Restore tables¶

Prerequisites¶

• Scylla Manager with CQL credentials to restore the destination cluster.

• Restoring the content of the tables assumes that the correct schema is already present in the destination cluster.

  Scylla Manager does NOT validate that, so it’s user responsibility to ensure it. The only form of validation that Scylla Manager performs is checking whether restored tables are present in destination cluster, but it does not validate their columns types nor other properties. In case destination cluster is missing correct schema, it should be restored first.

• It is strongly advised to restore the contents of the tables only into truncated tables.

  Otherwise, restored tables’ contents might be overwritten by the already existing ones. Note that an empty table is not necessarily truncated!

• It is important to remember current mode of the tombstone_gc option.

  In order to ensure data consistency, Scylla Manager alters restored tables options with tombstone_gc = {'mode': 'disabled'}. It’s user responsibility to reset the tombstone_gc option as a part of the restore tables follow-up action.

• Materialized views and secondary indexes must be recreated after the table is successfully restored.

Follow-up action¶

After the successful restore, it is important to perform the necessary follow-up action. After restoring the content of the tables, you should repair all restored keyspaces and tables. Without the repair, the restored content of the tables might not be visible, and querying it can return various errors. ONLY AFTER repairing restored tables, you should set their tombstone_gc option to desired mode (e.g. previous mode or strongly recommended tombstone_gc = {'mode': 'repair'}). Without that, purging tombstones for those tables will remain disabled. This can result in great memory consumption over time, as deleted rows won’t ever be removed from database’s memory.

The last step is to recreate Materialized Views and Secondary Indexes. To do it, just drop materialized views and indexes, then execute index creation and materialized views creation commands. See examples below.

// To list all indexes
SELECT keyspace_name, table_name, index_name FROM system_schema.indexes WHERE keyspace_name='<input>' AND table_name='input';

// To get the index creation command
DESCRIBE <keyspace_name>.<index_name>;

// To drop the index
DROP INDEX <keyspace_name>.<index_name>;

// To list all materialized views (please note that index creates the view as well - view name suffix is "_index" then)
// Do not recreate materialized view that are part of the index definition.
SELECT keyspace_name, view_name, base_table_name FROM system_schema.views WHERE keyspace_name='<input>' AND base_table_name='<input>';

// To get the materialized view creation command
DESCRIBE <keyspace_name>.<view_name>;

// To drop the materialized view
DROP MATERIALIZED VIEW <keyspace_name>.<view_name>;

Process¶

Restore procedure works iteratively over restored locations, manifests and tables.

• For each backup location:

  • Find live Scylla nodes that will be used for restoring current location

    • Find nodes local to location

    • If non can be found, find any node with location access

  • List backup manifests for specified snapshot tag

• Alter all restored tables with tombstone_gc = {'mode': 'disabled'} (ensure restored data consistency)

• For each manifest:

  • Filter relevant tables from the manifest

  • For each table:

    • Group SSTables to bundles by ID

    • Join bundles into batches containing batch size bundles each

• For each batch (in --parallel):

  • Select node with enough free space to restore given batch

  • Download batch to node’s upload directory

  • Call load and stream with primary_replica_only