https://bz.apache.org/bugzilla/show_bug.cgi?id=57653
--- Comment #8 from Mark Thomas <ma...@apache.org> --- (In reply to Christopher Schultz from comment #7) > Not knowing a great deal about the underlying issue... It is worth reading AprEndpoint.Poller.run() and the native implementation of poll() > If the fd is registered for both READ and WRITE, and we get multiple > "responses" from apr_pollset_poll, are those happening (somewhat) > simultaneously in two separate threads? No. > So one of them fetches the fd, then > the other fetches the same fd, then one of them removes the fd, then the > other tries to do that same thing? No. It happens sequentially in a single thread. > Or do we need to examine the return from apr_pollset_poll to find out if we > got the READ versus WRITE notification and make sure it's the one we were > expecting? That is handled in Poller.run() The fundamental problem is that poll() assumed that there would never be multiple notifications for the same socket. That assumption was not valid. -- 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
