hdf5-virtual-dataset · shipping · also HDF5 VDS

HDF5 virtual dataset

An HDF5 virtual dataset maps selections from source datasets into a single logical array without storing those source values in the virtual mapping itself.

Related Observatory article: HDF5 virtual dataset on hesela.com

Mapping: sources are identified by filenames, dataset paths and selections, and may be defined before the sources exist. Ordinary array reads follow the mapping. [1]

Missing input: unavailable mapped data can be returned as a fill value. A successful read therefore does not establish that every source was present. [2]

Resolution: HDF5_VDS_PREFIX can change where relative source names are found. A hash of the virtual file identifies its bytes, not every source selected by the reader environment. [3]

Measured example: Hesela's fixed-shape experiment returned zeros for a missing source and replacement integers after a prefix override, while the virtual file hash stayed unchanged. This is bounded API evidence, not a claim about every source failure. [4]

Relations

related

hdf5, hdf5-fill-value, backup-verification, hdf5-hyperslab

Citations

1. h5py 3.15.0: Virtual Datasets (accessed 2026-10-17)
2. The HDF Group: Introduction to the Virtual Dataset (accessed 2026-10-17)
3. The HDF Group: H5Pset_virtual source-file resolution and H5Pset_fill_value (accessed 2026-10-17)
4. Hesela: HDF5 virtual dataset dependency observations (accessed 2026-10-17)

JSON: /data/terms/hdf5-virtual-dataset.json