Recovery

Every destructive op (drop column, drop table) leaves an artifact — either a soft-drop tombstone or a real archive of the rows. zanith recover lists, inspects, restores, exports, and purges them.

#Five artifact kinds

Kind When Stored

**softDropColumn** An op declared soft-drop for a column. Column stays in the table, marked dropped in the artifacts table.

**softDropTable**

An op declared soft-drop for a table.

Table renamed to

_softdrop_, kept in the schema.**archiveColumn** Column dropped with archive policy. Rows moved to a __archive_ table, then column dropped.**archiveTable** Table dropped with archive policy. Table renamed to _archive_, schema kept.**backfillCheckpoint** A long-running backfill checkpointed. Row in the checkpoints table — restart picks up where it left off.
#The CLI surface
SHELL
7 lines
1zanith recover list                       # show every recoverable artifact
2zanith recover inspect <name>             # rows, size, recorded-at, source migration
3zanith recover restore-column <name>      # bring a soft-dropped or archived column back
4zanith recover restore-table <name>       # bring a soft-dropped or archived table back
5zanith recover export <name> --format csv # stream rows out to stdout (csv / jsonl)
6zanith recover purge <name>               # permanently delete the artifact
7zanith recover purge --older-than 30d     # bulk purge by age
#Restoring a column
SHELL
6 lines
1# A column was dropped two days ago — bring it back.
2zanith recover list
3# > orders_total_softdrop  · 14 days ago · 412 MB · from 20260420_120000_drop_orders_total
4 
5zanith recover restore-column orders_total_softdrop
6# > Re-attaches the column, restores the dropped index, marks the artifact restored.
Restore is idempotent — running it twice is a no-op the second time. The artifact stays in the recovery table marked restored until you purge it.
Heads up
Restoring an archived columnrebuilds the column and replays the rows from the archive table. That's a full-table operation — not what you want during peak hours on a multi-TB table.
#Exporting before purging
SHELL
8 lines
1# Take a CSV snapshot of an archive before purging.
2zanith recover export old_orders_archive_2026_03 --format csv > old_orders_2026_03.csv
3 
4# Or JSONL for downstream pipelines.
5zanith recover export old_orders_archive_2026_03 --format jsonl > old_orders.jsonl
6 
7# Then purge.
8zanith recover purge old_orders_archive_2026_03
#Retention policies
Each migration that creates an artifact can set retentionDays. purge --older-than sweeps anything past its retention window — a sensible cron entry on most production databases.
SHELL
2 lines
1# Daily housekeeping cron — purge anything past its retention window.
2zanith recover purge --older-than 30d --json > /var/log/zanith/purges.jsonl
#Reseeding into a new shape
The expand/contract reality: you archived a table, a NEW table took the old name with a different shape, and restore-table correctly refuses rather than clobber.reseed is the paved road — it pours the parked rows into the live table, with renames handled declaratively and unmapped columns falling back to their database defaults.
SHELL
5 lines
1# Old archive had "customer"; the new table calls it "customer_name".
2zanith recover reseed orders   --map customer_name=customer   --skip-conflicts        # ON CONFLICT DO NOTHING — re-runs are no-ops
3 
4# → ✓ Reseeded 1,204 row(s) into orders · columns: id, customer_name, total
5#   Archive kept — purge with: zanith recover purge <artifact-id>
Computed values (derive a column from the old row) live on the programmatic API, where functions fit:
TS
8 lines
1await reseedArtifact(adapter, artifact, {
2  map: {
3    customer_name: 'customer',                        // rename
4    status: (row) => row.total > 100 ? 'big' : 'std', // computed
5  },
6  skipConflicts: true,
7});
8// → { table, columns, read, inserted, skipped, batches }
Loud by design: unknown map targets and missing archive sources throw with the available columns named — a reseed never silently writes a partial shape.
#Programmatic API
The same operations are exposed as functions — see programmatic API for listArtifacts, findArtifact, restoreSoftDropColumn, restoreArchiveTable, readArtifactRows, reseedArtifact, purgeArtifact, purgeOlderThan.
← previous · MigrateShadow-DB verifyThe gate that catches drift before up touches production.next · Migrate →Audit + historyFour bookkeeping tables, the history CLI, snapshot replay.