| Re: [ecf-dev] Race conditions when exposing services? |
|
Hi Pablo,
Thanks for the report. This is an interesting problem...and I would like to enlist your help in diagnosing. I've read over both your email and your stackoverflow posting and have a couple of hypotheses about what's going on...as well as some questions that might give me a clearer idea of what's going wrong. I'm hopeful that with a little investigation we can diagnose and fix what's going wrong here. I think it might be best, though, to create a new bug on which to have the discussion, q and a, etc. Would you be willing to create a new bug to record the info that you've collected? Here is the ECF bug category: https://bugs.eclipse.org/bugs/enter_bug.cgi?product=ECF I know this is some additional work for you, but it would be better for you to create the bug and put the info you've collected there, rather than me or another committer...since we haven't yet actually experienced it in test or other code. A couple of initial thoughts/hypotheses/questions: 1) It seems possible to me that there is some timing problem with discovery (zookeeper in your case). Have you tried turning on the Zookeeper debug/tracing info? 2) Another possibility is that in some startup cases, the export of the remote services is not happening fast enough for ds (i.e. within the default 30s). It's very strange to me that it reports the equinox.event as the one being waited on however (the equinox.event is indeed used by the ECF RSA implementation...for import, however). Could you describe on the bug what service topology you have (i.e. which services are being exported by which nodes, and which services are being imported by which nodes). WRT this question: <stuff deleted> Is there any way to "refresh/force" the service
annoucement/search? (or avoid the warning message, which is the
thing that is affecting the services, probably).
[Scott] Yes. The OSGi RSA specification makes it possible to explicitly control the export and import of remote services, by creating and using a custom 'topology manager'. There's some info on this here: http://wiki.eclipse.org/Remote_Services_Admin And much more (of course) in the OSGi RSA spec. I'm hoping for this use case that won't be necessary, however (i.e. you can/could continue to use the ECF default topology manager), but if that becomes necessary it is an option. ECF's implementation of the topology manger can easily be extended/replaced...again if that proves to be necessary. But let's investigate this issue closely on this new bug first...as the issue you are seeing does look timing related...and may have more to do with a bug than with a need for a replacement topology manager. Thanks, Scott This issue is happening randomly in all the four
nodes of my network.
Thanks in advance!
[1] Message:
WARNING 26 [SCR - WorkThread] Timeout occurred! Thread was
blocked on processing [QueuedJob] WorkPerformer:
org.eclipse.equinox.internal.ds.SCRManager@19336006;
actionType 1
WARNING 26 [SCR] Enabling components of bundle
org.eclipse.equinox.event did not complete in 30000 ms
!SESSION 2013-01-14 20:46:11.168
-----------------------------------------------
eclipse.buildId=unknown
java.version=1.6.0_24
java.vendor=Sun Microsystems Inc.
BootLoader constants: OS=linux, ARCH=x86_64, WS=gtk,
NL=es_ES
Framework arguments: -vmargs
Command-line arguments: -vmargs -consoleLog -console
-configuration configuration -clean
!ENTRY org.eclipse.equinox.ds 2 0 2013-01-14 20:46:42.134
!MESSAGE [SCR] Enabling components of bundle
org.eclipse.equinox.event did not complete in 30000 ms
!ENTRY org.eclipse.equinox.ds 2 0 2013-01-14 20:46:42.135
!MESSAGE [SCR - WorkThread] Timeout occurred! Thread was
blocked on processing [QueuedJob] WorkPerformer:
org.eclipse.equinox.internal.ds.SCRManager@19336006;
actionType 1
Pablo García Sánchez Departamento de Arquitectura y Tecnología de Computadores Universidad de Granada http://geneura.ugr.es/~pgarcia
|