Memory Capacity Planning for Growing Workload

warningCapacity Planning

Proactively monitoring memory usage trends to determine when to scale Redis before hitting maxmemory limits and triggering evictions or OOM conditions.

Prompt: “Our Redis memory usage has grown from 60% to 82% of maxmemory over the past month and the trend is continuing. I want to understand if this growth rate means we'll hit our limit soon and start evicting keys. Should I scale up the instance now or can I wait? What's the safe memory threshold to stay under?”

Agent Playbook

When an agent encounters this scenario, Schema provides these diagnostic steps automatically.

When planning Redis memory capacity, start by checking whether evictions have already begun, then assess memory fragmentation impact which can accelerate hitting limits. Calculate your runway to the critical 90% threshold based on growth trends, and review historical peak usage patterns to identify recurring spikes. The goal is to scale proactively before sustained operation above 85-90% of maxmemory where eviction pressure becomes severe.

1Check if key evictions have already started

First thing I check is `redis.keys.evicted` to see if Redis is already evicting keys to stay within limits. Even a low but increasing eviction rate is a red flag. The `high-eviction-rate-indicates-memory-pressure` insight shows that when `redis.memory.used` exceeds 90% of `redis.memory.maxmemory`, evictions accelerate rapidly and cache hit rates decline. If you're seeing evictions at 82% usage, you likely have fragmentation issues or maxmemory set too aggressively.

High Eviction Rate Indicates Memory Pressure redis.keys.evictedredis.memory.usedredis.memory.maxmemory

2Assess memory fragmentation impact on actual usage

Check `redis.memory.fragmentation_ratio` — if it's above 1.5, your actual memory footprint is significantly larger than logical used memory, and you could hit physical limits before `redis.memory.used` reaches `redis.memory.maxmemory`. The `memory-fragmentation-amplifies-cache-eviction-pressure` insight explains how high fragmentation triggers premature evictions. At 82% logical usage with a 1.5x fragmentation ratio, you're actually consuming closer to 123% of reported usage, putting you at immediate risk.

Memory Fragmentation Amplifies Cache Eviction Pressure redis.memory.fragmentation_ratioredis.memory.usedredis.memory.maxmemory

3Calculate your runway to the 90% critical threshold

You've grown from 60% to 82% in a month — that's 22 percentage points, or roughly 0.73% per day. At this rate you'll hit 90% in about 11 days and 100% in 25 days, though growth often accelerates as workloads scale. Track daily `redis.memory.used` trends to detect acceleration. In my experience, you want to scale before hitting 85-90% of `redis.memory.maxmemory` to maintain headroom for traffic spikes and avoid the eviction pressure zone.

redis.memory.usedredis.memory.maxmemory

4Review historical peak memory patterns for spike risk

Compare `redis.memory.peak` to current `redis.memory.used`. If peak is significantly higher (>30% per the `peak-memory-exceeded-indicates-historical-capacity-issues` insight), you've experienced memory spikes that could recur. Look for daily or weekly patterns in peak usage — if these happen during batch jobs or traffic spikes, your current 82% could spike to 100%+ during these events, triggering immediate evictions even if average usage looks manageable.

Peak Memory Exceeded Indicates Historical Capacity Issues redis.memory.peakredis.memory.usedredis.memory.maxmemory

5Verify eviction policy alignment with workload expectations

Review your maxmemory-policy configuration to ensure it matches your use case. With noeviction, you'll start getting write errors at maxmemory. With allkeys-lru, Redis evicts any key to make room. The `high-eviction-rate-indicates-memory-pressure` insight notes that mismatched policies cause hot data evictions and cache miss storms. Also check if keys have appropriate TTLs — unnecessarily long-lived keys consume memory that could be reclaimed naturally without scaling.

High Eviction Rate Indicates Memory Pressure redis.memory.maxmemoryredis.db.keys