[IGNITE-16033] Wrong completion of an alter table operation for a slowing node

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0-alpha4
• Labels: ignite-3
• Assignee: Vladislav Pyatkov
• Reporter: Kirill Tkalenko
• Reviewer: Alexander Lapin
• Created: 01/Dec/21 09:59
• Updated: 06/Dec/21 07:06
• Resolved: 06/Dec/21 07:06

Description

Two nodes send a request to change the table (for example, adding the column "column1"), if one of the nodes slowed down (could not update the configuration in the metastorage due to the network), then one of the nodes will be able to perform the operation (the column will be added), and the second, after returning to the cluster there will be no error to perform the operation, since when a new configuration is received from the metastorrage, a notification about a schema change will occur, which will complete all current operations to change the table, although there should be an error (the column already exists). In fact, the error will be later, but she will not be able to complete the operation with it because the operation was completed on notification of a change in the table configuration.

Tests:

• org.apache.ignite.internal.runner.app.ItTablesApiTest#testAddColumnFromLaggedNode;
• org.apache.ignite.internal.runner.app.ItTablesApiTest#testAddIndexFromLaggedNode.

Issue Links

• is blocked by: IGNITE-16044 Configuration should provide a revision after a change is completed
• links to: GitHub Pull Request #488