Intro to Stateful Services

on YouTube ↗43m

" summary

Describes an architecture for low-latency, stateful services for a betting workload where stateless designs became too slow, unreliable and costly for 30 KB payloads at ~2000 updates per-second per tenant and end-to-end latencies spiking above one second when moving off single-node memory. The proposed approach uses Kafka change-streams to synchronize partitioned, in-memory stateful nodes keyed by tenant and evaluates tooling options (Pulsar, Faster, DuckDB and a custom stereoDB), with microbenchmarks reporting stereoDB handling ~3 million writes and millions of reads under optimized conditions.

AI-written summary · about

" chapters (11)

1. 0:00Intro to Stateful services
2. 4:41Traditional architecture problems
3. 8:13Stateful concept and architecture
4. 11:07Stateful performance benefits
5. 17:17Over-fetching and optimization
6. 22:22CPU Wait impact on performance
7. 25:52Stateful vs stateless reliability
8. 29:27Metal failures in high load
9. 30:52Tools for stateful systems
10. 34:04Stateful implementation experience
11. 35:48Q&A

" related talks

• 35mBuilding Scalable Stateful ServicesCaitie McCaffrey · Strange Loop 2015
• 37mBeyond Microservices: Streams, State & ScalabilityGwen Shapira · GOTO Chicago 2020
• 53mHow do we build high‑load microservicesAnton Moldovan · .NET Fest 2017
• 13mLow-Maintenance Backend Architectures for Scalable ApplicationsRinku Mohan · Conf42 Cloud Native 2025
• 59mBuilding Data Intensive Microservices with Apache KafkaYaniv Rodenski · NDC Sydney 2017
• 47mBuilding High Performance Real-time Stream Processing MicroservicesFawaz Ghali · GeeCON 2023
• 05mPREVIEW: Event-Driven Services with Stateful StreamsBen Stopford · Kafka Summit San Francisco 2017
• 48mIntegrating microservices by accepting the data dichotomyDamien Gasparina · Voxxed Days Microservices 2018