Everyone Expires at Once

After each deploy, a warmup job loads roughly 400,000 product entries into the cache in a few minutes, each with a flat 3600 second TTL. Exactly one hour after every deploy, the database sees a one minute burst of read traffic that is 20 times normal. What is the standard fix?

Raise the maxmemory limit so the warmup never forces an eviction and all 400,000 product entries stay resident for their whole lifetime

Randomize each entry's TTL within a band, for example a uniform draw between 3000 and 4200 seconds, so the expiries scatter

Raise the flat TTL from 3600 to 86400 seconds so the warmed entries outlive several deploy cycles before they ever need refilling

Switch the maxmemory policy from allkeys-lru to allkeys-lfu, so the entries with the most reads are the last ones the cache evicts