NOTE

3.6 Redis Hot Keys

How one disproportionately popular key overloads a Redis shard, how to detect it, and strategies such as local caches, replica reads, key replication, request coalescing, and data-model changes.

Redis / CacheCreated Updated 1 min readhistorical

This is a historical learning note and may contain outdated or incomplete understanding.

1. What Is a Hot Key?

A hot key receives a disproportionate share of reads or writes relative to the rest of the dataset.

There is no universal threshold such as “10,000 QPS.” A key is hot when its traffic becomes a bottleneck for the owning shard, network path, client pool, or downstream dependency.

2. Why Sharding Does Not Automatically Fix It

A single key hashes to one slot/shard. Adding more cluster nodes distributes other keys but does not split operations on that one key.

3. Detecting Hot Keys

Useful signals include:

  • application per-key metrics/sampling;
  • proxy metrics;
  • managed Redis hot-key analytics;
  • CPU/network asymmetry between shards;
  • command latency and access-frequency tooling.

Avoid full-keyspace instrumentation if its overhead exceeds the problem you are measuring.

4. Read-Hot Keys

Options include:

  • local in-process cache with bounded TTL;
  • CDN/edge cache for HTTP-visible data;
  • duplicate/sharded cache keys when stale replication is acceptable;
  • replica reads when the consistency model allows them;
  • request coalescing/singleflight so many misses cause one backend fetch.

5. Write-Hot Keys

Write hotspots are harder because duplicating the key creates a consistency problem.

Possible redesigns:

  • shard counters and aggregate later;
  • append events rather than rewriting one aggregate;
  • batch writes;
  • partition by user/object/time bucket;
  • move the invariant to a system designed for the required write concurrency.

6. Key Principle

A hot key is usually a data-model hotspot, not merely a Redis tuning issue. The strongest fix changes how traffic/state is partitioned.

Loading helpful count