At thefullstack we are building our platform on Google Cloud Platform.  Our key considerations  are favouring GCP managed services to avoid operational overhead of running infrastructure.  The less time we spend managing and maintaining infra means more time understanding our users and iterating the product. 
Our second key consideration are cloud infrastructure running costs. As we are a small team bootstrapping a startup cost control is important. To this end we are favour serverless 'pay of you go' services and avail of monthly free tier limits 

[View Larger Image](https://jamiegood.files.wordpress.com/2022/03/tfs_garch-2022-03-07-1444.png)

What follows is an overview of the GCP services:

### Cloud Run
Cloud run is a fully managed serverless platform to deploy scalable containerised services. It's the next step on from App Engine. 
Elastic scaling is achieved by setting a scaling threshold based of concurrent requests.  You are charged for CPU processing time and memory usage.  It allow you to set a minimum number of instances per microservice to manage costs.When our traffic low we can set minimum instances to 0. We will only be charged for the processing of http requests.  Also non production environments (development, testing) means a dev infra costs practically nothing.  
In production env however how beneficial depends on the cold start performance and can be  language/framework specific. If the cold start delay is acceptable depends on your context.  We'd noticed 3-4 seconds for lightweight low mem/cpu spec Node.js microservices to start up on the first request. Typically the service will scale down to the min instances value after 10 - 15 mins depending on the requests load.
As you user base grows you'll set a minimum beyond 0 to run continuously incurring 24/7 charges.



### Cloud Functions
Cloud Function differs from Cloud Run in that it further abstractions away infra. You don't build containers such with docker. 
Instead you can deploy your code via Cloud Build or similar directly to the Cloud Function system. 
You are charged for each Function invocation - Pay as you Go. 
Cloud Functions are intended for handle lightweight workloads and can response to HTTP request or events. Deep integration with rest of the GCP platform means Cloud Functions can act as a glue or integration mediator between your own and GCP managed services.
We use Cloud Functions to provide simple HTTP endpoints, to handle pub/sub events and as a lightweight ETL pipeline component.

### Pub/Sub
Handles async communication between microservices and wider GCP system.

### API Gateway
API Gateway is a fully managed service for creating, securing, caching, rate limiting and monitoring apis into our microservices. Again providers deep integration with wider GCP ecosystem. We configure via standard OpenApi configs.

### Cloud Storage
Cloud storage for user uploaded content such as post images, avatar images. 


### Databases
MongoAtlas a GCP hosted 'database-as-a-service' platform. 
NoSQL schemaless design is particularly suited to early stage start up needs allowing for constant adjustments. As the startup goals solidify we can introduce schema. Mongo hosts and maintains the encrypted database cluster and manages automated backups. We considered GCP NoSQL databases as our primary database however decided on MongoDB considering the flexibility of the DB and the existing experience and skillset of the founding team.

Firestore is an NoSql managed database integrated with the Firebase suite of products.
It keeps your client data in sync through realtime listeners.  In thefullstack we use the realtime feature for client side notifications such as comments, likes, mentions etc is streamed via Pub/Sub into Firestore. 

### Firebase 

Cloud Firestore discussed under Database above. 

#### Firebase Cloud Messaging
FCM provides push notifications to browsers and native mobile apps. 

#### Firebase Performance Monitoring
Firebase Performance Monitoring is a service that helps us gain insight into the performance characteristics of our client apps. Such as networks requests, traces, first contentful paint etc

#### Firebase Authentication 
Firebase Authentication provides SDKs and backend services to handle app authentication. Integrates with the other Firebase and wider GCP system via Cloud Functions. 


### VPC

Serverless VPC Access allows Cloud Run and Cloud Functions to access resources in a VPC network using private IPs.  VPC Network Peering allows connectivity between our VPC and our MongoAtlas. 


### Data & Analytics

Google Analytics and Firebase provides us with insight into our users and engagement. 
Further product specific KPIs such as metrics on content, posts, comment etc are surfaced in the data visualisation tool DataStudio
Cloud Scheduler is a cron style managed solution which trigger regular jobs.  We implement a lightweight ETL pipeline via Cloud Function reading, processing and storing data into BigQuery. 


### DevOps & CI/CD

A code commit kicks off a pipeline to production.  Cloud Build handles the build, testing and deployment to Cloud Run. Container images stored on Container Registry. Canary releases can be achieved by Cloud Build instructing Cloud Run to split traffic across new and existing service revision.  Secret Manager stores app secrets which Cloud build accesses during building. 


### Operations & Monitoring

Cloud Logging and Monitoring automatically collects logs and op monitoring data from all the services.
Importantly all are deeply integrated with GCP products mentioned so far including the cloud run microservices. 
Monitoring does alerting to our Slack or phones and Dashboards. We set performance SLAs and budget threshold for cost management. 
Error Reporting provides us with deeper insight into application errors and
Cloud Trace is a distributed tracing service that gives us insight into latency data to surface performance degradation. 



### Conclusion

Thanks for making it this far! I hope you found this google cloud architecture overview interesting. I look forward to sharing further posts on our experience with GCP. 

The next articles I hope to post are: 
* An article of running costs and the Free Tier Usage
* An article on considerations for scalability, availability and reliability 

If you are interested in a particular topic let me know.

The diagram was created using the [GoogleCloud Architecture Diagramming Tool](https://googlecloudcheatsheet.withgoogle.com/architecture?diagram=SimpleComputeApp)

At thefullstack we are building our platform on Google Cloud Platform.  Our key considerations  are favouring GCP managed services to avoid operational overhead of running infrastructure.  The less time we spend managing and maintaining infra means more time understanding our users and iterating the product. 
Our second key consideration are cloud infrastructure running costs. As we are a small team bootstrapping a startup cost control is important. To this end we are favour serverless 'pay of y