I actually discovered the same by easily reproducing the issue and just
checked in a fix. See
http://java.net/jira/browse/GLASSFISH-19116
for details. You are spot on with your analysis as well. You may like to
port this change to your workspace and let me know what you see.
Thanks,
Sahoo
On Saturday 29 September 2012 01:00 PM, alex Woo wrote:
> Sahoo,
>
> Thanks for your explanation.
>
> > I see webservices-extra-jdk-packages.jar, which is a framework
> extension bundle. So, surely we should be entering the
> > second if block as well.
>
> please note that, the value of 'systemBundleUpdationRequired' is
> just updated in BundleProvisioner#update(...) but not in
> BundleProvisioner#uninstall(...).
>
> this maybe the reason.
>
> Thanks
>
> Alex
>
>
>
> 2012/9/29 Sahoo <sanjeeb.sahoo_at_oracle.com
> <mailto:sanjeeb.sahoo_at_oracle.com>>
>
> Alex,
>
> As you must have figured out, if any bundle in modules/ is
> changes, hasAnythingChanged() returns true. But, not all kind of
> change requires a framework update. Only when framework extension
> bundles are involved, we need to update the framework and that's
> why we call bundleProvisioner.setBundleContext() in the second if
> block only. If framework is not updated, its bundle context
> remains valid, so there is no need to call
> bundleProvisioner.setBundleContext() in first if block.
>
> Looking at the list of bundles getting uninstalled as sent by you
> in your first email, I see webservices-extra-jdk-packages.jar,
> which is a framework extension bundle. So, surely we should be
> entering the second if block as well. Let me see if I can
> reproduce the issue locally.
>
> Thanks,
> Sahoo
>
>
> On Saturday 29 September 2012 09:19 AM, alex Woo wrote:
>> Sahoo,
>>
>> Thanks for you reply. Yes I am using GF 3.1.2.2. I merged the
>> code from trunk and build, but same error still occured.
>>
>> I debuged and found that it didn't enter the if block [if
>> (bundleProvisioner.isSystemBundleUpdationRequired()) ], but
>> entered into [ if (bundleProvisioner.hasAnyThingChanged())] .
>>
>> if (*bundleProvisioner.hasAnyThingChanged()*) {
>> bundleProvisioner.refresh();
>> deleteHK2Cache(); // clean hk2 cache so that
>> updated bundle details will go in there.
>> // Save the bundle ids for use during restart.
>> storeBundleIds(bundleIds.toArray(new
>> Long[bundleIds.size()]));
>> }
>> if
>> (*bundleProvisioner.isSystemBundleUpdationRequired()*) {
>> logger.logp(Level.INFO, "OSGiFrameworkLauncher",
>> "launchOSGiFrameWork", "Updating system bundle");
>> framework.update();
>> framework.waitForStop(0);
>> framework.init();
>>
>> bundleProvisioner.setBundleContext(framework.getBundleContext());
>> }
>>
>> Adding
>> [... bundleProvisioner.setBundleContext(framework.getBundleContext());]
>> in the second if block really fix this issue?
>> After bundleProvisioner#refresh() was called, is there no need
>> to do something to ensure the refresh() really compele?
>>
>> Thanks
>>
>> Alex
>>
>>
>> 2012/9/29 Sahoo <sanjeeb.sahoo_at_oracle.com
>> <mailto:sanjeeb.sahoo_at_oracle.com>>
>>
>> Alex,
>>
>> From the line numbers in the stacktrace reported by you, I
>> see you are using gf 3.1.2. You must have looked at gf 3.1.2
>> source then. There used to be an issue in 3.1.x branch, but
>> it has been fixed in trunk. In trunk, we now do the following:
>>
>> if
>> (bundleProvisioner.isSystemBundleUpdationRequired()) {
>> logger.logp(Level.INFO,
>> "OSGiFrameworkLauncher", "launchOSGiFrameWork", "Updating
>> system bundle");
>> framework.update();
>> framework.waitForStop(0);
>> framework.init();
>> *
>> bundleProvisioner.setBundleContext(framework.getBundleContext());*
>> }
>>
>> We do this specifically to avoid the IllegalStateException.
>>
>> Thanks,
>> Sahoo
>>
>> On Friday 28 September 2012 08:02 PM, alex Woo wrote:
>>> I saw the source code both GF and Felix, found that the
>>> reason my be this:
>>>
>>> When new/updated/removed bundles during subsequent
>>> restart, BundleProvisioner#refresh() was called,
>>>
>>> and will eventually call Felix#update() which will stop and
>>> restart the framework on a separated thread,
>>>
>>> after Felix#stop() was called, the BundleContext come to
>>> invalidate.
>>>
>>> so, BundleProvisioner#startBundles() Failure..
>>>
>>>
>>> thanks
>>>
>>> Alex
>>>
>>> 2012/9/28 Sahoo <sanjeeb.sahoo_at_oracle.com
>>> <mailto:sanjeeb.sahoo_at_oracle.com>>
>>>
>>> [ the original submitter was missing - adding him]
>>>
>>> On Friday 28 September 2012 06:44 PM, Sahoo wrote:
>>>> On Friday 28 September 2012 06:40 PM, JJ Snyder wrote:
>>>>> Let's try this one more time...
>>>>> Whenever I modify anything in modules I always wipe
>>>>> clean the osgi-cache directory. Stop GF, go to
>>>>> /glassfish3/glassfish/domains/domain1/osgi-cache and
>>>>> remove all files and dirctories, then restart GF.
>>>>>
>>>>> JJ
>>>> This is actually *not* required. I rarely clean the
>>>> cache. GF provisioner is coded such that it takes care
>>>> of new/updated/removed bundles during subsequent
>>>> restart. I need more info from user to see what's
>>>> happening. It could be a bug.
>>>>
>>>> Thanks,
>>>> Sahoo
>>>
>>>
>>
>>
>
>