[c3-users] c3-5.0.1-1 processes stuck with lots of output (patch included)
naughtont at ornl.gov
Thu May 13 11:52:19 EDT 2010
Yes, you've hit the right spot. I'll add this to the C3 tracker.
Thank you for the patch and bug report. We'll get these rolled into the
next cut of C3.
Thomas Naughton naughtont at ornl.gov
Research Associate (865) 576-4184
On Thu, 13 May 2010, Erik Jacobson wrote:
> Hello. It was unclear to me where to send this because the latest version
> of c3 isn't on the c3 web site, but rather, we found it over in the oscar
> project. I included the C3 contacts and the oscar-devel list in this
> c3-5.0.1-1.src.rpm from oscar's page has a problem when there is a lot of
> output. Basically, a write pipe blocks because the process that does to
> the reading is in waitpid and not actively reading from the pipe.
> This creates a deadlock.
> In this patch, I moved the waitpid stuff after the pipe reads in the
> area that caused the problem. I've attached a patch.
> To duplicate the problem, run the script chunk below with a command similar
> to the following (same problem with more than one node; just kept it to
> one node for simplicity).
> cexec -p rack_1:15-15 /tmp/runme 2500
> The script:
> #! /bin/bash
> while [ $cnt -le $1 ]; do
> echo "output line $cnt"
> #echo $cnt
> let cnt++
> Erik Jacobson - Linux System Software - SGI - Eagan, Minnesota
More information about the c3-users