Re: tapestry.thread-pool.max-pool-size not doing anything
You're rigth. If multiple threads are scheduling jobs and the pool has reached a threadcount of maxsize-1, chances are that a RejectedExecutionException is thrown. The offer-method should use a lock. Like so: public boolean offer(Runnable inJob) { _lock.lock(); try { if (_executor.getPoolSize() _executor.getMaximumPoolSize()) { return false; } return super.offer(inJob); } finally { _lock.unlock(); } } --Arjan Verstoep On 2 Jul 2014, at 09:52, Lance Java wrote: I'm not convinced this solution is thread-safe. On 1 Jul 2014 17:21, Arjan Verstoep a.j.verst...@alumnus.utwente.nl wrote: Yes, the default implementation is very counter-intuitive. I have invesigated once, and it was possible to 'repair' the ThreadPoolExecutor if you provide the 'right' work-queue. I don't know if you can apply this to Tapestry's ThreadPoolExecutor, but anyhow: here's how I do it: --Arjan Verstoep /** * Specific implementation of the {@link LinkedBlockingQueue} that won't * accept offered elements as long as the given {@link ThreadPoolExecutor} has * not yet reached its maximum capacity. * pThe intended effect is to have a ThreadPoolExecutor that will only start * adding Runnables to the Queue iafter/i the maximum number of Threads is * reached. The default implementation of the ThreadPoolExecutor is, that it * first starts new Threads when the Queue is full, which with an unbounded * Queue will inever/i be the case./p * pThis class will throw an * unchecked {@link java.lang.NullPointerException} if the ThreadPoolExecutor * is not set with {@link #setExecutor(ThreadPoolExecutor)} !/p */ class ThreadCreationTriggeringLinkedBlockingQueueT extends LinkedBlockingQueueRunnable { private static final long serialVersionUID = 1L; private transient ThreadPoolExecutor _executor; /** * pIf the Executor's current pool size is below the maximum pool size, * this method will not add the {@link Runnable} to the Queue.br/ * The {@link ThreadPoolExecutor} will then be forced to start a new Thread. * /p{@inheritDoc} */ @Override public boolean offer(Runnable inJob) { if (_executor.getPoolSize() _executor.getMaximumPoolSize()) { return false; } return super.offer(inJob); } /** * Sets the executor which will be used to decide whether or not to add the * Runnable to the Queue. * * @param inExecutor the ThreadPoolExecutor */ public void setExecutor(ThreadPoolExecutor inExecutor) { _executor = inExecutor; } } ThreadCreationTriggeringLinkedBlockingQueueRunnable queue = new ThreadCreationTriggeringLinkedBlockingQueueRunnable(); ThreadPoolExecutor threadpoolExecutor = new ThreadPoolExecutor(availableProcessors(), THREADPOOLMULTIPLICATOR * availableProcessors(), 10, TimeUnit.MINUTES, queue); queue.setExecutor(threadpoolExecutor); On 1 Jul 2014, at 16:04, Lance Java wrote: I feel the implementation is flawed. I would prefer that the pool size would increase from minimum to maximum under load to increase throughput and then revert back to the minimum when not under load. But instead, the pool size stays at minimum until the queue is full. In my opinion, the application is about to fall over at this point. Only then will the pool size increase which i think is crazy / stupid. Because of this, I've always seen core pool size and max pool size set to the same value. To achieve my preferred behaviour (above) I think you can specify different pooling behaviour. I think it required throwing exceptions and extending ThreadPoolExecutor or something nasty like that. On 1 Jul 2014 12:14, D Tim Cummings t...@triptera.com.au wrote: Ok, thanks for explaining this... I think :) I had a read of the ThreadPoolExecutor javadoc. It seems to me that core-pool-size should be described as the maximum pool size before queuing. Describing it as the minimum pool size is misleading because there is no minimum. The number of threads in the pool before any have been invoked is always zero. Tim On 1 Jul 2014, at 16:25, Lance Java lance.j...@googlemail.com wrote: If you read the javadoc for java.util.concurrent ThreadPoolExecutor you'll see that the number of threads will only increase when the queue has reached its capacity. Crazy / stupid behaviour if you ask me... But expected. I've been caught out by this before when I set core pool size to 1, expecting the thread size to increase under load. On 1 Jul 2014 02:53, D Tim Cummings t...@triptera.com.au wrote: Hi I am using Tapestry 5.3.7 and the ParallelExecutor. When I increase the max-pool-size I am not able to use any more threads. However when I increase the core-pool-size I am able to use more threads. It looks like the max-pool-size is not doing anything (or I am not understanding what
Re: tapestry.thread-pool.max-pool-size not doing anything
Yes, the default implementation is very counter-intuitive. I have invesigated once, and it was possible to 'repair' the ThreadPoolExecutor if you provide the 'right' work-queue. I don't know if you can apply this to Tapestry's ThreadPoolExecutor, but anyhow: here's how I do it: --Arjan Verstoep /** * Specific implementation of the {@link LinkedBlockingQueue} that won't * accept offered elements as long as the given {@link ThreadPoolExecutor} has * not yet reached its maximum capacity. * pThe intended effect is to have a ThreadPoolExecutor that will only start * adding Runnables to the Queue iafter/i the maximum number of Threads is * reached. The default implementation of the ThreadPoolExecutor is, that it * first starts new Threads when the Queue is full, which with an unbounded * Queue will inever/i be the case./p * pThis class will throw an * unchecked {@link java.lang.NullPointerException} if the ThreadPoolExecutor * is not set with {@link #setExecutor(ThreadPoolExecutor)} !/p */ class ThreadCreationTriggeringLinkedBlockingQueueT extends LinkedBlockingQueueRunnable { private static final long serialVersionUID = 1L; private transient ThreadPoolExecutor _executor; /** * pIf the Executor's current pool size is below the maximum pool size, * this method will not add the {@link Runnable} to the Queue.br/ * The {@link ThreadPoolExecutor} will then be forced to start a new Thread. * /p{@inheritDoc} */ @Override public boolean offer(Runnable inJob) { if (_executor.getPoolSize() _executor.getMaximumPoolSize()) { return false; } return super.offer(inJob); } /** * Sets the executor which will be used to decide whether or not to add the * Runnable to the Queue. * * @param inExecutor the ThreadPoolExecutor */ public void setExecutor(ThreadPoolExecutor inExecutor) { _executor = inExecutor; } } ThreadCreationTriggeringLinkedBlockingQueueRunnable queue = new ThreadCreationTriggeringLinkedBlockingQueueRunnable(); ThreadPoolExecutor threadpoolExecutor = new ThreadPoolExecutor(availableProcessors(), THREADPOOLMULTIPLICATOR * availableProcessors(), 10, TimeUnit.MINUTES, queue); queue.setExecutor(threadpoolExecutor); On 1 Jul 2014, at 16:04, Lance Java wrote: I feel the implementation is flawed. I would prefer that the pool size would increase from minimum to maximum under load to increase throughput and then revert back to the minimum when not under load. But instead, the pool size stays at minimum until the queue is full. In my opinion, the application is about to fall over at this point. Only then will the pool size increase which i think is crazy / stupid. Because of this, I've always seen core pool size and max pool size set to the same value. To achieve my preferred behaviour (above) I think you can specify different pooling behaviour. I think it required throwing exceptions and extending ThreadPoolExecutor or something nasty like that. On 1 Jul 2014 12:14, D Tim Cummings t...@triptera.com.au wrote: Ok, thanks for explaining this... I think :) I had a read of the ThreadPoolExecutor javadoc. It seems to me that core-pool-size should be described as the maximum pool size before queuing. Describing it as the minimum pool size is misleading because there is no minimum. The number of threads in the pool before any have been invoked is always zero. Tim On 1 Jul 2014, at 16:25, Lance Java lance.j...@googlemail.com wrote: If you read the javadoc for java.util.concurrent ThreadPoolExecutor you'll see that the number of threads will only increase when the queue has reached its capacity. Crazy / stupid behaviour if you ask me... But expected. I've been caught out by this before when I set core pool size to 1, expecting the thread size to increase under load. On 1 Jul 2014 02:53, D Tim Cummings t...@triptera.com.au wrote: Hi I am using Tapestry 5.3.7 and the ParallelExecutor. When I increase the max-pool-size I am not able to use any more threads. However when I increase the core-pool-size I am able to use more threads. It looks like the max-pool-size is not doing anything (or I am not understanding what it is supposed to do). When I use the defaults I get a maximum of 3 threads and other threads seem to wait in a queue and execute later. When I use the following settings in web.xml I still get only 3 threads context-param param-nametapestry.thread-pool.core-pool-size /param-name param-value3/param-value /context-param context-param param-nametapestry.thread-pool.max-pool-size/param-name param-value20/param-value /context-param When I use the following settings in web.xml I get 3 threads context-param param-nametapestry.thread-pool.core-pool-size /param-name param-value3/param-value /context-param context-param param-nametapestry.thread-pool.max-pool-size/param-name
Re: OutOfMemoryError after serving N pages
[EMAIL PROTECTED] wrote: We had the same thing, and in our case it was related to a small PermGen Space setting in our Tomcat container. My understanding of the Sun JVM is that it partitions up its allocated memory into heap, perm gen, etc. chunks and that the Perm Gen space is where all meta data about classes gets stored. In modern JEE programming with all the CGLIB and just-in-time abstract class overrides instantiation, the Perm Gen space tends to blow up quicker. Anyway, try throwing a -XX:MaxPermSize=256m on your JVM startup and see if that helps. Note: if you're actually running out of PermGen space, you'll see a reference to that in your OOME. If you are just getting plain old OOMEs, it sounds like something more sinister. HTH, Tom My JVM is complaining about heap space, so I fear that it is something sinister... 10:36:23,289 ERROR [MusiController4]:253 - Servlet.service() for servlet MusiController4 threw exception java.lang.OutOfMemoryError: Java heap space ~Arjan Verstoep - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Populating Tapestry ASO after successful Acegi Auth
[EMAIL PROTECTED] wrote: Greetings: I'm using Tapestry 4 and Acegi 1.0.1. I have acegi set up to perform authentication using the AuthenticationProcessingFilter. I have a requirement to place the domain object the Acegi UserDetails object is based upon into a Tapestry ASO for use during the user session. =20 My question is: what the best practice for filling the Tapestry ASO with the domain model's user object upon successful authentication with Acegi? Thanks, -jason Hi Jason, Set up your login-prcedure to redirect you to a certain page after successful login. Then, on that particular page put this code to get the Authentication-object. Then, You can put the Authentication object into the ASO-object. I currently use this code in my border-component, so it executes way too often. But I'll get to that eventually. ~ Arjan Verstoep private Authentication getAuthentication(IRequestCycle cycle) { Authentication auth = null; WebSession session = cycle.getInfrastructure().getRequest().getSession(false); if (session!=null) { SecurityContext sc = (SecurityContext) session.getAttribute(HttpSessionContextIntegrationFilter.ACEGI_SECURITY_CONTEXT_KEY); if (sc!=null) { auth = sc.getAuthentication(); } } return auth; } - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]