Real world microservices: Starting-up from scratch!
summary
Yolt starts in 2016 with Spring Boot services on Kubernetes, evolves to a 60+ service, event-driven platform using Kafka and Cassandra, and reaches 500,000 users. They replace blocking HTTP chains with Kafka events and a duplicated per-service user context to stop 200 TPS spikes on the user API, move heavy Python models to Scala (reducing instances from ~12 to 4), add ELK + trace IDs and Jaeger-style open tracing (found 19 Cassandra queries for one GET), and ship dashboards and team-level alerting.
AI-written summary · about
chapters (30)
- 0:00Introduction
- 7:122 Scale that we operate at
- 8:193. Innovative stack
- 10:12Our stack at the start
- 11:32This is hard!
- 13:57Fast forward...
- 14:57We can now answer these questions
- 16:04More insights needed
- 18:19HTTP challenges
- 20:34Hotspot users
- 22:07Time for action!
- 22:58Let's throw some events!
- 26:04User context object
- 27:48Back to our questions
- 28:24Growing the teams
- 29:53Generating architecture diagrams
- 30:40Scaling micro services with kubernetes
- 32:27Starting with microservices....
- 34:15Final question
- 34:30Monitoring, logging & alerting
- 34:48How we started
- 36:52Request tracing
- 37:09Kibana selector
- 38:01Kubernetes dashboard
- 38:58Grafana monitoring
- 40:43So we had logging, monitoring & alerting
- 41:37Let's improve
- 43:25Opentracing
- 44:51So how did we solve them?
- 45:42Wrap up and next steps
related talks
43mLife of a startup...
51mMicroservices and Microfrontends: Scaling, Learning, Growing
55mPatterns and practices for real-world event-driven microservices
24mMicroservices in Practice
18mMicroservices Out In the Wild
28mMicroservices – Boyan Dimitrov
23mLegacy, Open source, microservices and data architecture
01h 01mMicroservices Gone Wrong