This Bugzilla instance is a read-only archive of historic NetBeans bug reports. To report a bug in NetBeans please follow the project's instructions for reporting issues.
|Summary:||[2003-10-22]Deadlock during startup|
|Product:||javaee||Reporter:||Jan Lahoda <jlahoda>|
|Component:||HTTP Monitor||Assignee:||Ana.von Klopp <avk>|
|Issue Type:||DEFECT||Exception Reporter:|
|Bug Depends on:|
Two full thread dumps of the deadlock.
Links to diffs
Description Jan Lahoda 2003-10-22 13:48:35 UTC
WS-v2 build 031022, JDK1.4.2. Deadlock occurs sometimes during startup. The userdir was the same as used for issue #36745. I have archived it, so I can attach it if necessary. The full thread dumps are attached to this issue. Special info: first time the deadlock occured, I tested -XX:CompileThreshold=2 setting (the deadlock later occured even without this setting).
Comment 1 Jan Lahoda 2003-10-22 13:50:00 UTC
Created attachment 11936 [details] Two full thread dumps of the deadlock.
Comment 2 Peter Zavadsky 2003-10-22 14:04:20 UTC
Note: It in new winsys build. Assigning to Yarda. Please have a look at it and assign where the problem is.
Comment 3 Jaroslav Tulach 2003-10-22 16:06:43 UTC
It could be possible to break the deadlock by not creating node structure in constructor of MonitorAction or by not instantiating monitor action or the controler at all.
Comment 4 Ana.von Klopp 2003-11-13 17:56:43 UTC
I will look into changing the behaviour.
Comment 5 Ana.von Klopp 2003-11-14 00:25:20 UTC
I looked into the problem and realized that Jesse Glick had modified the files that load the transactions a couple of months ago with respect to how the nodes are created. These modifications introduced the Mutex code which apparently causes the hang now. There are numerous other problems associated with this change (which ought to have been obvious if the component had been tested at all afterwards - for example saving did not work). I'm backing out these changes for now (the component will be rewritten to use Looks as soon as I have time). I removed the creation of the nodes when the action is created anyway, since it's not necessary to do it on startup.
Comment 7 Jesse Glick 2003-11-14 11:22:55 UTC
May need to be re-fixed as part of issue #35833, TBD - the use of Mutex.EVENT was preventing critical non-thread-safe Node methods from being called from arbitrary threads due to abuse of Node/Children as a data model. Underlying problem will need some fix for promo-B.
Comment 8 L Martinek 2004-02-23 11:58:47 UTC