Choosing a reader strategy

This section covers reading sharded KV snapshots. Pick based on your code style and backend:

Reader	Code style	Backends	Best for
Sync SlateDB	Synchronous	SlateDB	Default, lowest friction
Sync SQLite	Synchronous	SQLite	SQL queries, download-and-cache or range-read VFS
Async SlateDB	asyncio	SlateDB	FastAPI, async workers
Async SQLite	asyncio	SQLite	Async SQLite wrappers

All readers share the same conceptual model: load _CURRENT → dereference manifest → open per-shard adapters → route lookups locally. See KV Storage Overview for the conceptual model.

Key concept: refresh

Readers pin all lookups to the manifest loaded at open time (or last refresh()). To observe a newly published snapshot:

changed = reader.refresh()  # True if newer manifest was loaded

This is the only way to advance — there is no automatic polling.