"Out of memory" entries from the memcached server: failed eviction or misconfigured memory limit?

The memcached server writes an "Out of memory" entry when its allocation path cannot secure memory for an incoming item. The goal is to determine what this documented failure condition reveals about a deployment before any configuration is changed.

The uncertainty lies in interpretation. Memcached reclaims space through per-slab-class LRU eviction rather than a single global policy, so one size class can lack a reclaimable item even when the server has not exhausted every page. It is unclear whether a logged "Out of memory" entry means an eviction was attempted and failed, or whether the server was started with eviction disabled so that errors are returned instead. Allocation behavior also differs across memcached releases, and other memory-related messages can point to different causes, so the exact phrase and the deployed version both matter for diagnosis.

• Under what exact conditions does the memcached server emit an "Out of memory" entry, and does the entry distinguish a failed eviction from a configuration that forbids eviction?
• What log evidence beyond the entry itself is needed to decide whether the memory limit, the item size distribution, or the eviction settings are responsible?