Redis "OOM command not allowed when used memory > maxmemory": choose what to do when full

1. Look at the numbers

redis-cli INFO memory | grep -E 'used_memory_human|used_memory_peak_human|maxmemory_human|maxmemory_policy|mem_fragmentation_ratio'

used_memory_human:2.00G
used_memory_peak_human:2.00G
maxmemory_human:2.00G
maxmemory_policy:noeviction
mem_fragmentation_ratio:1.08

noeviction is the default, and it’s what produces this error: when full, refuse writes. Now see what’s taking the space.

2. Choose a policy

redis-cli CONFIG SET maxmemory-policy allkeys-lru
redis-cli CONFIG REWRITE

CONFIG REWRITE keeps it across restarts. In Docker, or with a read-only config, put it in the redis-server command line or redis.conf instead.

One trap: with volatile-*, if no key has an expiry then nothing can be removed, and you get this same error. And if Redis holds queues or job data, such as with Sidekiq or BullMQ, don’t use allkeys-*, as it would silently delete your jobs. Those need noeviction, and more memory or fewer jobs.

3. Find what’s using the memory

redis-cli --bigkeys
redis-cli --memkeys

Both scan safely, without blocking the server. For one key:

redis-cli MEMORY USAGE session:12345
redis-cli TYPE session:12345
redis-cli TTL session:12345

TTL returns -1 if a key never expires, the most common cause of slow growth. To count keys by prefix, sample with SCAN, and never run KEYS * on a busy server, which blocks everything until it finishes:

redis-cli --scan --pattern 'session:*' | wc -l

4. Set expiry times

Code that writes without a TTL makes Redis grow until it’s full. Set one when you write:

SET session:12345 "data" EX 3600

await redis.set(`session:${id}`, data, { EX: 3600 });

To clean up old keys in batches, without blocking:

redis-cli --scan --pattern 'tmp:*' | xargs -r -n 500 redis-cli UNLINK

UNLINK frees the memory in the background, where DEL on a big key can stall the server.

5. Raise the limit, carefully

redis-cli CONFIG SET maxmemory 4gb
redis-cli CONFIG REWRITE

Check first what the machine has, with free -h, and leave room. Saving to disk forks the process, and on a busy server, it can use a lot of extra memory for a time. A limit of maxmemory too close to the machine’s total invites the OOM killer, which ends Redis itself. In a container, keep it below the memory limit too, or you get an exit code 137.

6. Memory used, but not by your data

If used_memory is low, and the process is large, look at mem_fragmentation_ratio. Well above 1.5 means fragmentation:

redis-cli CONFIG SET activedefrag yes

Output buffers may also take a lot: replicas falling behind, or clients that subscribe and don’t read. CLIENT LIST shows each client’s omem, and client-output-buffer-limit sets the limits.

If you see this on a replica, it follows its primary’s data, so the primary is what needs the attention.