[
https://issues.apache.org/jira/browse/CXF-2398?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12746074#action_12746074
]
William Tam edited comment on CXF-2398 at 8/21/09 9:54 AM:
-----------------------------------------------------------
While the new patch CXF-2398.patch2 addresses the problem (of prematurely
destroying clients) introduced CXF-2398.patch, the new patch also raises new
issues/questions.
* Creating more than one client proxy per
ManagedConnectionImpl.getConnection(subject, spec) method call is needed if we
are going to support Jee credential mapping. Currently, the subject is only
saved in the handles map for reporting the subject of the associated handle in
metadata (the new patch will break this, btw) but the subject could be used as
credential to create a client proxy connection handle in the future. FYI, If
you look at ManagedConnectionFactoryImpl.matchManagedConnections(), it
intentionally skips the subject as a matching criteria (so the application
server would call ManagedConnectionImpl.getConnection(subject, spec) with
different subjects). I'd recommend not to change it.
* I think what prompted the change was the observation of too many proxies
getting created. It is a problem in the
ManagedConnectionImpl.getConnection(subject, spec) method that does not check
the subject. We should not create new proxy if there is one already created
for a given subject. We could so something like this in getConnection() to fix
it.
{code}
Object handle = null;
for (Map.Entry<Object, Subject> entry : handles.entrySet()) {
if (subject == null) {
if (entry.getValue() == null) {
handle = entry.getKey();
break;
}
} else {
if (subject.equals(entry.getValue())) {
handle = entry.getKey();
break;
}
}
}
// only create a new handle, if hande is null
{code}
* I think it is fine to have the ManagedConnectionImpl.destroy() to destroy
client proxies. What about the cleanup() method? The JCA spec 6.5.4.4 says we
should invalidate all handles but at the same time it says cleanup() should not
close physical pipe. It does not fit well in our model because we invalidate
handle by destroying client (thus, closing connection). IMO, it is fine for
the cleanup() method to invoke client destroy(). Right now, we are not
invalidating handles. So, if you take my advice in the first bullet, we would
move the closed handle from handles map to another collection (say,
closedHandles) in handleCloseMethod().
{code}
handles.remove(proxy);
closedHandles.add(proxy);
{code}
When destroying the proxies, we just need to make sure we destroy the
closedHandles as well.
was (Author: wtam):
While the new patch CXF-2398.patch2 addresses the problem (of prematurely
destroying clients) introduced CXF-2398.patch, the new patch also raises new
issues/questions.
* Creating more than one client proxy per
ManagedConnectionImpl.getConnection(subject, spec) method call is needed if we
are going to support Jee credential mapping. Currently, the subject is only
saved in the handles map for reporting the subject of the associated handle in
metadata (the new patch will break this, btw) but the subject could be used as
credential to create a client proxy connection handle in the future. FYI, If
you look at ManagedConnectionFactoryImpl.matchManagedConnections(), it
intentionally skips the subject as a matching criteria (so the application
server would call ManagedConnectionImpl.getConnection(subject, spec) with
different subjects). I'd recommend not to change it.
* I think what prompted the change was the observation of too many proxies
getting created. It is a problem in the
ManagedConnectionImpl.getConnection(subject, spec) method that does not check
the subject. We should not create new proxy if there is one already created
for a given subject. We could so something like this in getConnection() to fix
it.
{code}
Object handle = null;
for (Map.Entry<Object, Subject> entry : handles.entrySet()) {
if (subject == null) {
if (entry.getValue() == null) {
handle = entry.getKey();
break;
}
} else {
if (subject.equals(entry.getValue())) {
handle = entry.getKey();
break;
}
}
}
// only create a new handle, if hande is null
{code}
* I think it is fine to have the ManagedConnectionImpl.destroy() to destroy
client proxies. What about the cleanup() method? The JCA spec 6.5.4.4 says we
should invalidate all handles but at the same time it says cleanup() should not
close physical pipe. It does not fit well in our model because we invalidate
handle by destroying client (thus, closing connection). IMO, it is fine for
the cleanup() method to invoke client destroy(). Right now, we are not
invalidating handles. So, if you take my advice in the first bullet, we would
move the closed handle from handle map to another collection (say,
closedHandle) in handleCloseMethod().
{code}
handles.remove(proxy);
closedHandles.add(proxy);
{code}
When destroying the proxies, we just need to make sure we destroy the
closedHandles as well.
> JMS Connections are not closed when JCA recycles CXF proxy
> ----------------------------------------------------------
>
> Key: CXF-2398
> URL: https://issues.apache.org/jira/browse/CXF-2398
> Project: CXF
> Issue Type: Bug
> Components: Integration
> Affects Versions: 2.1
> Reporter: Seumas Soltysik
> Assignee: Daniel Kulp
> Fix For: 2.1.7, 2.2.4
>
> Attachments: CXF-2398.patch, CXF-2398.patch2
>
>
> When close() is called on JCA ConnectionHandler, there is no cleanup of the
> underlying client proxy and its transport. Specfically when the JCA outbound
> use case is using JMS, JMS connections are not being closed.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
