Tutorial: Building with CleaveDB

REWIND — restore a document

When an erroneous update corrupts an important document, REWIND restores that document back to the exact state it had at a specified point in time, using CleaveDB's MVCC history ledger.

Prerequisite: Audited or versioned bucket

To restore earlier versions, the target bucket must have change logging enabled before the mutations occur:

-- Required once before changes occur
SHAPE BUCKET staff AUDITED

If a bucket is not audited or versioned, CleaveDB will reject the command with: No history for 'staff:a'. Run SHAPE BUCKET staff AUDITED ....

Restoring document state

Issue REWIND with the qualified document ID (bucket:id) and target timestamp:

-- Make a sequence of updates
CHANGE staff "a" SET age TO 36
CHANGE staff "a" SET age TO 37

-- Restore document state from 1 minute ago
REWIND "staff:a" TO "1 minutes ago"

-- Verify the restored document
FIND staff "a"

The fields of staff:a now match what was stored at that exact minute.

Temporal syntax variants

Graph bonds survive rewinds

Unlike relational cascades that break foreign keys, CleaveDB graph bonds are decoupled from document revisions. Rewinding a document modifies its internal attributes without severing incoming or outgoing bonds:

BOND "staff:a" TO "staff:b" AS "pal"
CHANGE staff "a" SET age TO 50
REWIND "staff:a" TO "now"

-- The graph bond remains completely intact
FIND "pal" OF "staff:a"

Any aggregations (DISTILL) run after the rewind instantly reflect the restored document attributes.