How Zalando improved offer creation throughput by x100 using Akka Sharded Cluster
summary
Zalando replaces a lock-heavy Postgres-based offer assembler with an actor-per-offer system using Akka Cluster Sharding, moving state into memory to eliminate DB locks, add per-node DB batching, and apply backpressure from a cluster singleton. The deployment hits ~100,000 events per second and >4 GiB/s (20x throughput), fixes GC pauses by switching to the G1 garbage collector, and stabilizes processing under real CyberWeek loads.
AI-written summary · about
chapters (12)
- 0:00Start
- 1:31Throughput
- 2:55Product Offer Pipeline
- 6:20Race Condition
- 6:45Pessimistic lock
- 7:48Akka Actors in Action
- 9:31Akka Cluster Sharding
- 13:35Stateful Sets
- 16:19Contact Point Discovery
- 21:05Failure Detector
- 22:22Garbage Collector - CMS to G1
- 25:00Sharding Strategy
related talks
25mGo distributed (and scale out) with Actors and Akka Clustering
47mApplying “web scale” patterns (...)
18m"Kafka's Metamorphosis"
27mFour Principles of Reactive Architectures
28mAkka - Enterprise Agentic AI with Tyler Jewel
41mHands on Introduction to Distributed Systems Concepts with Akka Clustering
37mReactive Architecture on AWS: from 0 to 1.6M events/minute
27mBUILDING A REAL-TIME AUCTION ENGINE USING EVENT SOURCING