[IGNITE-9663] MVCC: Data node failure can cause TX hanging.

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.7
• Component/s: cache, mvcc
• Labels: MakeTeamcityGreenAgain
• Assignee: Roman Kondakov
• Reporter: Andrey Mashenkov
• Created: 21/Sep/18 14:05
• Updated: 23/Oct/18 12:03
• Resolved: 23/Oct/18 12:01

Description

Next test hangs due to a bug:
CacheMvccTransactionsTest.testPutAllGetAll_ClientServer_Backups1_Restart_Scan

Possibly, it is caused with a race in MvccTracker and can be partially fixed in 1 or 
we have a bug in TX mapping in enlist phase on unstable topology.

 

I've tried to avoid races in Tracker and see TX hanged awaiting PME that version has just been completed. This need to be investigated.
 

[1] https://issues.apache.org/jira/browse/IGNITE-9428

Issue Links

• is caused by: IGNITE-7764 MVCC TX: make cache basic operations support Mvcc tx mode.
• is related to: IGNITE-9774 CacheMvccTransactionsTest.testPutAllGetAll_ClientServer_Backups1_Restart_Scan hangs
• relates to: IGNITE-5935 MVCC TX: Tx recovery protocol
• links to: GitHub Pull Request #5018
• links to: TC run