Closed mtitov closed 10 months ago
We didn't look at flux for quite a time, might be worthwhile to add to the tests, even though the deployment will eat up quite some time.
this sticks out:
2023-02-27T01:15:18.437598Z broker.err[0]: rc2.0: /bin/bash Broken pipe (rc=141) 1.0s Hard to tell why the pipe fails - we need to check our plumbing for the flux stdio grab I guess...
What is the priority on this? And what Flux version did you happen to use?
What is the priority on this?
~I've prepared a PR within ExaWorks to run RP-Flux as part of the testing, thus would be good to keep it open not for a long time - 1-2 sprints (?)~
And what Flux version did you happen to use?
that came out while running SDK docker container which has flux-core=0.28.0
, flux-sched=0.17.0
. Flux spack package has the latest versions, but our exaworks spack package failing due to swift/t dependency error.
For test purposes we can use a corresponding testing environment - https://github.com/ExaWorks/containerized-testing-environment/tree/main/psij/flux
Tried to change the command to start FLUX, to imitate tty
with no luck
Original command we have in RU
check = 'flux env; echo "OK"; while true; do echo "ok"; sleep 1; done'
start = 'flux start -o,-v,-S,log-filename=%s.log' % self._uid
cmd = '/bin/bash -c "echo \\\"%s\\\" | %s"' % (check, start)
(*) with the [relatively] new update in Flux, flux start
should be launched either within tty
or with executable (e.g., flux start <executable>
)
Work in progress..
Flux team responded with a recipe to run Flux w/o pty - @andre-merzky to apply patch
@andre-merzky could you open a PR draft while you work on this? The idea is to start to review early to speed up the process as much as possible (i.e., supporting Frontier).