DEV Community

zisam shaikh
zisam shaikh

Posted on

Benchmarking TypeScript KV & Cache Engines in 2026: KVDock vs. Unstorage vs. Keyv

Choosing a key-value or caching abstraction in TypeScript usually comes down to three questions:

  1. What happens when a hot cache key expires under 1,000 concurrent requests? (Does your database get slammed by a thundering herd?)
  2. Can I pair an in-process L1 memory cache with a distributed L2 store (like Redis or Cloudflare KV) without writing 200 lines of glue code?
  3. Does it work across Node.js, Bun, Deno, and Edge Workers with zero dependency bloat?

We ran a head-to-head architectural and benchmark comparison between KVDock, Unstorage, and Keyv.

🎮 Launch the Live Browser Stampede Simulator

⭐ Support open-source infrastructure: Star KVDock on GitHub ★


1. Architectural Feature Matrix

Capability kvdock unstorage keyv
Single-Flight Stampede Protection (getOrSet) 🟢 Built-in ❌ Manual ❌ No
Native Multi-Tier Caching (TieredDock L1+L2) 🟢 Built-in with auto-promotion 🟡 Overlay driver ❌ No
Zero Runtime Dependencies in Core 🟢 Yes (0 deps) 🟢 Yes ❌ No
Memory-Safe Streaming Scans (dock.scan()) 🟢 AsyncIterable cursor 🟡 Full array getKeys() 🟡 Iterator
Compare-And-Swap (cas) & Atomic Counters 🟢 Built-in (cas, increment) ❌ No ❌ No
Tag-Based Group Invalidation (invalidateTag) 🟢 Built-in ❌ No ❌ No
Next.js 15 App Router Incremental Cache 🟢 Official Handler 🟡 Community ❌ No
Distributed Locks with Fencing Tokens 🟢 Built-in (withLock) ❌ No ❌ No

2. The Thundering Herd Benchmark (1,000 Concurrent Misses)

When a cached key expires at the exact moment 1,000 requests hit your API, a standard get() + set() cache pattern causes all 1,000 callers to see null and query PostgreSQL simultaneously.

With kvdock.getOrSet(), concurrent requests for the same missing key attach to a single in-flight promise:

import { KVDock } from "kvdock";
import { MemoryAdapter } from "kvdock/adapters/memory";

const dock = new KVDock({ adapter: new MemoryAdapter() });

// 1,000 concurrent callers -> 1 database query, 999 coalesced waiters
const data = await dock.getOrSet("report:daily", async () => {
  return await db.query("SELECT * FROM heavy_aggregation");
}, { ttl: 60_000 });
Enter fullscreen mode Exit fullscreen mode

You can reproduce this benchmark locally in one command:

git clone https://github.com/zisam-glitch/kvdock.git
cd kvdock && npm install && npm run build
node scripts/benchmark-stampede.mjs
Enter fullscreen mode Exit fullscreen mode

Explore KVDock

If you appreciate zero-dependency TypeScript libraries with real benchmarks, please give zisam-glitch/kvdock a ★ Star on GitHub!


👤 About the Author

I'm Zisam Shaikh (@zisam-glitch), a systems & infrastructure engineer building zero-dependency caching and storage engines across TypeScript, Rust, Go, Swift, and Kotlin.

Top comments (1)

Collapse
 
alexshev profile image
Alex Shev •

Cache benchmarks become much more actionable when they separate steady-state throughput from operational behavior under failure. A short section on eviction, corrupted entries, and restart recovery would help teams choose a library for the workload they actually run, not only for a clean benchmark loop.