[DocDB] Support creating a backup in the middle of tablet splitting #29002
Description
Jira Link: DB-18752
Description
The new backup workflow (that supports creating backups in the middle of a DDL) doesn't handle the case of creating the backup in the middle of a tablet split correctly. Take the following example:
- CREATE TABLE mytbl (k INT PRIMARY KEY, v INT)
- Insert 200 rows.
- Check we have 1 tablet only:
- Split the tablet into 2 tablets and wait until we have 2 tablets (the parent tablet is not deleted).
- Create a backup.
- Restore the backup.
Restore will fail at import_snapshot with the following error:
[m-1] E1017 18:33:53.598845 77952 catalog_entity_info.cc:731] mytbl [id=00004000000030008000000000004000]: Two tablets {mytbl_tablet_id3}, {mytbl_tablet_id2} with the same partition key start and split depth: committed_consensus_state { current_term: 0 config { opid_index: -1 } } state: PREPARING table_id: "00004000000030008000000000004000" partition { partition_key_start: "" partition_key_end: "v\237" } colocated: false hosted_tables_mapped_by_parent_id: true and
The reason for the failure is that the created snapshot will include both the parent tablet and the two child tablets.
This leads import_snapshot to surface an error that there are 2 tablets with the same partition key start and split depth.
The two tablets has been created as part of repartition step of import snapshot:
[m-1] I1017 18:33:53.598722 77952 catalog_manager_ext.cc:2199] ImportTableEntry: Found existing table 00004000000030008000000000004000 for 00004000000030008000000000000000/mytbl (old table 000034d4000030008000000000004000) with schema public
[m-1] I1017 18:33:53.598752 77952 catalog_manager_ext.cc:1937] RepartitionTable: Repartition table 00004000000030008000000000004000 using external snapshot table 000034d4000030008000000000004000
[m-1] I1017 18:33:53.598778 77952 catalog_manager_ext.cc:1994] Created tablet {mytbl_tablet_id2} to replace tablet {mytbl_tablet_id1} in repartitioning of table 00004000000030008000000000004000
[m-1] I1017 18:33:53.598784 77952 catalog_manager_ext.cc:1994] Created tablet {mytbl_tablet_id3} to replace tablet f29cbec9fea8448b808f838010e8177e in repartitioning of table 00004000000030008000000000004000
[m-1] I1017 18:33:53.598794 77952 catalog_manager_ext.cc:1994] Created tablet {mytbl_tablet_id4} to replace tablet f4d7c73a92a846dc9bb73673b54b417a in repartitioning of table 00004000000030008000000000004000
[m-1] E1017 18:33:53.598845 77952 catalog_entity_info.cc:731] mytbl [id=00004000000030008000000000004000]: Two tablets {mytbl_tablet_id3}, {mytbl_tablet_id2} with the same partition key start and split depth
We should avoid including the parent tablet in the snapshot in case the two child tablets are registered.
Issue Type
kind/bug
Warning: Please confirm that this issue does not contain any sensitive information
- I confirm this issue does not contain any sensitive information.