Monitoring Redis

Introduction

Redis exposes 100+ INFO fields — focus on used_memory, evicted_keys, instantaneous_ops_per_sec, connected_clients, master_repl_offset lag, and latest_fork_usec. Export via redis_exporter to Prometheus; alert before users notice.

Dashboards: memory trend, hit rate (app metric), p99 command latency, replication lag, evictions per minute. SLOWLOG length growth correlates with deploy regressions.

Managed Redis adds cloud metrics — still validate application-side cache hit rate Redis alone cannot see.

Understanding the topic

Key concepts

• INFO memory/stats/replication/clients.
• redis_exporter → Prometheus → Grafana.
• SLOWLOG length and entries.
• Latency histogram LATENCY LATEST.
• Alert thresholds: memory 80%, evictions spike.
• Application cache hit rate complement.

Step-by-step explanation

1. Exporter scrapes INFO periodically.
2. Time-series stored in Prometheus.
3. Alertmanager fires on threshold breach.
4. On-call runbook links to redis-cli checks.
5. Post-incident review updates dashboards.

Syntax reference

Common commands

• instantaneous_ops_per_sec — traffic spike detection.
• mem_fragmentation_ratio — defrag trigger.
• master_link_down_since_seconds — repl broken.

INFO stats
INFO memory
INFO replication
SLOWLOG LEN
LATENCY LATEST

Informative example

Key metrics scrape script for cron until exporter deployed:

#!/bin/bash
H=$(redis-cli INFO memory | awk -F: '/used_memory_human/{print $2}' | tr -d '\r')
E=$(redis-cli INFO stats | awk -F: '/evicted_keys/{print $2}' | tr -d '\r')
OPS=$(redis-cli INFO stats | awk -F: '/instantaneous_ops/{print $2}' | tr -d '\r')
echo "memory=$H evicted=$E ops=$OPS" | logger -t redis-metrics

Replace with redis_exporter Helm chart in Kubernetes. Tag dashboards by environment and cluster name.

Real-world use

Real-world use cases

• On-call alert memory >80%.
• Detect replication lag during network event.
• Post-deploy slowlog regression.
• Capacity plan from ops/sec trend.
• SLA reporting for platform team.

Best practices

• Redis exporter on every instance.
• Alert evicted_keys rate not just memory.
• Track app cache hit rate alongside.
• Log correlation id on cache miss spikes.
• Weekly review LATENCY DOCTOR in staging.
• Runbook links from alert to redis-cli steps.

Common mistakes

• Monitoring only CPU not memory.
• No alert on replication disconnect.
• Ignoring evictions as normal.
• Dashboard without app hit rate context.

Advanced interview questions

Summary

INFO memory/stats/replication — core scrape set.