Car Pooling Service Challenge

Design/implement a system to manage car pooling.

At Cabify we provide the service of taking people from point A to point B. So far we have done it without sharing cars with multiple groups of people. This is an opportunity to optimize the use of resources by introducing car pooling.

You have been assigned to build the car availability service that will be used to track the available seats in cars.

Cars have a different amount of seats available, they can accommodate groups of up to 4, 5 or 6 people.

People requests cars in groups of 1 to 6. People in the same group want to ride on the same car. You can take any group at any car that has enough empty seats for them. If it's not possible to accommodate them, they're willing to wait until there's a car available for them. Once a car is available for a group that is waiting, they should ride.

Once they get a car assigned, they will journey until the drop off, you cannot ask them to take another car (i.e. you cannot swap them to another car to make space for another group).

In terms of fairness of trip order: groups should be served as fast as possible, but the arrival order should be kept when possible. If group B arrives later than group A, it can only be served before group A if no car can serve group A.

For example: a group of 6 is waiting for a car and there are 4 empty seats at a car for 6; if a group of 2 requests a car you may take them in the car. This may mean that the group of 6 waits a long time, possibly until they become frustrated and leave.

Evaluation rules

This challenge has a partially automated scoring system. This means that before it is seen by the evaluators, it needs to pass a series of automated checks and scoring.

Checks

All checks need to pass in order for the challenge to be reviewed.

• The acceptance test step in the .gitlab-ci.yml must pass in master before you submit your solution. We will not accept any solutions that do not pass or omit this step. This is a public check that can be used to assert that other tests will run successfully on your solution. This step needs to run without modification
• "further tests" will be used to prove that the solution works correctly. These are not visible to you as a candidate and will be run once you submit the solution

Scoring

There is a number of scoring systems being run on your solution after it is submitted. It is ok if these do not pass, but they add information for the reviewers.

API

To simplify the challenge and remove language restrictions, this service must provide a REST API which will be used to interact with it.

This API must comply with the following contract:

GET /status

Indicate the service has started up correctly and is ready to accept requests.

Responses:

• 200 OK When the service is ready to receive requests.

PUT /cars

Load the list of available cars in the service and remove all previous data (reset the application state). This method may be called more than once during the life cycle of the service.

Body required The list of cars to load.

Content Type application/json

Sample:

[
  {
    "id": 1,
    "seats": 4
  },
  {
    "id": 2,
    "seats": 6
  }
]

Responses:

• 200 OK When the list is registered correctly.
• 400 Bad Request When there is a failure in the request format, expected headers, or the payload can't be unmarshalled.

POST /journey

A group of people requests to perform a journey.

Body required The group of people that wants to perform the journey

Content Type application/json

Sample:

{
  "id": 1,
  "people": 4
}

Responses:

• 200 OK or 202 Accepted When the group is registered correctly
• 400 Bad Request When there is a failure in the request format or the payload can't be unmarshalled.

POST /dropoff

A group of people requests to be dropped off. Whether they traveled or not.

Body required A form with the group ID, such that ID=X

Content Type application/x-www-form-urlencoded

Responses:

• 200 OK or 204 No Content When the group is unregistered correctly.
• 404 Not Found When the group is not to be found.
• 400 Bad Request When there is a failure in the request format or the payload can't be unmarshalled.

POST /locate

Given a group ID such that ID=X, return the car the group is traveling with, or no car if they are still waiting to be served.

Body required A url encoded form with the group ID such that ID=X

Content Type application/x-www-form-urlencoded

Accept application/json

Responses:

• 200 OK With the car as the payload when the group is assigned to a car. See below for the expected car representation

  {
    "id": 1,
    "seats": 4
  }

• 204 No Content When the group is waiting to be assigned to a car.
• 404 Not Found When the group is not to be found.
• 400 Bad Request When there is a failure in the request format or the payload can't be unmarshalled.

Tooling

At Cabify, we use Gitlab and Gitlab CI for our backend development work. In this repo you may find a .gitlab-ci.yml file which contains some tooling that would simplify the setup and testing of the deliverable. This testing can be enabled by simply uncommenting the final acceptance stage. Note that the image build should be reproducible within the CI environment.

Additionally, you will find a basic Dockerfile which you could use a baseline, be sure to modify it as much as needed, but keep the exposed port as is to simplify the testing.

⚠️ Avoid dependencies and tools that would require changes to the acceptance step of .gitlab-ci.yml, such as docker-compose

⚠️ The challenge needs to be self-contained so we can evaluate it. If the language you are using has limitations that block you from solving this challenge without using a database, please document your reasoning in the readme and use an embedded one such as sqlite.

You are free to use whatever programming language you deem is best to solve the problem but please bear in mind we want to see your best!

You can ignore the Gitlab warning "Cabify Challenge has exceeded its pipeline minutes quota," it will not affect your test or the ability to run pipelines on Gitlab.

Requirements

• The service should be as efficient as possible. It should be able to work reasonably well with at least $10^4$ / $10^5$ cars / waiting groups. Explain how you did achieve this requirement.
• You are free to modify the repository as much as necessary to include or remove dependencies, subject to tooling limitations above.
• Document your decisions using MRs or in this very README adding sections to it, the same way you would be generating documentation for any other deliverable. We want to see how you operate in a quasi real work environment.

Feedback

In Cabify, we really appreciate your interest and your time. We are highly interested on improving our Challenge and the way we evaluate our candidates. Hence, we would like to beg five more minutes of your time to fill the following survey:

• https://forms.gle/EzPeURspTCLG1q9T7

Your participation is really important. Thanks for your contribution!

Car Pooling Challenge Explanation

Framework and languages

• NodeJS / TypeScript
• Express
• Jest and supertest (testing)
• K6 (performance testing)

Methodology and architecture

The whole challenge was designed with clean architecture pattern and TDD methodology following Red-Green-Refactoring approach. Each feature was developed starting with a new git branch named feature/nameOfFeature created from the develop branch and then merged with it when the feature was finished.

Process

I separated the API in 5 main layers:

• Controllers
• Usecases
• Repositories
• Data
• Domain (which contain Entities and Usecases)

Each of them, except domain, contain 3 "modules" :

• Cars - add, find cars
• journey - join, find or delete a group with a car
• waitingList - keep groups that can't join a car

The application is "Usecase" centric. All the requested features were developed by defining the usecase and writing his Unit Test with Jest. Then I created repositories to connect usecases to the datasources. Finally, Controllers were developed to be able to send and retrieve data trough a REST API.

When all the project was tested and completed, I performed a spike-test using k6.

Performance

Here is the 3 steps of the test:

• put 50.000 cars

Then virtual users:

• Start a journey
• Locate a group
• Drop Off a group

The test loop from 0 to 4000 virtuals user during 2min. Key results are:

• Total requests ~200.000
• Avg req duration (from request to response) ~20ms
• 99% of requests take <1s
• 100% 200 status response from the 3 endpoints

(You will find a screenshot of the test in the spike-test folder)

To be able to handle a lot of data, I saved all dataSource in Javascript Map object. Cars are grouped by free seats and Groups in waitingList by people. Journey is a Map with group ID as key and a Map containing CarEntity (key car) and GroupEntity (key group) Cars and Groups can be found directly by the number of needed seats. When a group start a journey, I search for the first car in the Map getted by the number of people in the group. When a group stop his journey, I get the Map(car,group) thanks to the ID of the dropedOff Group. With that information I can quickly get the Car from his DataMap and add the number of leaving people to his freeSeats then search in the waitingList to complete the car with 1 or more groups.

Launch the project

You can launch the project with

npm run dev

or 

npm run docker:build:dev || npm run docker:build:prod
npm run docker:run:dev || npm run docker:run:prod

To run the spike-test you need to install k6 k6-installation-link

Then

cd spike-test
k6 run script.js