Back-end Monolith of PokerGloss.com. Front-end repo
TL;DR Microservices are great if you have money.
I love Docker, Kafka and all that cloud stuff, just don't have enough money:)
Each folder in this repository used be either a library or a microservice having it's own
Grpc API, Rest API, CI/CD and Kubernetes config. Everything was deployed to GKE.
I shut down the project in 2021 because I was low on funds and burned out.
I always wanted to keep it online but the cost of GKE would prevent that.
In 2024 I felt positive enough to fix it. My goal was to run it as cheap as possible (under 10$ a month).
That means no Kubernetes, Docker, gRPC, Message Queues and the whole Microservice Architecture.
I decided to merge all microservices into a monolith and deploy it as a single Go binary to a VPS instance.
The original project's name was PokerBlow. I knew it was a bit cheeky, but then too many people pointed out that it sounds sexual. After the migration from the cloud, I renamed it to PokerGloss as pokergloss.com was available and it sounds good to me!
- Thanks to
api/{service_name}/**pattern it was easy to combine all the routes in one HTTP server. - gRPC was replaced with simple function calls. I'm glad I decided to create gRPC only for internal calls, instead of having the easy gRPC-HTTP transcoding way.
- Pub/Sub was replaced with simple Go channels, (I used this to mock it). I'm glad GCP didn't offer Kafka as a Service because it would be harder to get rid of. However, table timeouts needed to be persistent and I kept Pub/Sub just for them.
- Each service would still have its own database, because it creates no additional costs.
- I deleted all the k8s yaml files, because now the result product of the whole back-end is just a binary running behind nginx on a small VPS.
- The microservices used to run on Golang 1.15. Some of the old packages were no longer available e.g.
google.golang.org/grpcSo I had to reinstall packages, updating Go to 1.23. - I had a nice CI/CD pipeline test->build->deploy->build_client-publish-client for each service. But remembering how much pain it was to configure dockertest for GitLab (the last lines of dockertest readme are mine), I decided to stick with local bash scripts and run tests on my M3 macbook, which is a lot faster than any ci runner. Maybe later I'll add the GitHub CI/CD.
- To replace Kubernetes' features I created Yetis
I have used the MongoDB license because I don't want anyone using this commercially without my permission. If you'd like to use something from this project commercially please email me at [email protected]