Glossary

Use this page when you need a short definition before reading a deeper concept page.

Terms

Plugin — Product-specific code that reads source items and shapes output data. See Plugins.

Run — One firecube ingest ... execution against one product target.

Batch — A group of source items processed as one unit of work.

Product URI — The full file:// or s3:// location of the output product, such as file:///data/products/MY_PRODUCT.zarr.

Product root — The storage location for one Firecube product. It contains the readable data store and ChunkManager state.

Data store — The product data that users and downstream tools read, such as Zarr arrays, Parquet files, or a Tensogram .tgm file.

ChunkManager — Firecube's product lifecycle surface. It records runs, spans, write claims, snapshots, and cleanup state so products can be inspected, resumed, recovered, and cleaned up.

ChunkManager record — A logical Firecube record about run, span, claim, or snapshot state. It is not a physical Zarr array chunk.

Zarr chunk — A physical storage chunk inside a Zarr array. Zarr chunks are controlled by output-format tuning options and affect read/write performance.

Claim — A write-coordination record used to prevent conflicting work against the same product area.

Span — A record of what one run wrote, usually with group or time coverage.

Snapshot — A derived ChunkManager view used by inspection and cleanup commands.

Storage driver — The I/O implementation selected for a run, such as fsspec or obstore.

Output format — The shape Firecube writes: Zarr for gridded datacubes, Parquet for tabular records, or Tensogram for packaging finished Zarr products.

Write mode — The write strategy selected for ingestion. staged writes to a temporary local product before committing; direct writes to the target product path directly.

Pipeline worker — A worker inside one Firecube run that can prepare or write batches concurrently when the plugin class and output format allow it.

Slot range — A fixed index range assigned to a direct-Zarr worker. Slot ranges must be disjoint and chunk-aligned for safe parallel writes.

Next Steps

• Concepts — return to the chapter map
• Storage & ChunkManager — understand product state, recovery, claims, and cleanup
• Output Formats — choose Zarr, Parquet, or Tensogram
• Parallelism — choose a safe parallel model