Hi Jason,
> On Mon, Jan 25, 2010 at 6:54 PM, Steve Huston
> <[email protected]> wrote:
> > alwaysReadableFd is for a pipe - is there a chance you've closed it?
>
> I think there is. It looks like the static scope of
> alwaysReadable and alwaysReadableFd cause them to get
> initialized before main(). When my daemon class closes open
> file descriptors while fork()ing they get closed. I still
> have to figure out how to resolve the conflict.
Right - this seems to be the main problem to solve here. Could you
please open a jira for it?
> There is, however, still a bug in the code I originally cited:
>
> PollerPrivate() :
> epollFd(::epoll_create(DefaultFds)),
> isShutdown(false) {
> QPID_POSIX_CHECK(epollFd);
> ::sigemptyset(&sigMask);
> // Add always readable fd into our set (but not
> listening to it yet)
> ::epoll_event epe;
> epe.events = 0;
> epe.data.u64 = 0;
> QPID_POSIX_CHECK(::epoll_ctl(epollFd, EPOLL_CTL_ADD,
> alwaysReadableFd, &epe));
> }
>
> If an exception is thrown by the second QPID_POSIX_CHECK then
> the file descriptor allocated by
> epollFd(::epoll_create(DefaultFds)) is left dangling.
Ahh, right... The destructor won't be called if the second
QPID_POSIX_CHECK throws. Could you please also open a jira for this?
> A ::close(epollFd) would be needed in the catch() block to free
> it. There are a number of functions with a similar design in
> EpollPoller.cpp that might be similarly affected.
If you could note those in the jira, that would be appreciated.
Thanks,
-Steve
--
Steve Huston, Riverace Corporation
Total Lifecycle Support for Your Networked Applications
http://www.riverace.com
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
