// problem · 12 march 2026 · 1 min

Redis data appears stale

Redis was doing exactly what I told it to do. I had told it to lie for a while.

Time wasted: 2 hours

Problem

Periodic spikes in database load, or a UI showing a value that had already been updated. “We use Redis” was true and unhelpful.

Laravel Redis PHP-FPM MySQL

Homepage TTFB spiked every N minutes, or a setting change took until TTL to appear. MySQL showed the same expensive query many times in parallel.

The next request after a write should see the write. A hot page should not stampede MySQL when a key expires.

Inspected the key, TTL, and the remember() callback. Checked whether the write path invalidated anything.

Flushed Redis (it “fixed” it). Increased TTL (it hid it). Blamed the database.

Cache-aside without invalidation on write, and/or a hot key expiry with no lock around rebuild. Redis stored what it was given.

Invalidate on write. Lock around expensive rebuilds. Precompute the hot key from a job if the request path should never rebuild.

Redis is not psychic. TTL is not a strategy. A stampede is many workers believing they should recompute at once.

If you cannot write the invalidation in the same change as remember(), the cache is not finished.

Found something wrong?

These notes can be incomplete, or wrong in a different environment. Suggest a correction on GitHub or email me at me@t4mer.net.

// article

How I think about Redis caching in Laravel applications

Cache aside, stampedes, and the difference between faster and wrong.

// til

TIL: Redis isn't just a cache

The same process can hold a cache, a queue, a lock, and a session. Those are different contracts.

// problem

HTTP 421 Misdirected Request caused by an incorrect Host header

A 421 that looked like Nginx until the outbound Host header showed up.