[HDDS-3118] Possible deadlock in LockManager

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 0.5.0
• Component/s: Ozone Manager
• Labels: pull-request-available
• Target Version/s: 0.5.0
• Assignee: Bharat Viswanadham
• Reporter: Attila Doroszlai
• Created: 02/Mar/20 21:52
• Updated: 10/Mar/20 15:36
• Resolved: 04/Mar/20 09:43

Description

LockManager has a possible deadlock.

1. Number of locks is limited by using a GenericObjectPool. If N locks are already acquired, new requestors need to wait. This wait in getLockForLocking happens in a callback executed from ConcurrentHashMap#compute while holding a lock on a map entry.
2. While releasing a lock, decrementActiveLockCount implicitly requires a lock on an entry in ConcurrentHashMap.

Issue Links

• links to: GitHub Pull Request #627
• links to: GitHub Pull Request #647