A single post, product, or event page can suddenly draw a large share of all traffic.

The cache node holding that post, profile, comment thread, or counter can saturate while other nodes sit idle. The system may have 100 cache nodes, but the hot key is still owned by one shard unless you change how that key is served.

This is the hot key problem: traffic concentrates on one key or one partition so strongly that normal horizontal scaling stops helping.

Hot keys appear in viral posts, flash-sale products, breaking news articles, popular live streams, global counters, and high-volume queue partitions. The right fix depends on what is hot: a read-mostly object, a write-heavy counter, a scarce inventory record, or an ordered stream.

In interviews, do not jump straight to "add a cache." First identify the access pattern and correctness requirement. A viral article can tolerate stale reads. A like counter can tolerate approximate aggregation. A ticket inventory key cannot be treated like an approximate counter without risking oversell.