Thomas Watson created FELIX-6140:
------------------------------------
Summary: possible deadlock in ResolverImpl.EnhancedExecutor.await()
Key: FELIX-6140
URL: https://issues.apache.org/jira/browse/FELIX-6140
Project: Felix
Issue Type: Bug
Components: Resolver
Affects Versions: resolver-1.16.0, resolver-2.0.0
Reporter: Thomas Watson
The org.apache.felix.resolver.ResolverImpl.EnhancedExecutor class is used by
the ResolverImpl in order to run tasks in parallel during a resolve process.
When there is work that can happen in parallel the resolver will compute a
number of tasks and dispatch each task to the executor and then await for all
the tasks to complete. To do this the current code uses an AtomicInteger to
count up the number of tasks and then waits for each task to decrement the
count back to zero.
The way the code builds up the tasks recursively leaves it open to incrementing
the count and then getting some system failure. For example an
OutOfMemoryException, or before OSGi R7 we used to have the executor throw a
runtime exception on timeout.
This can make it such that the count will never reach zero and we end up
hanging.
See https://bugs.eclipse.org/bugs/show_bug.cgi?id=547830 and possible fix that
I plan to release soon at https://git.eclipse.org/r/#/c/143026/
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
