Case Studies Manufacturing
Global power tools manufacturer
System Twin — IoT
Architecture remediation for a manufacturing-execution streaming pipeline—lifecycle enrichment across global plants. Error rate from >5% to <0.5%, e2e latency from 120 minutes to <1 second, zero incidents over an operating year.
Apache Kafka · Kafka Streams · Apache Flink · OpenTelemetry · RocksDB · Kubernetes
-
Customer
A global power tools manufacturer running a streaming pipeline in manufacturing execution—enriching parts as they are produced across global plants and tracking lifecycle end to end.
-
Challenge
Major performance issues on the manufacturing-execution streaming path.
A large Kafka Streams topology owned enrichment and lifecycle tracking. Error rates and consistency failures showed up as missed join windows, state-store explosion, and poor data quality. The platform could not hold latency or correctness under plant-scale load.
-
Work
Owned architecture remediation. Vastly simplified the Kafka Streams topology and moved fault-prone segments to Flink for consistent snapshots and faster recovery. RocksDB and related state-store tuning. Institutionalized data-quality and freshness metrics. Scaled Flink (Flink operator) and Kafka Streams with partitioning so jobs could run in parallel safely. OpenTelemetry for observability across the path.
-
Impact
- Error rate down to <0.5% (from >5%)
- End-to-end latency from 120 minutes to <1 second
- Self-scaling / self-healing architecture with lower cost to operate
- Zero incidents over an entire operating year
