https://issues.apache.org/bugzilla/show_bug.cgi?id=49159
--- Comment #8 from sylvain.laur...@gmail.com 2010-09-25 18:42:53 EDT --- Created an attachment (id=26074) --> (https://issues.apache.org/bugzilla/attachment.cgi?id=26074) proposed patch to renew threads one by one with a delay I explored your idea and managed to force threads to "kill themselves" after a context is stopped. Here are the main points of my new patch : - I save the last time a context is stopped in a static variable of StandardContext (this could be improved to record it somewhere at the Service or Engine level, in a instance variable) - The "suicide" is done in ThreadPoolExecutor.afterExecute(), in case the thread was created before the last context was stopped. - the rate at which threads are renewed is throttled to avoid renewing all threads at the same time (otherwise it would have been the same as my initial proposal to renew the whole pool). - I had to make TaskQueue extend ArrayBlockingQueue instead of LinkedBlockingQueue because the latter does not propose to have a "fair" policy. Fairness is required so that every thread of the pool is given the opportunity to be killed. - Unlike LinkedBlockingQueue, ArrayBlockingQueue requires a reasonable size (not Integer.MAX_VALUE). I set it to 100 by default (configurable with the maxQueueSize property at the Connector or Executor level) I'm quite happy with this patch except the fairness requirement : - The javadocs of ArrayBlockingQueue say that it may decrease performance, but I don't how much - Since each thread of the pool is used in turn, we might lose some caching effects - and it also prevents the size of the pool to be decreased in case of low activity : without fairness, some Thread could timeout waiting for an element from the queue while other threads could serve new requests. Eventually the pool could decide to decrease it size. What do you think ? If we don't use a fair queue, we do not have a deterministic behavior and some thread might still trigger a memory leak... -- Configure bugmail: https://issues.apache.org/bugzilla/userprefs.cgi?tab=email ------- You are receiving this mail because: ------- You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
