[IGNITE-22969] Sql. Replanning query on unstable topology

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.0.0-beta1
• Fix Version/s: 3.0
• Component/s: sql
• Labels: ignite-3
• Epic Link: Processing sql on unstable topology
• Assignee: Evgeny Stanilovsky
• Reporter: Andrey Mashenkov
• Created: 09/Aug/24 10:47
• Updated: 07/Nov/24 07:09
• Resolved: 07/Nov/24 07:09

Description

ExecutionTargetFactory implementations calculates execution nodes for the query fragments and then effectively finds colocated fragments operating with nodes order (int) instead of nodes ids (string).

To do that, we need to know all available nodes. So, we read LogicalTopology before mapping query fragments. Then we read assignments for all the tables used in the query.
On unstable topology, we may observe a node in assignment, which wasn't part of LogicalTopology at the previous step. E.g. the missed node can be just failed node or can be just joined node.

In IGNITE-22861 was decide to ignore unknown nodes.

So, possible solution

• left "as is"
• throw an exception when unknown node was found, then restart query planning phase
• add a relation between LogicalTopology and partition Assignments to avoid this case
• use single source (e.g. assignments) for getting cluster nodes.

Issue Links

• Discovered while testing: IGNITE-22861 Sql. Calculation for mapOnBackups parameter erased.
• incorporates: IGNITE-22855 Sql. Missing ClusterNode can be involved into tx enlist
• links to: GitHub Pull Request #4629