Failed to acquire the pool semaphore in empty pool

[garrydias]

Hello... i'm facing a problem when looking up my bean in test phase.
To stress this bean, I'm using a JMeter test case with 100 simultaneously threads. Each thread perform an EJB lookUp for myWorkerBean defined in my applicationContext.xml as follows

<jee:local-slsb lookup-home-on-startup="false" id="myWorkerBean" business-interface="MyWorkerInterface"
        jndi-name="MyWorkerBean/local" />

Each looked up instance performs a call to MyWorkerInterface.execute(WorkRequest) method that slleps around 60.000 mills. But only first 30 lookups will perform succesfully and the remaining 70 will fail with "Fail to aquire pool semaphore, strictTimeout=-1" error because my bean class is annotated with strict max pool size = 30 and will be discarded.
At this moment, I have 30 bean instances filling the pool during 60 seconds and if I check the MBean for MyWorkerBean in jmx-console shows AvailableCount attribute equals 30.

After 60 seconds all 30 instances perform succesfully. Then, I run 100 more jmeter threads again, but this time all lookups blows with "Fail to aquire pool semaphore, strictTimeout=-1" and the MBean for MyWorkerBean in jmx-console shows AvailableCount equals 0 (but it should mark 30 again). What it means: for some reason that first 30 bean instances were not released.

See the code:

public interface MyWorkerInterface
{
    public WorkResponse execute(WorkRequest request);
}

@javax.jws.WebService(portName = "MyWorkerWSPort", name = "MyWorker", serviceName = "MyWorker")
public class MyWorkerWS extends SpringSupport implements MyWorkerInterface {
 
    @javax.jws.WebMethod(operationName = "execute")
    @javax.jws.soap.SOAPBinding(parameterStyle = ParameterStyle.BARE)
    @javax.jws.WebResult(name = "workResponse", partName = "workResponse", targetNamespace = "http://ws.working.my.com/")
    public WorkResponse execute(
        @javax.jws.WebParam(mode = Mode.IN, name = "work", partName = "work") 
        WorkRequest request) 
    {
 
        final int startingDepth = NDC.getDepth();
        WorkResponse response = null;
        try {
            //Here I'm looking up my bean via Spring
            MyWorkerInterface worker = (MyWorkerInterface) super.getBeanFactory().getBean("myWorkerBean");
 
            //Performs around 60.000ms
            response = worker.execute(request);            
        } catch (org.springframework.beans.BeansException beansException) {
            LoggerHelper.error("Application error", beansException);
        } catch (EJBException ejbException) {
 
            final String errorMessage = 
                org.apache.commons.lang.StringUtils.defaultString(ejbException.getMessage());
 
            if (errorMessage.startsWith("Failed to acquire the pool semaphore")) {
                // MyWorkerInterface.execute(WorkRequest) method is delayed for more than 20000ms.
                // When the execute method backs to its normal execution time this EJBException is still happening
                // but the pool seems to be empty
                LoggerHelper.error("Pool is full because MyWorkerInterface.execute(WorkRequest) is" +
                        "taking more time than normal to perform.", ejbException);
            } 
            else {
                LoggerHelper.error("ejb error", ejbException);
            }
        }
        return response;
    }
}

@Clustered
@Local(MyWorkerInterface.class)
@Stateless(name = "MyWorkerBean")
@Interceptors(value = { SpringBeanAutowiringInterceptor.class, LoggingInterceptor.class })
@RolesAllowed(value = "client")
@Pool(value = PoolDefaults.POOL_IMPLEMENTATION_STRICTMAX, maxSize = 30)
public class MyWorkerBean implements MyWorkerInterface {
 
    @RolesAllowed("client")
    @TransactionAttribute(TransactionAttributeType.NOT_SUPPORTED)
    public WorkResponse execute(final WorkRequest workRequest) {
 
        long start = System.currentTimeMillis();
 
        WorkExchange exchange = new WorkExchange();
        exchange.setWorkRequest(workRequest);
 
         //Please ignore try/catch
        Thread.sleep(60000);
 
        LoggerUtil.info("Operation time: " + (System.currentTimeMillis() - start));
 
        // WorkExchange.workResponse have been set by MDB
        WorkResponse response = exhange.getWorkResponse();
 
        return response;
    }
}

public class SpringSupport {
 
    protected final org.apache.log4j.Logger logger = 
        org.apache.log4j.Logger.getLogger(this.getClass());
 
    @javax.annotation.Resource
    private javax.xml.ws.WebServiceContext webServiceContext;
 
    private org.springframework.context.ApplicationContext cachedApplicationContext;
 
    protected org.springframework.beans.factory.BeanFactory getBeanFactory() {
        if (cachedApplicationContext != null) {
            return cachedApplicationContext;
        }
        if (webServiceContext == null) {
            return null;
        }
        javax.xml.ws.handler.MessageContext messageContext = webServiceContext.getMessageContext();
        if (messageContext == null) {
            return null;
        }
        javax.servlet.ServletContext servletContext = 
            (javax.servlet.ServletContext) messageContext.get(
                    javax.xml.ws.handler.MessageContext.SERVLET_CONTEXT);
 
        if (servletContext == null) {
            return null;
        }
        this.cachedApplicationContext = 
            (org.springframework.context.ApplicationContext) servletContext
            .getAttribute(
                    org.springframework.web.context.WebApplicationContext.ROOT_WEB_APPLICATION_CONTEXT_ATTRIBUTE);
        return cachedApplicationContext;
    }
}

I appreciate any help

thanks

[garrydias]

I made a mistake.

The error "Failed to acquire pool semaphore" occurs at method invocation. The lookup occurs normally. The exception is thrown in this line

response = worker.execute(request);

(still searching a solution...)

[zefina]

You create a proxy which implements MaterialDaoLocal (well the lookup does that) and you expect to convert it to a concrete implementation (MaterialDaoBean) because it is a dynamic proxy that isn't obviously isn't going to work. You should program to the interface (MaterialDaoLocal)...