[ARTEMIS-2347] JournalStorageManager::stopReplication can deadlock while stopping

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.8.1
• Fix Version/s: 2.9.0
• Component/s: Broker
• Assignee: Unassigned
• Reporter: Francesco Nigro
• Created: 21/May/19 15:11
• Updated: 29/May/19 19:17
• Resolved: 29/May/19 19:17

Description

JournalStorageManager::stopReplication needs to:

1. acquires the manager write lock
2. acquires (if any) large message intrinsic locks during performCachedLargeMessageDeletes
3. acquires the manager read lock on confirmPendingLargeMessage

JournalStorageManager::stop needs to:

1. acquires (if any) large message intrinsic locks during performCachedLargeMessageDeletes
2. acquires the manager read lock on confirmPendingLargeMessage

A racing call to JournalStorageManager::stopReplication while stopping could deadlock
the broker:

1. JournalStorageManager::stop acquires a large message intrinsic locks during performCachedLargeMessageDeletes
2. JournalStorageManager::stopReplication acquires the manager write lock
3. JournalStorageManager::stop wait is blocked on the manager write lock before releasing the lock on large message
4. JournalStorageManager::stopReplication is blocked on the large message intrinsic lock before releasing the manager write lock
5. deadlock occurred: none can proceed

Issue Links

• links to: GitHub Pull Request #2675