[Swift-devel] trunk-cobalt block task ended prematurely
Mihael Hategan
hategan at mcs.anl.gov
Wed Mar 4 15:37:18 CST 2015
Right. However, you mentioned before that it works fine when you switch
staging from "direct" to "swift". That causes swift to use the plain
_swiftwrap, which pretty much has the same code:
if [ "$STDIN" == "" ]; then
if [ "$SWIFT_GEN_SCRIPTS" != "" ]; then
genScripts
fi
if [ -n "$TIMECMD" ] && [ -n "$TIMEARGS" ]; then
"$TIMECMD" "${TIMEARGS[@]}" "$EXEC" "${CMDARGS[@]}"
1>"$STDOUT" 2>"$STDERR"
else
"$EXEC" "${CMDARGS[@]}" 1>"$STDOUT" 2>"$STDERR"
fi
else
...
So that's what's puzzling to me.
Can you try changing this line in _swiftwrap.staging:
"$EXEC" "${CMDARGS[@]}" 1>"$STDOUT" 2>"$STDERR"
to
"$EXEC" "${CMDARGS[@]}" 1>"$STDOUT" 2>"$STDERR" </dev/null
and then remove stdin= from the app in the swift script and seeing if
that works?
While it seems like a wise choice to do this in general, I'm trying to
see if this fixes this particular issue.
Mihael
On Wed, 2015-03-04 at 15:22 -0600, Ketan Maheshwari wrote:
> Hi Mihael,
>
> The code in _swiftwrap.staging branches based on if STDIN is present or
> not. See the snippet below:
>
> if [ "$STDIN" == "" ]; then
> if [ "$SWIFT_GEN_SCRIPTS" != "" ]; then
> echo "#!/bin/bash" > run.sh
> echo "\"$EXEC\" \"${CMDARGS[@]}\" 1>\"$STDOUT\" 2>\"$STDERR\"" >>
> run.sh
> chmod +x run.sh
> fi
> "$EXEC" "${CMDARGS[@]}" 1>"$STDOUT" 2>"$STDERR"
> else
> if [ "$SWIFT_GEN_SCRIPTS" != "" ]; then
> echo "#!/bin/bash" > run.sh
> echo "\"$EXEC\" \"${CMDARGS[@]}\" 1>\"$STDOUT\" 2>\"$STDERR\"
> <\"$STDIN\"" >> run.sh
> chmod +x run.sh
> fi
> "$EXEC" "${CMDARGS[@]}" 1>"$STDOUT" 2>"$STDERR" <"$STDIN"
> fi
>
> When "stdin=" is not provided the code takes the first branch and hangs. It
> works otherwise.
>
> It is possible that it hangs because if mpich bug Mike mentioned.
>
> I agree we should stick in a </dev/null in there.
>
> --Ketan
>
>
> On Wed, Mar 4, 2015 at 3:12 PM, Hategan-Marandiuc, Philip M. <
> hategan at mcs.anl.gov> wrote:
>
> > I'm still confused. I don't see any difference in stdin handling between
> > _swiftwrap and _swiftwrap.staging (which is used for direct staging).
> >
> > Maybe we should always feed the app a /dev/null if there is no stdin=
> > specified.
> >
> > Mihael
> >
> > On Wed, 2015-03-04 at 08:50 -0600, Ketan Maheshwari wrote:
> > > I added stdin="/dev/null" to app invocation line and it has worked now.
> > > --Ketan
> > >
> > > On Wed, Mar 4, 2015 at 8:44 AM, Ketan Maheshwari <ketan at mcs.anl.gov>
> > wrote:
> > >
> > > > Please find one with 59 minutes attached. --Ketan
> > > >
> > > > On Tue, Mar 3, 2015 at 11:17 PM, Mihael Hategan <hategan at mcs.anl.gov>
> > > > wrote:
> > > >
> > > >> You are using coasters, so what gets queued is the block, not the job.
> > > >>
> > > >> You should specify execution.options.maxJobTime = "00:59:00".
> > > >>
> > > >> Then you can probably do a walltime of about "00:50:00". But 7 minutes
> > > >> vs. 5 minutes isn't much of a difference.
> > > >>
> > > >> Mihael
> > > >>
> > > >> On Tue, 2015-03-03 at 22:28 -0600, Ketan Maheshwari wrote:
> > > >> > Attached is a log for maxWalltime set to 7 minutes beyond which the
> > job
> > > >> > does not get submitted because of the 1 hour walltime limit of
> > Cetus.
> > > >> > --Ketan
> > > >> >
> > > >> > On Tue, Mar 3, 2015 at 10:15 PM, Ketan Maheshwari <
> > ketan at mcs.anl.gov>
> > > >> wrote:
> > > >> >
> > > >> > > When I check queue with qstat, I see the job is submitted for 40
> > > >> minutes.
> > > >> > > When I try to increase maxWallTime the workflow does not get
> > submitted
> > > >> > > because on Cetus maximum allowed walltime is 60 minutes. --Ketan
> > > >> > >
> > > >> > > On Tue, Mar 3, 2015 at 10:03 PM, Hategan-Marandiuc, Philip M. <
> > > >> > > hategan at mcs.anl.gov> wrote:
> > > >> > >
> > > >> > >> Hi,
> > > >> > >>
> > > >> > >> Looks like almost exactly 5 minutes to me:
> > > >> > >>
> > > >> > >> 2015-03-04 01:45:43,943+0000 INFO Execute TASK_STATUS_CHANGE
> > > >> > >> taskid=urn:R-3-0-2-1425432781969 status=2
> > > >> > >> workerid=0304-3301040-000000:000000
> > > >> > >> 2015-03-04 01:50:44,676+0000 INFO Execute TASK_STATUS_CHANGE
> > > >> > >> taskid=urn:R-3-0-2-1425432781969 status=5 Walltime exceeded
> > > >> > >>
> > > >> > >> Which is what the config file is asking for:
> > > >> > >>
> > > >> > >> app.bgsh {
> > > >> > >> env.SUBBLOCK_SIZE: "16" # [R]
> > line
> > > >> 27
> > > >> > >> executable: "/home/ketan/SwiftApps/subjobs/bg.sh" # [R]
> > line
> > > >> 25
> > > >> > >> maxWallTime: "00:05:00" # [R]
> > line
> > > >> 26
> > > >> > >> }
> > > >> > >>
> > > >> > >> Again, the wrapper log shows the app as still running. Last line
> > is:
> > > >> > >> Progress 2015-03-04 01:45:43.971393118+0000 EXECUTE
> > > >> > >>
> > > >> > >> Please do me a favor and increase the walltime to one hour and
> > let's
> > > >> see
> > > >> > >> what happens then.
> > > >> > >>
> > > >> > >> If it still doesn't finish after one hour, we could try to
> > strace it
> > > >> and
> > > >> > >> see what is happening there.
> > > >> > >>
> > > >> > >> Mihael
> > > >> > >>
> > > >> > >> On Tue, 2015-03-03 at 19:53 -0600, Ketan Maheshwari wrote:
> > > >> > >> > Please find the log attached. --Ketan
> > > >> > >> >
> > > >> > >> > On Tue, Mar 3, 2015 at 7:03 PM, Hategan-Marandiuc, Philip M. <
> > > >> > >> > hategan at mcs.anl.gov> wrote:
> > > >> > >> >
> > > >> > >> > > On Tue, 2015-03-03 at 15:42 -0600, Ketan Maheshwari wrote:
> > > >> > >> > > > Slow network looks unlikely to be a cause:
> > > >> > >> > >
> > > >> > >> > > It's the only variable obvious, so I wouldn't say that.
> > > >> > >>
> > > >> > >> I meant "only obvious variable" there.
> > > >> > >>
> > > >> > >>
> > > >> > >>
> > > >> > >
> > > >>
> > > >>
> > > >> _______________________________________________
> > > >> Swift-devel mailing list
> > > >> Swift-devel at ci.uchicago.edu
> > > >> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel
> > > >>
> > > >
> > > >
> >
> >
> >
More information about the Swift-devel
mailing list