kotlin. While you can work on it in any editor, such as vim or VS Code, it is significantly easier to use an IDE. We recommend IntelliJ. If you are using Intellij, the Kotlin plugin must be at least 1.3.70.

Java 11

This project targets the current LTS version of Java: 11. You are welcome to use any of the JDK implementations locally, but deployments will be done using an AdoptOpenJDK build. If you want to use Oracle's official JDK, it can be dowloaded here. We recommend, however, that you install the JDK using a package manager, such as brew on macOS, to easily stay up to date.

Maven

Maven is a dependency management and build tool that is commonly used in the Java world. It is available for installation via package managers, such as brew on macOS, for easy updating (recommended way) but can also be installed manually.

Install manually

  • Download Maven from here
  • Follow these installation instructions here
  • Add the full path to your ~/.bash_profile, ~/.zshrc, or similar.

Docker

Docker is used to run external services locally for front-end development or tests instead of requiring access to a cloud environment. Some of the resources can be run without docker, but some don't really have local install options. Install docker manually by following this guide or via a package manager.

Running Locally

Docker Compose

We have a docker compose file that will spin up all the pieces necessary and expose the API on your local machine. There are several environment variables that can be used to configure it:

  • DB_PORT: Used if you want to connect to the gremlin server manually and play with the graph.
  • APP_VERSION: By default, it will use latest. If you need to run a specific version, or one that isn't in DockerHub yet, set this to the version tag you want.
  • `APP_PORT**: By default, it will be 9000.

Not changing the backend

To run the backend when not intending to work on it (useful if you are working only on the front end or testing it) you can pull it in from dockerhub with

docker-compose -f docker-compose-full-stack.yml upp app

Alternately, run just dependencies from docker-compose and build/develop on your machine

$ mvn clean install
$ docker-compose -f ./docker-compose-dependencies.yml up database aws

Manually with Docker

Database - Gremlin

  • Pull the Gremlin server image
    $ docker pull tinkerpop/gremlin-server
  • Run the gremlin image. By default, the Gremlin server does not support the string ids we are using with Neptune. As such we need to override the settings so it will.
    $ docker run -d \
            -p 8182:8182 \
            -v $(pwd)/src/test/resources/tinkergraph-overrides.properties:/opt/gremlin-server/conf/tinkergraph-empty.properties \
            --name=zerobase-db \
            tinkerpop/gremlin-server:latest

By default, there are no credentials for the local install

Localstack - AWS Fakes

Follow their startup documentation: https://github.com/localstack/localstack.

App

$ docker run -d --name=zerobase-api \
    -e WRITE_ENDPOINT=<gremlin-api host> \
    -e DB_PORT=<gremlin mapped port> \
    -e AWS_SES_ENDPOINT=<localstack ses http url (including port)>
    zerobaseio/smart-tracing-api:<version>

Manually without Docker

Run Gremlin without Docker

If you're unable to use Docker to run Gremlin, there is an alternative available. The desktop version can be downloaded from here. The configuration instructions are found here. The default configuration should suffice.

AWS SES

Follow the documentation for a non-Docker SES fake. Here's one: https://github.com/csi-lk/aws-ses-local

Project

After cloning the project there are multiple ways to deploy it locally: from dockerhub, using an IDE or via the command line. There is also a docker-compose for setting up all dependencies. By default, the app listens on port 9000. You can override that with an environment variable of PORT if you need to. The local-config.yml defaults to localhost and 8182 for the database. Both can be overriden with environment variables, using WRITE_ENDPOINT and DB_PORT respectively.

Running in an IDE

The following directions use Intellij as the IDE, but the steps should be similar if you are using a different IDE.

  • If necessary, navigate to File/Project Structure. Update the SDK for the project to JDK11.
  • Open Main.kt
  • In Main.kt, click on the run button next to main and select Create Main.kt to create a run configuration.

main

  • Set the name to something reasonable. If running locally, it's recommended to use the local-config.yml instead of config.yml. The configuration will target running locally against Gremlin, while the normal config will target AWS Neptune.
  • Set the program arguments as server <the config file you want to use>. Options for config files are:
    • src/main/resources/config.yml
    • src/main/resources/local-config.yml
  • By default, the local-config.yml expects the database to be running at localhost:8182. If you need to override the port, set an environment variable of DB_PORT to the port your gremlin server is running on.

env

  • Click OK and run the configuration you just made.

Running from the command line.

  • From the project's root directory, build the project.

    $ mvn clean install

    Alternately, to do a build with fewer checks

      mvn package -Dbasepom.check.skip-all
  • Set the environment variables (if needed), either as an export or inline, and run the jar.

    You should not have to set environment variables at all if running the default configuration for dependencies via docker-compose-dependencies

    • Export
  •   $ export DB_PORT=12345
      $ java -jar target/smart-tracing-api.jar server target/classes/local-config.yml
    ```
    
    • Inline
        $ DB_PORT=12345 java -jar target/smart-tracing-api.jar server target/classes/local-config.yml

Debugging / Calling end points

  • Once you're running the project, you can double check if all is well by visiting http://localhost:8081 in your browser. You shoudl see a simple page that exposes metrics and healthcheck results. You can also curl http://localhost:8081/healthchecks if you prefer the command line.
  • App endpoints are available at http://localhost:9000

Database Model

Database Model