[seam-issues] [JBoss JIRA] Commented: (SEAMCRON-13) Try workaround for WELD-862 using deep copy

[ https://issues.jboss.org/browse/SEAMCRON-13?page=com.atlassian.jira.plugi... ] Peter Royle commented on SEAMCRON-13: ------------------------------------- Another portable solution suggested by Sturat Douglas: (04:37:00 PM) stuartdouglas: and I had an idea for a protable way to implement it (04:37:09 PM) PeteRoyle: awesome (04:37:49 PM) stuartdouglas: the basic idea is use a threadlocal to control when then interceptor is actually run (04:38:17 PM) stuartdouglas: and then look up and re-invoke on the bean in a new thread (04:38:43 PM) stuartdouglas: as long as the @Async interceptor is first, this gives pretty much the same result (04:39:07 PM) stuartdouglas: as long as the bean being invoked on is normal scoped (04:40:25 PM) daniel_hinojosa [~danno(a)c-68-35-182-153.hsd1.nm.comcast.net] entered the room. (04:40:55 PM) PeteRoyle: I don't follow all the way. So are there multiple interceptors? (04:41:48 PM) stuartdouglas: the basic idea is to re-run the whole invocation in a new thread, but have a thread local set that will stop the @Asyn interceptor from firing again (04:42:28 PM) PeteRoyle: ok I'm getting warmer I think (04:42:51 PM) stuartdouglas: something that did just occur to me is that weld will probably not work very well in this new thread anyway, unless you set the TCCL to the correct one (04:43:07 PM) PeteRoyle: TCCL? (04:43:16 PM) stuartdouglas: Thread Context Class Loader (04:43:28 PM) stuartdouglas: weld uses it a lot, even for stuff that should not need it to be set (04:43:51 PM) stuartdouglas: so if it is not set you end up with highly informative 'Singleton is not set' errors (04:44:30 PM) PeteRoyle: Ahah! I've been getting those sporadically on redeployment (04:44:45 PM) PeteRoyle: But to do with scheduling, not asynch (04:44:49 PM) PeteRoyle: (so far) (04:45:20 PM) stuartdouglas: those are pretty much always due to the wrong class loader being set as the TCCL (04:45:54 PM) stuartdouglas: also the only beans that will be accessible is @ApplicationScoped and @Dependant (04:46:10 PM) stuartdouglas: unless you put in some non-portable code to set up the contexts (04:46:11 PM) PeteRoyle: Is there a typical cure for that which doesn (04:46:29 PM) PeteRoyle: ''t require knowledge of how classloading works? (04:47:34 PM) stuartdouglas: not really, it depends where you get the errors, it should only be a problem when you start trying to use threads that you have spawned yourself to interact with weld (04:48:31 PM) stuartdouglas: for @Async the solution is to set the TCCL to the same as the original threads TCCL, and set it back in a finally block (04:49:11 PM) stuartdouglas: and it should really be run as a PriviliedAction, because a security manager might not let you set it (04:50:22 PM) marekn [~mnovotny@nat/redhat/x-wgsrrhabedyogdbb] entered the room. (04:50:39 PM) PeteRoyle: I get what you're saying, but I'm not sure how to do that stuff (04:50:54 PM) PeteRoyle: (setting the TCCL and running as PrivAction) (04:51:57 PM) stuartdouglas: https://github.com/stuartwdouglas/jboss-as/blob/master/weld/src/main/java... (04:52:11 PM) stuartdouglas: https://github.com/stuartwdouglas/jboss-as/blob/master/weld/src/main/java... (04:52:43 PM) stuartdouglas: note that SecurityActions is package private, otherwise any class can use it to set the TCCL (04:53:42 PM) stuartdouglas: basically it just wraps Thread.currentThread().set/getContextClassLoader (04:54:38 PM) PeteRoyle: Should I copy this code into Cron, (or Solder) to make it portable? (04:55:09 PM) stuartdouglas: it needs to be copied each time (04:55:38 PM) stuartdouglas: otherwise any code can get/set the TCCL (04:56:02 PM) kevinpollet [~kevinpoll(a)217.112.54.72] entered the room. (04:56:56 PM) PeteRoyle: OK so do I set the TCCL as the first thing in the new thread? (04:56:59 PM) amitev [~amitev(a)212.25.36.84] entered the room. (04:57:06 PM) stuartdouglas: yes (04:57:11 PM) PeteRoyle: (frm within the thread, say inside the run() method)? (04:57:17 PM) stuartdouglas: yes (04:57:34 PM) PeteRoyle: and unset it from a finally block from the original thread which screate the new thread (04:57:52 PM) PeteRoyle: screate=created (04:58:21 PM) stuartdouglas: yes, although you can just set it to null rather than saving and restoring the existing TCCL (04:59:01 PM) stuartdouglas: otherwise if you are using a thread pool the TCCL can hang around after redeployment, which results in a memory leak (04:59:47 PM) PeteRoyle: actually it looks like all the examples in WeldContainer do both those things in the same thread. Can I unset the TCCL at the end of the run() method (in a finally)? (04:59:58 PM) maschmid [~maschmid@nat/redhat/x-fjnivohxvbmhnpfw] entered the room. (05:00:00 PM) stuartdouglas: yes (05:00:06 PM) PeteRoyle: ok (05:00:33 PM) stuartdouglas: oops, sorry, I mis-read your earlier question (05:00:50 PM) stuartdouglas: you should only manipulate it inside the run method (05:00:57 PM) PeteRoyle: ok gotchya (05:01:47 PM) PeteRoyle: So about this thread local. What exactly will it be storing? (05:02:16 PM) PeteRoyle: Just any kind of marker? (05:02:26 PM) PeteRoyle: Boolean (05:02:30 PM) stuartdouglas: just a boolean (05:03:11 PM) PeteRoyle: ok so if it's false I set it and start the thread. If it's true I return (05:03:29 PM) stuartdouglas: if it is true set it to false and then proceed (05:03:39 PM) PeteRoyle: ctx.proceed()? (05:03:43 PM) stuartdouglas: that will allow additional @Asyn invocations for the @Async thread to run (05:03:46 PM) stuartdouglas: yes -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira