Edge computing: nearby execution, explicit data distance

An edge runtime can reduce the client-to-compute leg. It cannot repeal network distance to authoritative data, make every store globally writable, or promise a fixed latency. Routing, cold state, queueing, storage location and provider limits all affect observed percentiles.

Runtime and concurrency

Cloudflare Workers isolates can handle interleaved requests, may be evicted, and are not guaranteed to receive the next request for a user. Global mutable variables are therefore neither durable storage nor a synchronization primitive. Use a named durable coordinator or transactional store for shared effects and stable operation IDs for ambiguous retry.

Data choices

• Workers KV is read-optimized and eventually consistent. Remote cached values, including negative lookups, can remain stale; concurrent writes can overwrite.
• D1 has one write primary. Read replicas update asynchronously. Sessions and bookmarks provide sequential consistency for a logical session.
• A worker near an Asia user that queries a US database still pays the WAN and database round trips. Measure, co-locate, reduce chattiness, or choose a data model whose consistency contract fits.

No scenario references an implicit or nonexistent regional node: every runtime target and hop is present in the topology.

Scenarios

Sources

• How Workers works
• How Workers KV works
• D1 read replication
• Vercel Edge runtime and data-source distance