State At The Edge
Edge functions often need a little data, such as routing rules, feature flags, or session lookups. Edge key value storage provides this as a globally distributed map optimized for fast reads near the user.
The Read Optimized Design
These stores favor reads over writes.
- Data is replicated to many POPs so reads are local and fast.
- A write at one place propagates to all replicas, which takes time.
- The system offers eventual consistency, meaning replicas converge after a delay.
A value you just wrote may not appear everywhere immediately. Designs must tolerate brief staleness.
What It Suits
- Configuration: flags and rules read on every request.
- Lookups: mapping a short code to a destination.
- Light personalization: small per user hints.
It is not a transactional database. Frequent writes, strong consistency, or large objects belong elsewhere.
Reasoning About Latency
- Reads: typically very fast because they hit a nearby replica.
- Writes: slower and only eventually visible globally.
- Cache layer: many platforms cache hot keys in the function memory, trading a little extra staleness for speed.
Key idea
Edge key value storage replicates small data to many POPs for fast local reads, accepting eventual consistency on writes, which makes it ideal for config and lookups but not for transactional workloads.