Data Vault families

Twelve families write the vault layer. This page documents seven of them: vault-entity, vault-control, and the five load families.

The other five render the combined load, which emits one artifact per source rather than one per vault object: vault-combined-load (databricks/data_vault/sql/combined_load.jinja, roots source, tiers, load and behavior) and the four include-only pieces it composes, vault-piece-hub, vault-piece-link, vault-piece-satellite and vault-piece-link-satellite, each taking a single object root. They are indexed on the overview page and their field-by-field sections are still to be written. The combined load is generated but not yet routed into a build, so a change you make in those five does not reach a deployed pipeline until that wiring ships.

Vault templates split by how the object loads, not by what it is called: hubs and links share the keyed insert, all three satellite kinds share the versioned insert, and reference, bridge, and point-in-time each have their own. Two conventions hold across all five load families: this is the vault object being written (the catalog row passed through whole, with the fully qualified name dialect-cased and a snake-cased name added), and bridge and point-in-time carry no source roots at all. That last one is deliberate: they rebuild from the whole raw vault, and their own bridge.steps[] and pit.satellites[] already name every table they read.

Shared ghost-record block (ghost), where a family carries it:

Field	Type	Meaning
ghost.enabled	boolean	Whether the ghost-row flow renders; defaults to true when ghost handling is unset.
ghost.zeroKeySql	string	The all-zeros hash-key literal, UNHEX(REPEAT('0', ...)) at the hash algorithm’s width.
ghost.loadDateSql	string	The epoch load date stamped on ghost rows: TIMESTAMP '1970-01-01 00:00:00'.
ghost.recordSource	string	The SYSTEM record-source constant stamped on ghost rows.

vault-entity

Templates: databricks/data_vault/link.jinja

Roots: this, cdc, source, sources, hubHashKeys, dependentChildColumns (conditional), ghost

The streaming link entity definition. The link’s own columns are derived from the keys and hash-key roots rather than enumerated, so this.columns is empty and honest.

Field	Type	Meaning
this.name	string	The link’s logical name.
this.fqn	string	The link’s fully qualified name.
this.physicalName	string	The finished physical table name (the link prefix applied to the snake-cased name).
this.description	string	The link’s documentation text.
this.columns	array	Empty and honest: the link’s columns come from the key roots.
this.linkConfig.hashKeyColumn	string	The link’s own hash-key column.
this.linkConfig.loadDateColumn	string	The link’s load-date audit column.
this.linkConfig.recordSourceColumn	string	The link’s record-source column.
this.linkConfig.hubFqns	string[]	Fully qualified names of the participating hubs, in declared order.
this.linkConfig.hubHashKeyColumns	string[]?	Per-hub hash-key names stamped at decompose time; wins over pattern derivation when it length-matches hubFqns.
this.linkConfig.dependentChildColumns	string[]?	Dependent-child column source names designated on the link itself.
cdc.keys	string[]	The change-feed KEYS clause: the link’s hash key.
cdc.sequenceBy	string	The change-feed SEQUENCE BY column.
source / sources	table ref + nameSnakeCase	The streaming source, named rather than enumerated (columns empty); source.nameSnakeCase precomputed. sources holds the identical object.
hubHashKeys	string[]	Participating hub hash-key column names, in hubFqns order (the stored stamp, or pattern-derived for legacy rows).
dependentChildColumns	string[]	Conditional: dependent-child columns resolved to their persisted physical names; absent entirely when none are designated.
ghost	object	The shared ghost-record block above.

vault-control

Templates: databricks/data_vault/control.jinja

Roots: source, tiers, threadCount

The per-source control notebook that runs a source’s vault loads in tiers. References, business vault, point-in-time, and bridge are deliberately absent: they run in the vault workflow’s pre-stage and post-stage, not in a source’s control notebook.

Field	Type	Meaning
source.name	string	Leaf name of the staging source this control notebook drives.
source.fqn	string	Dialect-cased fully qualified name of that source.
tiers[]	array	Ordered execution tiers; an empty tier is filtered out before render.
tiers[].label	string	Tier heading: hubs and links (tier one), satellites and link satellites (tier two).
tiers[].steps[].name	string	The load’s leaf name, for example hub_customer.
tiers[].steps[].notebookPath	string	Relative path to the load notebook, ../loads/<leaf>.
threadCount	number	Parallelism for the tier’s thread pool; the Build emits 4.

vault-load-keyed

Templates: databricks/data_vault/loads/keyed_insert_load.jinja

Roots: kindLabel, this, keyColumn, source (conditional), sources, ghost, ghostHashKeyColumns, sequenceColumn, recordSourceColumn

The idempotent insert for the four keyed kinds: hub, link, business hub, business link. The one vault family whose columns carry ghost sentinels.

Field	Type	Meaning
kindLabel	string	Human label for the header comment: Hub, Link, Business hub, or Business link.
this	object	The vault object being written: the catalog row spread whole, fqn dialect-cased (replacing the raw one), nameSnakeCase added.
this.columns[].ghostSentinelSql	string	Per-column null-safe sentinel for the ghost row, by data type: 0 numeric, false boolean, DATE '1900-01-01', TIMESTAMP '1900-01-01 00:00:00', X'' binary, '' otherwise.
this.hubConfig / .linkConfig / .bhubConfig / .blinkConfig	object or null	The kind’s configuration payload; each carries recordSourceColumn.
keyColumn	string	The object’s own hash-key column, which the insert keys on.
source	table ref?	Conditional: sources[0], identical object; absent when the resolver returned none.
sources[]	table ref[]	Every upstream staging table feeding this load, in resolver order; columns ordinal-sorted, empty when unresolvable.
ghost	object	The shared ghost-record block.
ghostHashKeyColumns	string[]	Every hash-key column taking the zero key on the ghost row: the object’s own keys, then the participating hub hash keys (links only, so a link ghost points at hub ghost rows).
sequenceColumn	string	The column the merge sequences on.
recordSourceColumn	string	The record-source column from whichever configuration payload the row carries; empty when none.

vault-load-versioned

Templates: databricks/data_vault/loads/versioned_insert_load.jinja

Roots: kindLabel, this, source (conditional), sources, keyColumns, sequenceColumn

The versioned insert for the three satellite kinds. No ghost roots by design: a versioned load emits no ghost row.

Field	Type	Meaning
kindLabel	string	Satellite, Link satellite, or Business satellite.
this	object	Same shape as the keyed family’s this, except columns are raw (no ghost sentinels).
source / sources	table ref	Identical semantics to the keyed family.
keyColumns	string[]	The parent hub or link hash-key column(s) the versioning partitions by.
sequenceColumn	string	The merge sequence column; when the change-feed default _commit_timestamp resolves for a satellite kind, the vault load-date column substitutes.

vault-load-reference

Templates: databricks/data_vault/loads/reference_upsert_load.jinja

Roots: this, codeColumn, source (conditional), sources

Field	Type	Meaning
this	object	The reference table being written; same shape as the versioned family’s this.
codeColumn	string	The reference table’s code column: the upsert’s match key. Empty when absent.
source / sources	table ref	Identical semantics to the keyed family.

vault-load-bridge

Templates: databricks/data_vault/loads/bridge_rebuild_load.jinja

Roots: this, bridge

The bridge rebuild. No source roots (the ratified exception): the traversal below carries every table the load reads.

Field	Type	Meaning
this	object	The bridge table being written.
bridge.anchor.fqn	string	The anchor hub (or business hub) the traversal starts from.
bridge.anchor.hashKeyColumn	string	The anchor’s hash-key column.
bridge.steps[]	array	Ordered traversal steps, each link joined to its far hub.
bridge.steps[].alias	string	SQL alias for the step’s link.
bridge.steps[].linkFqn / .linkHashKeyColumn	string	The step’s link and its hash-key column.
bridge.steps[].hubFqn / .hubHashKeyColumn	string	The far hub this step reaches and its hash-key column.
bridge.steps[].joinFromTableAlias / .joinFromHashKeyColumn	string	The alias this step joins from (the anchor, or an earlier step’s hub) and the hash-key column on it.

vault-load-pit

Templates: databricks/data_vault/loads/pit_snapshot_load.jinja

Roots: this, pit

The point-in-time snapshot rebuild. No source roots, same exception as bridge.

Field	Type	Meaning
this	object	The point-in-time table being written.
pit.anchor.fqn	string	The anchor hub or link supplying the spine’s keys.
pit.anchor.hashKeyColumn / .loadDateColumn	string	The anchor’s hash-key and load-date columns.
pit.satellites[].fqn	string	A satellite the snapshot reaches back into.
pit.satellites[].columnPrefix	string	Output column prefix, de-duplicated when the satellite’s name already carries it (never sat_sat_customer_detail).
pit.snapshotDateColumn	string	The snapshot spine’s date column.
pit.cadenceIntervalSql	string	The spine cadence as an INTERVAL literal: 1 HOUR, 1 DAY, 7 DAY, or 1 MONTH (daily when unset).
pit.satLoadDateColumn	string	The load-date audit column every satellite is scanned at-or-before.