The upgrade to Go 1.24 initially introduced a memory regression that increased physical memory usage across Datadog's services, prompting a collaboration with the Go community to identify and fix the issue. Surprisingly, the new update led to a significant decrease in memory usage in high-traffic environments, driven by the new Swiss Tables implementation that improved memory efficiency for large in-memory maps like the shardRoutingCache. By examining live heap profiles, it was discovered that Swiss Tables allowed for faster probing and eliminated the need for overflow buckets, yielding substantial memory savings. Further reductions in memory usage were achieved by optimizing the data structure, particularly by removing unused fields and using smaller data types. These changes not only mitigated the initial regression but also enhanced overall memory efficiency, demonstrating the importance of detailed runtime metrics and the potential for small optimizations to yield significant performance improvements.