[OPENJPA-2547] When two threads attempt to get a Pessimistic Lock, one thread gets a 'false' lock.

• Type: Bug
• Status: Closed
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 2.1.2, 2.2.2, 2.2.1.1, 2.3.0, 2.4.0
• Fix Version/s: 2.1.2, 2.2.2, 2.2.1.1, 2.4.0
• Component/s: jdbc
• Assignee: Unassigned
• Reporter: Heath Thomann
• Created: 25/Nov/14 16:58
• Updated: 14/Jan/15 22:49
• Resolved: 14/Jan/15 22:49

Description

I have created a scenario where two threads both attempt to get a pessimistic lock on an object, where one thread legitimately gets the lock, and the other gets a 'false' lock. To describe this issue, lets look at a test. I will provide (attache) a full working test, but for now lets look at this snippet of the test which is at the heart of the issue:

PessimisticLockEntity entity = oem.find(PessimisticLockEntity.class, pKey);

boolean locked = false;
while (!locked) {
try

{ oem.getFetchPlan().setLockTimeout(5000); oem.lock(entity, LockModeType.PESSIMISTIC_READ); locked = true; }

catch (PessimisticLockException ple) {

With this test, imagine the case where two threads call this code at roughly the same time. In this case, one thread should receive a lock, and the other thread should receive a PessimisticLockException (PLE). In my test, this is what happens.....so far so good. However, the thread with the lock (call it T1) then sleeps for a while, and the thread which got the PLE (call it T2) attempts to get the lock over and over again. When T2 tries to get a lock while T1 holds the lock, T2 should continue to receive a PLE. However, I've found that T2 "gets" a lock. That is, a PLE is never thrown because OpenJPA doesn't execute SQL to obtain the lock. OpenJPA skips that steps and thinks T2 has a lock. This gives T2 a false lock.

I've found that prior to JIRA OPENJPA-2449, we got the expected results (i.e. T2 continues to get a PLE while T1 holds the lock). It is only with OPENJPA-2449 that the incorrect behavior occurs.

I will provide a patch which contains a potential fix, and a test that does two things: 1) tests the scenario above and 2) provides a test to verify OPENJPA-2449 (we never provided a test when we committed OPENJPA-2449). My fix, if correct, simply adds one additional check to the code added via OPENJPA-2449.

Thanks,

Heath

Issue Links

• is broken by: OPENJPA-2449 refresh(PESSIMISTIC_WRITE) generates seperate SQL for the lock