[IGNITE-23454] Sql. Worker node left the cluster during transaction initialization phase

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.0.0-beta1
• Component/s: sql
• Labels: ignite-3
• Epic Link: Processing sql on unstable topology
• Assignee: Unassigned
• Reporter: Evgeny Stanilovsky
• Created: 16/Oct/24 05:56
• Updated: 25/Nov/24 12:13
• Resolved: 25/Nov/24 12:13

Description

Case: Partitions are enlisted alongside the ClusterNode which is considered to be the owner of the primary replica of the particular partition. Before enlistment, ClusterNode is resolved by consistentId of the replica, which was returned from PlacementDriver. If at the moment of resolution node with given consistentId has left the cluster, query execution must be restarted from the Mapping phase. Worth to mention, that new assignment must be acquired from PlacementDriver.

Definition of done:
1. Correctly process the case when node by consistentId is unavailable.
2. Check tx enlist process with consistentId change (suppose we have explicit tx and sequential queries with topology change between execution)

Issue Links

• duplicates: IGNITE-22855 Sql. Missing ClusterNode can be involved into tx enlist
• incorporates: IGNITE-22855 Sql. Missing ClusterNode can be involved into tx enlist