[ 
https://issues.apache.org/jira/browse/TOMEE-2108?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16116148#comment-16116148
 ] 

Romain Manni-Bucau commented on TOMEE-2108:
-------------------------------------------

Hi Svetlin,

that's actually what we do: undeploy on failure, think the tck coding to not 
wrap some deployment exception like bval and cdi ones forgot to "keep" the 
destroyApp: 
https://github.com/apache/tomee/blob/master/container/openejb-core/src/main/java/org/apache/openejb/assembler/classic/Assembler.java#L1042

> TomEEContainer (arquillian) does not clean up after failed deployment.
> ----------------------------------------------------------------------
>
>                 Key: TOMEE-2108
>                 URL: https://issues.apache.org/jira/browse/TOMEE-2108
>             Project: TomEE
>          Issue Type: Bug
>            Reporter: Svetlin Zarev
>
> h4. Problem:
> ---
> When the deployment fails, the DeployerEjb throws javax.ejb.EjbException and 
> the "archiveName" is not added  to "moduleIds". As a result when 
> undeploy(Archive) is called, the associated DeployedApp is not found and 
> clean up is not performed -> this results in tomcat's internal structures 
> still keeping data for the failed webapp, which might break other 
> (re)deployments with the same webapp name.
> h4. Steps to reproduce:
> ---
> Execute any of the CDI TCK test that are expected to fail. I tested with 
> (passing.xml):
> {code}
>   <test name="JSR-346 TCK">
>     <classes>
>       <class 
> name="org.jboss.cdi.tck.tests.decorators.definition.broken.nodecoratedtypes.DecoratorWithNoDecoratedTypes1Test"/>
>  
>       <class 
> name="org.jboss.cdi.tck.tests.decorators.definition.broken.nodecoratedtypes.DecoratorWithNoDecoratedTypes2Test"/>
>  
>     </classes>
>   </test>
> {code}
> The classes are deployed sequentally using the same war name. Both 
> deployments will fail as expected. After undeploy is called for teh first 
> one, a WARN message is logged:
> {code}
> VIII 04, 2017 9:59:47 PM org.apache.openejb.arquillian.common.TomEEContainer 
> undeploy
> WARNING: 2fea4e92ef1679e3f8eaf0cc2ce557a56f68cb26.war was not deployed
> {code}
> So far so good, the deployment failed, so we might assume it was not really 
> deployed and dismiss the warning. Unfortunately Tomcat did never perform a 
> clean up for that app, so when the next test is deployed (with the same WAR 
> name) we can see in the logs:
> {code}
> 04-Aug-2017 21:59:47.233 SEVERE [http-nio-35825-exec-5] 
> sun.reflect.NativeMethodAccessorImpl.invoke Creation of the naming context 
> failed: [javax.naming.OperationNotSupportedException: Context is read only]
> {code}
> That's not what we expected to see. This happens because in 
> org.apache.naming.ContextAccessController#setSecurityToken, when the security 
> token is being associated with the context name, it checks only if a token 
> exists for that name, but not if it's thE same token:
> {code}
>     public static void setSecurityToken(Object name, Object token) {
>         SecurityManager sm = System.getSecurityManager();
>         if (sm != null) {
>             sm.checkPermission(new RuntimePermission(
>                     ContextAccessController.class.getName()
>                             + ".setSecurityToken"));
>         }
>         if ((!securityTokens.containsKey(name)) && (token != null)) {
>             securityTokens.put(name, token);
>         }
>     }
> {code}
> And the root cause for that is that no clean up was performed (kind of 
> related to: TOMEE-2098). 
> h4. Proposal:
> ---
> Do clean up if the deployment fails. This would be as simple as adding 
> catch(EjbException) in 
> org.apache.openejb.arquillian.common.TomEEContainer#deploy(org.jboss.shrinkwrap.api.Archive<?>):
> {code}
> } catch (OpenEJBException | EJBException re) { // clean up in undeploy needs 
> it
>     moduleIds.put(archiveName, new DeployedApp(file.getAbsolutePath(), file));
>     re;
> }
> {code}
> Note that I've changed it to use the absolute path, because otherwise the 
> ProvisioningResolver is not able to understand the relative path:
> {code}
> 04-Aug-2017 22:47:42.436 SEVERE [http-nio-42119-exec-9] 
> org.apache.openejb.core.transaction.EjbTransactionUtil.handleSystemException 
> EjbTransactionUtil.handleSystemException: Not a local file but no prefix 
> specified: target/workdir/1/2fea4e92ef1679e3f8eaf0cc2ce557a56f68cb26.war. So 
> can't resolve
>  java.lang.IllegalArgumentException: Not a local file but no prefix 
> specified: target/workdir/1/2fea4e92ef1679e3f8eaf0cc2ce557a56f68cb26.war. So 
> can't resolve
>       at 
> org.apache.openejb.loader.provisining.ProvisioningResolver.protocol(ProvisioningResolver.java:133)
>       at 
> org.apache.openejb.loader.provisining.ProvisioningResolver.realLocation(ProvisioningResolver.java:115)
>       at 
> org.apache.openejb.loader.ProvisioningUtil.realLocation(ProvisioningUtil.java:158)
>       at 
> org.apache.openejb.assembler.classic.Assembler.getAppInfo(Assembler.java:633)
>       at 
> org.apache.openejb.assembler.DeployerEjb.undeploy(DeployerEjb.java:410)
> {code}
> The issue is that the resolver cannot correctly resolve the absolute path 
> from the relative one. It thinks it is:
> {code}
> /home/svetlin/git/open-source/tomee/tck/cdi-tomee/target/tomee/apache-tomee-plus-7.0.4-SNAPSHOT/target/workdir/0/2fea4e92ef1679e3f8eaf0cc2ce557a56f68cb26.war
> {code}
> While it really is:
> {code}
> /home/svetlin/git/open-source/tomee/tck/cdi-tomee/target/workdir/0/2fea4e92ef1679e3f8eaf0cc2ce557a56f68cb26.war
> {code}
> h4. Summary
> ---
> *  when the deployment fails, no clean up is done
> *  do clean up after failed deplyments
> **  catch javax.ejb.EjbException besides OpenEjbException
> **  Use absolute path for DeployedApp.path in case of failed deployment
> What do you think ? 



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Reply via email to