What happens when Redis runs out of memory, and how would you design for it?

In production I would monitor memory usage, fragmentation, evicted keys, expired keys, rejected writes, key count, hit rate, and source load after evictions. I would avoid using one Redis database for unrelated workloads with different eviction tolerance, such as disposable product cache beside session state. If eviction rises, I would check which key patterns and payload sizes consume memory before changing capacity. The better fix may be key discipline, shorter TTLs, smaller values, splitting workloads, or capacity changes, not only raising the limit.