> On Jan. 24, 2014, 6:16 p.m., Ian Downes wrote: > > 3rdparty/libprocess/include/process/subprocess.hpp, line 120 > > <https://reviews.apache.org/r/17306/diff/1/?file=447773#file447773line120> > > > > should we check their exit codes since dup2 can fail even with valid > > descriptors.
I've handled EINTR now, and updating os::close to handle EINTR as well in a separate review. It doesn't appear that the other error codes are possible here: EBADF: oldfd was opened above since we check the result of pipe, newfd is in the valid range. EBUSY: (Linux), doesn't look like we're susceptible to a race condition between open and dup2. EINTR: Handled this now. EINVAL: Not possible here. EMFILE: Since dup2 will close newfd (stdin), it does not seem possible to for dup2 to cause the number of open file descriptors in the subprocess to be exceeded. - Ben ----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/17306/#review32726 ----------------------------------------------------------- On Jan. 24, 2014, 7:06 a.m., Ben Mahler wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/17306/ > ----------------------------------------------------------- > > (Updated Jan. 24, 2014, 7:06 a.m.) > > > Review request for mesos, Benjamin Hindman, Ian Downes, and Jie Yu. > > > Bugs: MESOS-943 > https://issues.apache.org/jira/browse/MESOS-943 > > > Repository: mesos-git > > > Description > ------- > > This adds an asynchronous mechanism for subprocess execution, per MESOS-943. > > What started simple was made a little more complex due to the following > issues: > > 1. Who is responsible for closing the input / output descriptors? > > Placing this burden onto the caller of subprocess() seems likely to yield > leaked open file descriptors. This introduced the notion of a shared_ptr / > destructor / copy constructor / assignment constructor to ensure that the > file descriptors are closed when the handle to the file descriptors are lost. > However, even with my implementation, one may copy these file descriptors, at > which point they may be deleted from underneath them. > > 2. What does discarding the status entail? Does it kill the process? > > The current implementation kills the process, which requires the use of an > explicit Promise to deal with the discard from the caller not affecting the > reaper's future. If discard() is a no-op, we must still use an explicit > Promise to preserve the notification from the Reaper (so that we can know > when to delete the Reaper). > > > That's about it, I've added tests that demonstrate the ability to communicate > with the subprocess through stdin / stout / stderr. > > Please let me know if you find any simplifications that can be made! (Other > than C++11 lambdas, of course :)) > > > Diffs > ----- > > 3rdparty/libprocess/Makefile.am 40f01a7b3803696ccca440c8326e1d6d7c377459 > 3rdparty/libprocess/include/process/subprocess.hpp PRE-CREATION > 3rdparty/libprocess/src/tests/subprocess_tests.cpp PRE-CREATION > > Diff: https://reviews.apache.org/r/17306/diff/ > > > Testing > ------- > > Tests were added and ran in repetition. > > > Thanks, > > Ben Mahler > >
