Evgeny Stanilovsky created IGNITE-18377:
-------------------------------------------
Summary: Sql statistics become broken after node restart with
enabled persistence.
Key: IGNITE-18377
URL: https://issues.apache.org/jira/browse/IGNITE-18377
Project: Ignite
Issue Type: Bug
Components: sql
Affects Versions: 2.14
Reporter: Evgeny Stanilovsky
Assignee: Evgeny Stanilovsky
It tuned out that for some reasons SQL query may be dispatched to data node
which is performing rebalance. Eventually this will lead to falling back to
full table scan usage instead of proper tree index.
Scenario:
1. Cluster in normal state with multiple nodes, under workload (both cache
writes/updates and SQL selects)
2. Particular node A is restarted
3. Since there were writes during the time when node A was offline, node A
would has to perform historical rebalance
4. During performing historical rebalance SQL select might be dispatched to be
executed on node A as well, but at this moment node A will return 0 primary row
count for caches under rebalance, so we will initialize table statistics
incorrectly and further will calculate unrealistic costs.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
