[IGNITE-19061] Node restart rewrites data nodes with too actual nodes (Design ticket)

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Done
• Fix Version/s: 3.0
• Labels: ignite-3
• Epic Link: Distribution Zones [Restart]
• Assignee: Mirza Aliev
• Reporter: Mirza Aliev
• Created: 17/Mar/23 13:44
• Updated: 04/Sep/24 16:10
• Resolved: 31/May/23 09:30

Description

Currently, in the DistributionZoneManager.start method we have a logic that initialises data nodes in Meta Storage. It takes logical topology from Vault and write it to the Meta Storage as a data nodes.

This behaviour is not correct, lets consider the scenario:

1. We have cluster [A, B, C], new node E was added to the logical topology, scale up timer was scheduled to add that node to a data nodes of some zone Zone_1 after 1 hour.
2. Right after that node A from cluster was restarted, so it will rewrite data nodes for Zone_1 according the logic from the DistributionZoneManager.start
3. Data nodes of Zone_1 become [A, B, C, E], but user was expecting, that data nodes for Zone_1 will be changed after 1 hour.

UPD:

Current ticket is reclassified to the ticket for design, so the proper solution for the restart and restoring timers must be provided.

Issue Links

• blocks: IGNITE-18755 Empty data nodes list of a distribution zone.