Details
-
Bug
-
Status: Resolved
-
Blocker
-
Resolution: Fixed
-
M4.5
-
None
Description
In further investigating various alter table bugs, I think I discovered another one: our Schema object doesn't maintain any "next_unused_column_id_" member or anything. So, if you issue an ALTER which drops a column, and then a separate ALTER which adds a new one, you may end up resuscitating old data. Even worse, it might be old data from an entirely unrelated column with a different type.
There's a test which attempts to show this doesn't happen in tablet-schema-test, but it's too localized to Tablet and reuses a single SchemaBuilder which does properly avoid column id reuse. We need an integration test for this case.
Attachments
Issue Links
- is related to
-
KUDU-382 Add coverage for and fix concurrent AlterSchema
- Resolved