[HIVE-20773] Query result cache might contain stale MV data

• Type: Bug
• Status: Open
• Priority: Critical
• Resolution: Unresolved
• Affects Version/s: 4.0.0
• Component/s: Materialized views
• Target Version/s: 4.0.0
• Assignee: Unassigned
• Reporter: Oliver Draese
• Created: 18/Oct/18 22:30
• Updated: 23/Jun/21 19:43

Description

A user can specify “required freshness” of data in a materialized view via

hive.materializedview.rewriting.time.window=<time>

 

If a query is executed against stale data, its result will be cached by HiveServer2 to avoid re - execution at a later point in time. This cache will now contain stale data. A different query session might not be able to tolerate stale data or have a shorter acceptable time window. Right now, it will receive the query result cache content independent of its own hive.materializedview.rewriting.time.window setting, potentially leading to unexpected results.