CDV ❯ fairness policy for ReentrantLock and ReentrantReadWriteLock
-
Bug
-
Status: Open
-
2 Major
-
Resolution:
-
Byte Code Transform
-
-
eng group
-
Reporter: teck
-
December 21, 2007
-
1
-
Watchers: 0
-
December 09, 2009
-
Description
ReentrantLock and ReentrantReadWriteLock both have constructors that let an application decide if they want a fair or non-fair lock policy (the default is non-fair),
The terracotta instrumentation changes this and always forces the fair policy no matter what the application passes on the constructor.
For non-shared locks instances, this seems pretty odd (especially since we’re forcing the non-default)
Comments
Chris Dennis 2009-12-09
Chris Dennis 2009-12-09
For the record the failures are due to this known “bug” (or bad specification) in the JDK 1.5 RRWL:
http://markmail.org/message/jlxczl4ilrt43kz3 http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6315709
This behavior has actually caused failures in the monkeys, as we use the instrumented forced fair RRWL in the L1 code, it actually caused a deadlock in the new lock manager code, I’m going to code a workaround for the lock manager, but I’m very keen on seeing this fixed.
Log –> http://kong.terracotta.lan/output/archive/trunk/rev14172/cruisecontrol-log-trunk-alltests-rev14172-rh5mo2-20091208-2147.log.txt.gz Archive –> http://kong.terracotta.lan/output/archive/trunk/rev14172/build-archive-trunk-alltests-rev14172-rh5mo2-20091208-2147.zip