Cloudflare Workers KV shifts edge state management from Redis and

Edge execution and performance

Cloudflare Workers KV provides low-latency access to globally distributed data across 300+ locations. This architecture places code within milliseconds of every user. Traditional centralized models force users in Tokyo to wait for round-trips to Virginia because light travels at a finite speed. Because Cloudflare Workers uses V8 isolates instead of containers, request starts happen in under 1ms, which eliminates the large 100-1000ms cold start delays that developers see in traditional AWS Lambda environments.

KV is eventually consistent.

Developers use KV for read-heavy workloads where a single key changes less than once per second. For these tasks, KV provides under 10ms p99 read latency. Workers provides 128MB of memory for all plans. Paid plans provide 30 seconds of CPU time, while free plans limit users to 50ms per request. Wrangler provides local development through accurate emulation of KV, D1, and Durable Objects. It allows one-command deployment to production. KV handles feature flags and configuration. It manages redirect maps. It serves static assets. Cloudflare allows 1,000 namespaces per account. KV is fast.

Data limits and consistency

Technical limits define the boundary between KV and Redis. KV allows only one write per second to the same key on every plan. This restriction prevents using KV for counters, live leaderboards, or session state that updates on every request. Upstash Redis lacks this per-key write limit.

KV uses a hybrid push/pull replication model. Writes go to a central tier and reads pull data into a regional cache. This caching makes KV eventually consistent. Changes take up to 60 seconds to propagate globally. Negative lookups also cache the absence of a key, meaning a newly created key might not appear in other regions for up to 60 seconds. For storage above the first GB, Cloudflare charges $0.50 per GB-month. Upstash does not charge for bandwidth on pay-as-you-go plans. You already know that this lag breaks session management.

KV lacks atomicity.

Will propagation delays eventually render KV obsolete for real-time apps?

Comparative benchmarks

Deno’s 2026 benchmark placed Cloudflare KV last for global latency. The test measured Upstash Redis as the leader because KV’s heavy caching hurts latency when reading shortly after a write. A cold read in KV takes 30ms, while a write takes over 170ms. For hot-key reads where data stays in the regional cache, KV is faster. Upstash makes one HTTPS request to the nearest region for every operation, which is about 6 ms from IAD to us-east-1. Upstash also supports the standard Redis protocol on port 6379 with TLS.

Upstash Redis handles complex data structures like sorted sets, streams, or hashes. KV only supports five operations: get, put, delete, list, and getWithMetadata. For atomic operations or strong consistency, Cloudflare recommends using Durable Objects. AWS DynamoDB provides strong consistency for read-after-write operations and uses 17 global locations. DynamoDB also includes local and global secondary indexes for different access patterns. KV lacks the ability to perform atomic operations. Deno KV is built on FoundationDB and provides sequential consistency.

KV is limited.

The verdict: Cloudflare Workers KV replaces Redis for simple, read-heavy edge configurations but fails as a substitute for complex, write-heavy stateful coordination.