Snowflake restore lets you recover dropped or point-in-time objects using Time Travel, Fail-safe, and zero-copy clones.
Use Time Travel’s UNDROP or CLONE to recover the table instantly. UNDROP rebuilds the original object, while CLONE creates a new copy at a chosen timestamp.
Snowflake keeps dropped objects for your account’s DATA_RETENTION_TIME.Running UNDROP TABLE restores the object exactly as it was at the drop moment, including data and permissions.
Pick CLONE when you need to test fixes without overwriting production. A zero-copy clone shares storage with the source but presents an independent object.
Specify the object, timestamp, or offset. Wrap timestamps in quotes and set the time zone if needed.
1) Detect the bad DELETE on Orders
.2) Identify the safe timestamp. 3) Run CREATE TABLE Orders_fix CLONE Orders AT (TIMESTAMP => '2023-10-01 12:00:00')
. 4) Swap tables or merge rows.
• Verify retention period is long enough for your RPO.
• Store timestamps of risky deployments.
• Use CLONE for hotfixes; UNDROP for simple rollbacks.
• Grant OWNERSHIP only to controlled roles.
UNDROP costs zero extra storage; CLONE adds storage only for new changes after creation.Compute charges apply while querying or copying restored data.
The same commands work at higher levels. Use UNDROP SCHEMA
or CREATE DATABASE db_clone CLONE mydb AT (OFFSET => -3600)
to roll back one hour.
After Time Travel, Fail-safe allows Snowflake support to recover data for seven additional days, but expect longer wait times and a support ticket.
.
Yes, but retention periods vary. Standard offers 1 day, Enterprise and above offer up to 90 days.
No. Zero-copy clones share micro-partitions; storage grows only for post-clone changes.
No. Time Travel works only within the same account. Use shares or database replication for cross-account recovery.