Skip to content

yggdrasil.io.delta.snapshot

snapshot

Snapshot — collapsed table state at a specific Delta version.

Snapshot dataclass 

Snapshot(
    version: int,
    table_root: "Path",
    protocol: Protocol = Protocol(),
    metadata: Optional[Metadata] = None,
    active_files: "Dict[str, AddFile]" = dict(),
    txns: "Dict[str, int]" = dict(),
    domain_metadata: "Dict[str, DomainMetadata]" = dict(),
)

clustering_columns property 

clustering_columns: 'List[str]'

Liquid-clustering columns for this table, in cluster-key order.

Databricks records liquid clustering in a domainMetadata action under the delta.clustering domain — not in metaData.partitionColumns (a clustered table is unpartitioned). The domain config looks like::

{"clusteringColumns": [["region"], ["id"]],
 "domainName": "delta.clustering"}

where each entry is a physical column path (a list, to address nested struct fields). We flatten each path with . so a top-level column region surfaces as "region" and a nested addr.zip as "addr.zip" — the same dotted name the read predicate uses.

Older / non-Databricks writers sometimes stamp the columns in metaData.configuration['delta.clusteringColumns'] as a comma-joined string instead; we accept that shape as a fallback so a table written either way reports the same key. Returns [] when the table isn't clustered.

total_bytes property 

total_bytes: int

Sum of the active data files' on-disk sizes (from AddFile.size).

advanced 

advanced(
    log: DeltaLog, commit_files: "Iterable[Path]", version: int
) -> "Snapshot"

Return a new snapshot = this state + the actions in commit_files.

Used to advance a cached snapshot to a newer version by replaying only the commits after it — no checkpoint read, no re-reading the prior commits. commit_files are the (ascending) commit JSONs for versions (self.version, version] (see :meth:DeltaLog.commits_after).