As our game scales from thousands to millions of players, our real-time metrics infrastructure needs to evolve beyond simple aggregation. We're exploring event streaming architectures using Kafka, Kinesis, or Pulsar to handle massive throughput while maintaining low latency. The challenge is designing a system that can scale horizontally, handle backpressure gracefully, ensure exactly-once processing semantics, and recover from failures without losing critical metrics. We're also considering lambda architecture versus kappa architecture - should we maintain separate batch and streaming pipelines, or unify everything into a single streaming model? Partitioning strategies, consumer group management, and state management in stream processing are all critical decisions. What architectural patterns have you successfully implemented for scaling real-time game metrics? How do you handle hot partitions when certain game servers or events generate disproportionate traffic? Any lessons learned from production incidents? Would love to hear about your journey scaling from prototype to production-grade real-time systems!