Redis latency spikes during traffic bursts

Redis latency spikes during traffic bursts are usually caused by one or more of these:

Common causes

• CPU saturation: Redis is single-threaded for command execution, so bursts can queue requests.
• Slow commands: KEYS, large LRANGE, big SMEMBERS, SORT, Lua scripts, etc.
• Large payloads: big values increase network and serialization time.
• Memory pressure / swapping: if the OS swaps Redis, latency jumps badly.
• Fork-related pauses: BGSAVE, BGREWRITEAOF, or replication can cause copy-on-write overhead.
• Connection storms: many clients reconnecting at once.
• Persistence overhead: AOF fsync settings can add spikes.
• Evictions / expiration cycles: too many key expirations or eviction activity during bursts.
• Network bottlenecks: NIC saturation, packet loss, or cross-zone latency.

What to check first

1. Redis slowlog
   • SLOWLOG GET
2. Latency doctor
   • redis-cli --latency
   • redis-cli LATENCY DOCTOR
3. Instantaneous ops and CPU
   • INFO commandstats
   • INFO stats
   • INFO cpu
4. Blocked / slow clients
   • CLIENT LIST
5. Persistence / fork events
   • INFO persistence
6. OS metrics
   • CPU, memory, swap, disk I/O, network, context switches

Typical fixes

• Replace expensive commands with more efficient patterns.
• Break large values into smaller ones.
• Add client-side backpressure and connection pooling.
• Use pipelining to reduce round trips.
• Avoid big bursts with rate limiting or queueing.
• Disable or tune persistence if acceptable:
  • appendfsync everysec instead of always
  • avoid unnecessary BGSAVE
• Ensure Redis has:
  • no swap
  • enough RAM
  • CPU headroom
• Scale out:
  • read replicas for read-heavy traffic
  • sharding/cluster for write-heavy or high-QPS workloads
• If using Lua, keep scripts short and deterministic.

Quick diagnostic command set

redis-cli LATENCY DOCTOR
redis-cli SLOWLOG GET 20
redis-cli INFO cpu
redis-cli INFO stats
redis-cli INFO persistence
redis-cli CLIENT LIST | head

If you want, I can help you diagnose your specific spike by looking at:

• LATENCY DOCTOR
• SLOWLOG GET
• INFO persistence
• INFO stats
• your Redis version and deployment setup