Skip to content

Latest commit

 

History

History
137 lines (98 loc) · 6.91 KB

README.md

File metadata and controls

137 lines (98 loc) · 6.91 KB
Raw

@deblasis' Space Traffic Control

E2E Integration Tests Go Report

This is an implementation of the corresponding Topcoder Challenge

TLDR;

make run-fast

then navigate to http://localhost:8081 and you will be redirected to the swagger UI From there you can play with the API

Side note: I used a playful tone, normally I keep it more professional 😉.

⚠ "Root" User Token

In order to get the pre-seeded Command user token, just run the following command, I would never put usernames and password in source control, that's why I am following best practices... 😂 not

curl -X 'POST' \
  'http://localhost:8081/auth/login' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "username": "deblasis",
  "password": "password!"
}'

Otherwise follow the instructions below for seeding your own user(s).

Architecture

The architecture of this implementation is as follows:

There is also a single Consul leader instance and two additional Agents offering DNS for APIGateway and ClessidraSvc.

Consul

All services report health status via Grpc to Consul, that, along with the design, enables horizontal scaling.

Services

Nodes

Gateway

All http requests requests are served by a front facing API Gateway that implements GRPC-Gateway. It proxies requests, converts them into GRPC and also takes care of handling authorization headers.

AuthZ / AuthN

I used JWTs RSA RS256 and I implemented handlers, interceptors and ACL myself instead of using out-of-the-box solutions. Token expiration is configurable via the jwt.tokenduration key in the AuthService's app.yaml hat is also overridable via environment variable DEBLASIS_JWT_TOKENDURATION supplied obviously to the AuthService container.

TLS

It's implemented but disabled by default to simplify testing/verification/debugging

Backend

In the backend, the services communicate with each other via GRPC but also expose HTTP endpoints for debugging purposes. Ideally, they should be disabled in a real-world production environment via a feature-flag of some sort.

Persistence

On the persistence side of things, there are specific microservices for the two databases in this implementation.

CentralCommandDBSvc has a peculiarity which is a feature-flagged Auxiliary GRPC endpoint that serves the purpose of exposing a method that's used to cleanup the database during integration tests.

Database

Nothing special, I used views and functions to fulfill the requirements and I leveraged the multi-model features provided by pg-go to have a very very simple CQRS

Seeding

There's a small binary that seeds the database using a CSV file and a staging table, when the staging table contains record, the AuthDBSvc picks up the records, hashes the passwords and deletes the table. This happens at service start, which implies that a restart has to be triggered, this is by design for security reasons.

In order to seed the database:

make seed-auth_dbsvc

The input file is located here.

There's also an initial user named deblasis added from the migrations with the same mechanism. This is only for convenience of testing/running the solution. Otherwise there would have been the need for an additional manual seeding step.

CI/CD via Github Actions seeds the database with the additional user in order to test the seeder itself.

Docker

The docker/docker compose setup takes advantage of the various features available such as caching, build images and parallel builds.

I applied layering here too, for example having a docker-compose.hostports.yml file used to expose ports in local environment, or a docker-compose.integrationtests.yml that configures the databases to be ephemeral and adds the container that executes E2E tests when it spins up.

Makefile

The makefile is used as a productivity enhancer, there are targets that are meant to be used only in local development, such as run-fast that builds the binaries on the host in parallel (way faster especially if using multi-core machines) and then copies them into the respective containers.

Certificates

They can be generated on the host or inside a container for a containerized build depending on the needs. There's also a target in the Makefile to copy the certificates to the host from the containerized builder for debugging purposes.

Worker

There's a worker service called ClessidraSvc (Clessidra is Italian for hourglass) that simulates passage of time, making sure that the ships that are supposed to leave, leave their docks and that the reserved docks are freed after a configurable timeout. The service registers itself with consul via the agent sidecar.

Prerequisites

  1. Docker
  2. Docker-Compose
  3. Go 1.11+

Running Tests

make run-fast-integrationtests

Spins up the whole stack and executes E2E tests in a dedicated container with logging enabled at the database level too.

For the same environment but to attach only the the integrationtester run (this is used in CI/CD checks):

make integrationtests-run

Local Deployment

make run-fast-integrationtests

Production Build and Installation

To spin up a version that doesn't expose container ports apart from the entrypoint (ApiGateway)

make run-fast

You can then navigate to http://localhost:8081 to the Swagger UI

Notes

  • I introduced the concept of reservation when a ship requests-landing because let's suppose that there's only one dock available, I don't want multiple ships to be offered the same dock. First come first serve.

  • When a ship has to land, in the requirements there's no mention of the dockId to land to. Since when we request landing we receive a dockId (from dockingStation in the response with command = land) I made it compulsory to have a land request that requires not only the time but also the dockId to land to.

  • I believe this was a fun challenge, I am sorry I didn't see much participation and submissions so far (we are 9d from submission end at the time of writing).

  • Because of the latter reason, I didn't spend too much time cleaning up the code, refactoring, etc. I might do that on Github after the competition is completed.