[IGNITE-9290] Make remove explicit locks async when node left.

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 2.8
• Component/s: cache
• Labels: deadlock
• Assignee: Andrey Mashenkov
• Reporter: Andrey Mashenkov
• Created: 16/Aug/18 08:08
• Updated: 04/Feb/19 10:19
• Resolved: 11/Jan/19 14:51

Description

GridCacheMvccManager.removeExplicitNodeLocks() run synchronously in discovery and exchange threads. This introduce unnecessary delays in discovery and exchange process.

Also, this may cause a deadlock on node stop if user transaction holds an entry lock and awaits some Ignite manager response (e.g. cache store or DR or CQ), as manager stops right after last exchange has finished so managers can't detect node is stopping. 

 

[1] http://apache-ignite-developers.2346864.n4.nabble.com/Synchronous-tx-entries-unlocking-in-discovery-exchange-threads-td33827.html 

Issue Links

• causes: IGNITE-9053 testReentrantLockConstantTopologyChangeNonFailoverSafe can hang in case of broken tx
• is duplicated by: IGNITE-10511 disco-event-worker can be deadlocked by BinaryContext.metadata running is sys striped pool waiting for cache entry lock
• links to: GitHub Pull Request #4556
• links to: GitHub Pull Request #4556
• links to: TC run results