Maxwait exceeded causes pool capacity leak

• Key: POOL-144
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Major
• Assignee: Unassigned
• Reporter: Phil Steitz
• Affects Version/s: 1.5
• Fix Version/s: 1.5.1

Description

When exhausted action is set to WHEN_EXHAUSTED_BLOCK, maxwait is positive and client threads time out waiting for idle objects, capacity can be "leaked" from GenericObjectPools and GeneritCkeyedObjectPools.

Test case and patch for GOP attached.

Comments

[Phil Steitz]

Looking at the patch again and the same pattern in case WHEN_EXHAUSTED_FAIL, I am concerned that while the exception-generating thread waits to acquires the lock on the pool to remove the latch, another thread could sneak in and do something (e.g. return and object) that triggers an allocate, which will leak an object onto the soon-to-be-abandoned latch.

[Mark Thomas]

I agree with your analysis. I'd also add that similar problems can occur in each of the case statements. When a thread enters a sync block, there needs to be a check that another thread hasn't allocated an object whilst the current thread was waiting for a lock.

I have a patch for GOP and GKOP. I'll commit it shortly.

[Mark Thomas]

Fixed in trunk. Thanks for catching this.

Fixed in 1.5.1 release.