Amichai Rothman created SLING-2990:
--------------------------------------
Summary: race condition in scheduler could cause tasks to be lost
Key: SLING-2990
URL: https://issues.apache.org/jira/browse/SLING-2990
Project: Sling
Issue Type: Bug
Components: Commons
Reporter: Amichai Rothman
In QuartzScheduler.java, binding a task/job checks if the scheduler is null and
then either registers the service reference or adds it to a list for later
registration. When the scheduler is activated all items on the list are
registered and the list is cleared.
All accesses to the list are properly synchronized, but the scheduler null
check is not, so they are not done atomically together - there may be a race
condition here that can cause a task to be lost: if the scheduler is checked
and found to be null, then a context switch occurs and the scheduler is
activated and processes the list, and then the original thread resumes and adds
the task to the list, then the task will remain unprocessed in the list and
never be registered.
The same thing might happen inversed with unbinding tasks/jobs.
btw the bind/unbind methods are duplicated for jobs and tasks, so it might be
nicer to extract the common implementation and delegate to it with a job/task
parameter. Then there will be less places to apply the fix in :-)
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
