Optionaloptions: ChunkPrefetcherOptionsPrivatestorePrivatemaxPrivateenabledPrivatedebugPrivateinPrivatequeuePrivateprocessingPrivateisPrivate Static ReadonlyMAX_PrivateseenPrivateparsedCache for parsed chunk indices (avoids repeated regex parsing per access)
PrivatemaxUpper bounds per array path for suppressing out-of-range prefetch requests
Called by store after L2 hit or L3 fetch. Enqueues adjacent chunks for prefetching.
PrivateaddAdd a key to the queue, unless it is already queued or in flight.
PrivateprocessProcess the prefetch queue with concurrency limiting. Race-condition safe via processing flag and queueMicrotask re-trigger.
PrivateparseRegister array shape and chunk sizes to enable adjacency prefetch. Neighbors outside the array's chunk grid are skipped.
Base path of the array (e.g., 'gsplats_t0023/centers')
Array shape (e.g., [2096, 4])
Chunk sizes (e.g., [1024, 4])
PrivategetGenerate adjacent chunk keys (±1 in each dimension). Returns [] unless bounds are registered for the key's array path.
Dispose the prefetcher, clearing all internal state and stopping processing.
Get prefetch statistics.
PrivatelogConditional logging based on debug mode.
Chunk prefetcher that proactively fetches adjacent zarr chunks.
Triggered on L2 hits and L3 fetches to hide network latency by prefetching adjacent chunks (±1 in each dimension) before they are explicitly requested.
Key features:
See: docs/guides/specs/CACHE_PREFETCHING_SPEC.md