Gringotts is an email delivery API, that supports multiple third party email delivery services with a simple configuration change.
Install Java.
NOTE:
Javaversion must be >=1.8.0_40
Install Maven.
NOTE:
Mavenversion must be >=3.1.1
In the base directory, copy the application.properties.tpl file to application.properties:
$ cp application.properties.tpl application.properties
You'll then need to populate any of the empty properties with your own API keys, or service-specific URLs.
NOTE: If you skip this step, emails will not be able to be delivered.
Assuming you have successfully installed Maven and Java, and mvn and java are both now on your $PATH, you're ready to compile and run gringotts.
First, compile the code using maven:
$ mvn package
If this compiles successfully, fire that bad boy up:
$ java -jar target/gs-spring-boot-0.1.0.jar
PROTIP: You can also set this up in the IDE of your choice (if you're into IDEs) by importing it as a new Maven project.
At this point, you should be able to send an email! Try the following:
$ curl --request POST \
--url http://localhost:8080/v1/email \
--header 'content-type: application/json' \
--data '{
"to": "joshua.wyse@gmail.com",
"to_name": "Josh Wyse",
"from": "noreply@gringotts.com",
"from_name": "Mr Gringotts",
"subject": "Yay, I used Gringotts!",
"body": "<h1>Hooray!</h1>"
}'
Now, if you want to swap out which email service is being used under the hood, simply stop the app, change the email.service property in your application.properties file, and start it back up.
Lastly, the unit tests run as part of the build, but if at any point you want to run just the unit tests, you can do so like this:
$ mvn test
Initially, I was going to spin up a quick node express app to solve this issue. I find them very lightweight, and zero to "hello world" incredibly fast for quick apps like this. That being said, the directions explicitly mentioned demonstrating OO principles, which I just have more experience designing and building in Java, and with statically typed languages, in general. The directions also mentioned the ability to easily modify a configuration file to swap out which 3rd party email provider is being used. To me, that screamed polymorphism and using dependency injection to quickly swap out the underlying implementation of some interface. All of that being said, after a bit of deliberation, I chose Java. I also chose Spring to handle IoC. To be completely honest, I think Spring is a little heavyweight for a quick app like this, as is often my feeling with Spring, but using Spring Boot, I had a simple web server up and running in less than five minutes, with only a few lines of code. Don't judge me too much for using Spring here, although I still feel a little wrong using such a heavyweight library for such a simple application.
I also used maven as the build system because I am familiar with it and it just works. I don't love XML (read: hate XML), but it does the job just fine.
I used JUnit for both unit tests and integration tests. The unit tests run as a part of each maven build. The integration tests do not, since right now, they are going out and making API calls to each email service.
The main entry point of the app is the Application class. This is where Spring Boot is configured, and the proper EmailService bean is created based on the specified email.service property. Spring Boot starts up an embedded web server (Tomcat), and then scans for any RestController beans. I have one of these beans configured, which is the EmailController class. These controllers are where you define the API methods, and tie them to a specific URL. This is where the /v1/email API is defined. That API function has a validator tied to it, which is where any incoming API calls will have their JSON payloads validated to ensure all of the required email fields are legit. Then, the sendEmail function gets called, which delegates to the configured EmailService. The main goal with this, is that the controller/API layer knows about HTTP/HTTP responses/HTTP codes/etc, while the service layer has no knowledge of that at all. It simply knows how to take an Email object, make an API call out to its specific service, and throw any service exceptions if something goes wrong. The controller then handles the service exceptions, and goes about converting those into the proper HTTP response codes, etc.
Given more time, I would abstract out the Unirest HTTP client code that is used into a wrapper. I started down this route, but for the sake of time stopped. The direction I was going, was to create an Http interface and then have a UnirestHttpImpl. This would allow gringotts to be more loosely coupled with any specific HTTP implementation.
To handle the different email services, it felt like the perfect fit for polymorphism. This kept the API controller agnostic of what service it's actually using, since it doesn't need to know. It just gets the EmailService injected into it, and knows how to call the sendEmail API on it. This keeps the controller and service layers loosely coupled from each other. I also wrapped all UnirestException in the service layer and am throwing our generic EmailServiceException so that the unirest code doesn't leak up into the controller layer. This loose coupling allows us to swap out the email service implementation with zero code changes, and loose coupling in general leads to less fragile code and code that is easier to refactor at a later date.
I wanted to keep the response from gringotts consistent back to the client, regardless of what the underlying email service returned. I started with a normalized id and message field for Mandrill and Mailgun, but then Sendgrid just returns a 202, and no fields at all. I debated on what was best here, as Sendgrid isn't even in the requirements, so I didn't want to compromise the design because of that. All of that being said, I ended up going with an empty response and just returning a 202 for all POST /email APIs (which I guess contradicts my previous sentence), but I thought that was sufficient for this API. As a general rule, when I'm designing API, I like to return the fully hydrated created resource with a 201. I do understand sending an email is a little different than creating a new server-side RESTful resource though.
I leveraged Github's issues functionality in order to ensure I listed out all of the requirements that were given for this application. I also included a handful of other items that weren't requirements and were more things I would do if I were to take more time on this. Please visit the "Issues" tab in Github to see some of those outstanding issues. Some of those need to be fleshed out quite a bit more.
Currently, the error handling in gringotts is pretty poor. After making an API call to each service, I just check to see if the response was a 20x response, and if it's not, I serialize the entire JSON response into a String and send it back in an error. Given more time, I'd try to clean this up a bit, and do more testing to ensure I'm handling errors from each service appropriately. I would also test more framework level errors and handle them in a little prettier way (404s, etc.).
Currently, I'm not testing any of the EmailService implementations at all. The way I would do this,
is mock the underlying HTTP endpoints using something like WireMock. This would make it easy to test all of the service code itself, while not depending on the third party email service itself. All in all, I don't have very great test coverage. I'm very passionate about testing, and don't feel like I did a great job communicating that in this app. Given more time, I would spruce up the current tests, as well as create a CI/CD pipeline where we depend on these tests to 1) merge code into Github and 2) before the code gets deployed to production. I also am a subscriber to the "testing pyramid" philosophy, and I would ensure that my unit test:integration test ratio improved compared to where it is now.
The instructions said to "organize, design, document and test your code as if it were going into production". There are quite a few things I would focus on right away if this were to be deployed in production, and most are around deployment and infrastructure. I would not use an embedded tomcat servlet container, but use Tomcat itself. I would also centralize all logging, and look into containerized deployments in order to easily scale.
As part of the Mandrill account setup, they require you to setup a "sending domain" before you can properly deliver emails. I had two of the three steps done to verify this process, however the email verification wasn't working as I wasn't receiving the verification email from them. I reached out to their support to get some help, and for some reason my personal domain email (joshua@joshuawyse.com) had been blacklisted as deliveries had failed there in the past. They reset that and I immediately got my verification email. After finishing the setup for my "sending domain" (DNS entries and email verification), I realized I needed to purchase monthly plan and a minimum of 25k transactions ($30 cost). Spoke with Neil, and he verified I didn't need to go this far. So all that being said, I wasted quite a bit of time on that and I apologize for the delay in turning this in!
As someone who wants to support local business, as well as the best email SaaS provider out there, I also included an integration with Sendgrid to send emails as well. #SupportLocalBusiness 😄