Taming the BPF LRU - Eliminating TCP Resets in Cilium

Explore a critical production issue in this lightning talk where Cilium users experienced mysterious TCP connection failures due to unexpected RST packets at scale. Dive deep into the root cause analysis of a bug in Cilium's BPF-based SNAT implementation and its LRU eviction policy that prematurely terminated active sessions. Learn how the eBPF datapath contributed to this problem and discover the elegant upstream fix merged in Pull Request #37747, which proactively restores original NAT entries on reverse traffic paths. Understand how this real-world solution reduced connection failures from up to 10% to nearly zero in production environments. Gain valuable insights for operators debugging network instability and developers working with eBPF challenges, including clear diagnostic techniques for identifying this "silent killer" and key principles for building robust, high-performance cloud networking solutions.