This session covers how Attentive scaled mutual exclusivity in its event-driven architecture by evolving from Redis-based locks to Apache Pulsar’s FAILOVER and KEY_SHARED subscription modes. Operating at the scale of 620M texts/day during peak events, we share how we addressed the “split-brain” effect, improved the latency of event consumption, and navigated the trade-offs across subscription models. We’ll also dive into key lessons learned in building a reliable, low-latency, high-throughput message orchestration system.