@hello.interview: Your cache just expired. 10,000 requests hit your database at once. That's a cache stampede, and most engineers only know one way to fix it. Cache locking. One request grabs a mutex and rebuilds the cache, everyone else waits in line. It works, but if that rebuild fails or takes too long, you've got thousands of requests timing out. Fragile. Request coalescing. Instead of queuing requests, you collapse them. Every duplicate in-flight request gets the same result from a single upstream call. Cloudflare does this at the CDN layer. Your backend goes from getting hammered by 10,000 requests to getting exactly one per app server. Probabilistic early refresh. Before the TTL expires, each request has a small random chance of triggering a background refresh. At minute 50 of a 60-minute TTL, maybe 1% of requests refresh it. At minute 59, maybe 20%. Refreshes spread out naturally. No thundering herd. Background refresh. A worker proactively recomputes hot keys before they expire. Best for predictable traffic, like a Ticketmaster event page where you know exactly what's going to get hammered. Start with request coalescing. It's the easiest win and works at the CDN or app layer. Add probabilistic early refresh for unpredictable spikes. Use background refresh only when you can predict what's hot. Skip mutex locking unless you're already on Redis and need a quick patch. Full caching breakdown linked in bio.

Hello Interview

Open In TikTok:

Region: US

Wednesday 06 May 2026 16:08:49 GMT