Why Redis Becomes Slow and How to Optimize Its Performance

This article analyzes why Redis may become slow and offers a comprehensive set of diagnostics and optimizations.

1. Intrinsic latency : Use ./redis-cli --intrinsic-latency 120 to measure the maximum latency within 60 seconds; values above the baseline (e.g., >2 ms on low‑end hardware) indicate slowdown.

2. High‑complexity commands : Check the slowlog and avoid O(N) or higher commands such as SORT , SUNION , ZUNIONSTORE , or large N values, as they increase CPU usage.

3. Big keys : Scan for big keys with redis-cli -h 127.0.0.1 -p 6379 --bigkeys -i 1 and limit their size; large keys also worsen fork and copy‑on‑write overhead.

4. Expiration spikes : Randomize expiration times to avoid mass key deletions at the same moment, and consider enabling lazy‑free expiration (e.g., lazyfree-lazy-expire yes ).

5. Memory limit reached : When maxmemory is hit, Redis evicts keys according to the configured policy (e.g., allkeys-lru , volatile-lru ); choose a policy that suits your workload.

6. Fork overhead : Forking for RDB/AOF rewrite copies the process memory; large instances cause long pauses. Monitor latest_fork_usec via INFO and keep instance size below ~10 GB.

7. Huge pages : If transparent huge pages are enabled ( cat /sys/kernel/mm/transparent_hugepage/enabled shows [always] ), disable them to avoid 2 MB copy overhead.

8. AOF configuration : Choose an appropriate appendfsync policy; appendfsync everysec balances safety and performance, but watch for disk I/O contention during AOF rewrite.

9. CPU binding : Bind Redis to multiple logical cores of the same physical CPU, or use Redis 6.0's server_cpulist , bio_cpulist , aof_rewrite_cpulist , etc., to reduce contention.

10. Swap usage : Detect swap with cat /proc/ $pid /smaps | egrep '^(Swap|Size)' ; avoid swap by adding memory or scaling out.

11. Memory fragmentation : Check mem_fragmentation_ratio via INFO memory ; if >1.5, consider restarting (Redis <4.0) or enabling activedefrag (Redis ≥4.0).

Finally, the article provides a checklist of optimization actions, such as limiting O(N) commands, using lazy‑free deletion, adjusting eviction policies, controlling instance size, disabling huge pages, tuning AOF, and monitoring key metrics like expired_keys and latest_fork_usec to maintain low latency.