Data on the hub is tiered. Everything stays browsable at all times, so opening a dataset, listing its variables and reading its metadata works whether or not the data itself is on disk.
What changes is the first read. Data nobody has touched for a long time is moved to tape, and reading it again triggers a recall:
If you are working with a dataset intensively and would rather it stayed on disk throughout, ask us to pin it. That is a deliberate decision with a name and a reason attached, rather than something to work around.
Contact supportThe hub is built from a few small services. Each has its own documentation; this page is the index.
blobmap
Decides which zarr objects move to tape together, and reads that decision back. A store is millions of objects, which is too many to track individually and too few to treat as one unit: blobs are what sits in between and makes tiering tractable.
DocumentationManifest format| Data Browser | search the catalogue |
| STAC Browser | the same holdings, as STAC |
| Concepts | why the data is laid out this way |
| blobmap | how the tiering decision is made |
