菜单
Deep domain knowledge for Octant's N-dimensional data system, storage backends (Zarr, Icechunk), OctantBlock slicing, LRU caching (BlockCache), and background prefetching (BlockPrefetcher). Use when modifying or debugging src/data/.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/lazarusA/octant --skill octant-data-engine命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | octant-data-engine |
| description | Deep domain knowledge for Octant's N-dimensional data system, storage backends (Zarr, Icechunk), OctantBlock slicing, LRU caching (BlockCache), and background prefetching (BlockPrefetcher). Use when modifying or debugging src/data/. |
This skill guides development and debugging of Octant's N-dimensional tensor extraction, caching, and storage backends in src/data/.
Storage Backend (Zarr/Icechunk)
│
▼
BlockStore Trait (inspect, fetch_block, fetch_blocks)
│
▼
DatasetManager (StoreHandle registry & lifecycle)
│
▼
BlockPrefetcher (bounded background lookahead pool)
│
▼
BlockCache (LRU cache of OctantBlock hyperslabs)
│
▼
OctantBlock (N-D strided memory block)
│
▼
MatrixData / Volume (f32 renderable GPU payload)
BlockStore Trait (src/data/block_store.rs)To add a new storage backend:
backend_name(&self) -> &'static strvariables(&self) -> &[VariableInfo]inspect(&self) -> &DatasetMetadatafetch_block(&self, request: &BlockRequest) -> Result<OctantBlock, DataError>fetch_blocks(&self, requests: &[BlockRequest]) -> Result<Vec<OctantBlock>, DataError>object_store::ClientOptions) must always configure explicit timeouts (.with_timeout(Duration::from_secs(30)) and .with_connect_timeout(Duration::from_secs(10))) to prevent worker starvation.OctantBlocksrc/data/octant_block.rsArc<[f32]> for $O(1)$ zero-copy sharing between cache, resamplers, and GPU upload buffers.slice_2d(dim_x, dim_y, fixed_indices) -> Result<MatrixData, DataError> extracts 2D matrices for GPU rendering.volume(dim_x, dim_y, dim_z, fixed_indices) -> Result<MatrixData, DataError> extracts 3D volumetric slabs.shape.iter().copied().try_fold(1usize, |acc, d| acc.checked_mul(d)).unwrap_or(usize::MAX)
BlockCache & BlockPrefetcher (src/data/block_cache.rs, src/data/block_prefetch.rs)BlockCacheKey hashes dataset URI, variable name, and selection ranges.BlockPrefetcher runs a dedicated worker pool to fetch upcoming frames along the animation dimension before the UI requests them.sync_channel(N)) to ensure backpressure when the UI consumer lags behind prefetch workers.try_recv() inside the main UI loop.tokio multi-threaded runtime for parallel I/O.wasm-bindgen-futures. Avoid blocking threads on WASM.