[FLINK-27316] Prevent users from changing bucket number

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: table-store-0.1.0
• Fix Version/s: table-store-0.1.0
• Component/s: Table Store
• Labels: pull-request-available
• Assignee: Jane Chan
• Reporter: Jane Chan
• Created: 20/Apr/22 07:42
• Updated: 14/Jun/22 02:51
• Resolved: 21/Apr/22 02:26

Description

Before supporting this feature, we should throw a meaningful exception to prevent data corruption which is caused by

 ALTER TABLE ... SET ('bucket' = '...');
 ALTER TABLE ... RESET ('bucket');

 

How to reproduce

-- Suppose we defined a managed table like
CREATE TABLE IF NOT EXISTS managed_table (
  f0 INT,
  f1 STRING) WITH (
    'path' = '...'
    'bucket' = '3');

-- then write some data
INSERT INTO managed_table
VALUES (1, 'Sense and Sensibility),
(2, 'Pride and Prejudice), 
(3, 'Emma'), 
(4, 'Mansfield Park'), 
(5, 'Northanger Abbey'),
(6, 'The Mad Woman in the Attic'),
(7, 'Little Woman');

-- change bucket number
ALTER TABLE managed_table SET ('bucket' = '5');

-- write some data again
INSERT INTO managed_table 
VALUES (1, 'Sense and Sensibility'), 
(2, 'Pride and Prejudice'), 
(3, 'Emma'), 
(4, 'Mansfield Park'), 
(5, 'Northanger Abbey'), 
(6, 'The Mad Woman in the Attic'), 
(7, 'Little Woman'), 
(8, 'Jane Eyre');

-- change bucket number again
ALTER TABLE managed_table SET ('bucket' = '1')

-- then write some record with '-D' as rowkind
-- E.g. changelogRow("-D", 7, "Little Woman"),
-- changelogRow("-D", 2, "Pride and Prejudice"), 
-- changelogRow("-D", 3, "Emma"), 
-- changelogRow("-D", 4, "Mansfield Park"), 
-- changelogRow("-D", 5, "Northanger Abbey"), 
-- changelogRow("-D", 6, "The Mad Woman in the Attic"), 
-- changelogRow("-D", 8, "Jane Eyre"), 
-- changelogRow("-D", 1, "Sense and Sensibility"), 
-- changelogRow("-D", 1, "Sense and Sensibility") 

CREATE TABLE helper_source (
  f0 INT, 
  f1 STRING) WITH (
  'connector' = 'values', 
  'data-id' = '${register-id}', 
  'bounded' = 'false', 
  'changelog-mode' = 'I,UA,UB,D'
); 

INSERT INTO managed_table SELECT * FROM helper_source;

-- then read the snapshot

SELECT * FROM managed_table

which will get wrong results

Issue Links

• relates to: FLINK-28035 Support rescale overwrite
• links to: GitHub Pull Request #96