Community
Participate
Working Groups
Zookeeper is using a node path in this form 'root/child/grandchild'. We have used the ECF GUID class to get an unique node. However, we have discovered (haha) that the GUID sometimes contains slashes. This is not according to the Zookeeper specs and will result in services not being published. We have a patch.
Hi Wim. If the patch can be applied in next couple of days and tested and reviewed by a couple of committers before next Saturday 5/5, we should be able to get into RC4+1 (Monday, 5/7).
Setting target milestone to 3.3.0. I hope we can get this in for RC4+1.
Created attachment 171092 [details] Patch - See description Here is a patch that fixes it along with other instabilities with the failing test on ecf2. I have setup a stresstest where every 15 seconds a service is activated by DS, picked up by Zookeeper and delivered to consumers. The host and three consumers are running already for three days non-stop and they work as expected. From the stack trace, the following method was behind the failing hudson-zoodiscovery test org.eclipse.ecf.provider.zookeeper.core.ZooDiscoveryContainer.getServices(ZooDiscoveryContainer.java:384) Ahmed changed it so that it never returns a null but an empty collection instead even if the watchManager is null. Ahmed also changed method NodeReader.dispose() because it has the potential of throwing a null pointer exception. Some unnecessary/error prone threads/ static fields were taken care of.
(In reply to comment #3) > Created an attachment (id=171092) [details] > Patch - See description > > Here is a patch that fixes it along with other instabilities with the failing > test on ecf2. I have setup a stresstest where every 15 seconds a service is > activated by DS, picked up by Zookeeper and delivered to consumers. The host > and three consumers are running already for three days non-stop and they work > as expected. > > From the stack trace, the following method was behind the failing > hudson-zoodiscovery test > org.eclipse.ecf.provider.zookeeper.core.ZooDiscoveryContainer.getServices(ZooDiscoveryContainer.java:384) > > Ahmed changed it so that it never returns a null but an empty collection > instead even if the watchManager is null. > Ahmed also changed method NodeReader.dispose() because it has the potential of > throwing a null pointer exception. > Some unnecessary/error prone threads/ static fields were taken care of. Thanks Wim. I would like to get this patch reviewed, approved, and released to head by end of day tomorrow (Sat 6/5), for RC4+1 build (Sunday evening pacific time). I will try to review this later today/Friday and give my +1. I would like Markus to review as well...and we have Wim's review of course.
Hi, +1 just one issue: - org.eclipse.ecf.provider.zookeeper.DiscoveryActivator.start(...).new ServiceTracker() {...}.removedService(ServiceReference, Object) has to call super.removedService(...) in org.osgi.util.tracker.ServiceTracker.removedService(ServiceReference, Object) And please for patches this close to a release only change what is strictly necessary to fix the bug (especially formatting). Makes it a lot easier for us reviewer. :-)
Hi, Thanks for the review. And sorry for the work, you know how it goes... I have added the super call.
+1 on patch 171092. Thanks for the quick work and testing. Please release to HEAD and we'll introduce for rc4+1 build 5/6.
released to HEAD
