System Design: The Sharded Counters

A popular post can concentrate updates on one counter even when the rest of a service is quiet. This chapter spreads that write load across partial counters, then examines the extra work needed to read and preserve their total.

Problem statement

Views, likes, and comments can create a skewed update workload: a few items receive a large fraction of events. These frequent items are heavy hitters. Their counters become hot keys when the required update rate exceeds one key’s serving capacity. The bottleneck depends on per-item traffic, not merely the total number of counters.

The following historical media examples illustrate concentrated attention. Use their counts to understand skew, not as current traffic measurements or a capacity benchmark: