<antirez>
8 hours ago
- Redis persistence uses fork(), which causes latency spikes on Xen-based virtual servers like some EC2 instances.
- Stripe's latency graphs show delays up to one second during fork events, affecting the 99th percentile.
- Latency spikes every 30 minutes impact very few page views (about 0.05%), unlike evenly distributed spikes.
- For quality of service, the distribution of latency spikes matters more than average latency metrics.
- Redis is optimizing for EC2 with specific documentation and safer master-slave replication with disabled persistence.
- A quick fix for EC2 with disabled persistence is to disable automatic restarts and use Sentinel for failover.