From a99c97ba66cb9b42136129dbe9a4d44b5578190d Mon Sep 17 00:00:00 2001
From: Jan David Mol <mol@astron.nl>
Date: Fri, 19 Aug 2016 09:20:01 +0000
Subject: [PATCH] Task #8415: Use docker-run-slurm.sh script, to contain
resources and propagate kill signals in SLURM
---
CEP/Pipeline/recipes/sip/pipeline.cfg.CEP4.tmpl | 2 +-
MAC/Services/src/PipelineControl.py | 2 +-
RTCP/Cobalt/GPUProc/src/scripts/runObservation.sh | 2 +-
3 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/CEP/Pipeline/recipes/sip/pipeline.cfg.CEP4.tmpl b/CEP/Pipeline/recipes/sip/pipeline.cfg.CEP4.tmpl
index e8d8adae319..138835c5bb1 100644
--- a/CEP/Pipeline/recipes/sip/pipeline.cfg.CEP4.tmpl
+++ b/CEP/Pipeline/recipes/sip/pipeline.cfg.CEP4.tmpl
@@ -70,4 +70,4 @@ globalfs = yes
# /bin/bash -c
#
# Required because the pipeline framework needs some bash functionality in the commands it starts.
-cmdline = ssh -n -tt -x localhost srun --exclusive --ntasks=1 --cpus-per-task={nr_cores} --jobid={slurm_job_id} --job-name={job_name} docker run --rm -u {uid} -v /data:/data --net=host {docker_env} lofar-pipeline:${LOFAR_TAG} {command}
+cmdline = ssh -n -tt -x localhost srun --exclusive --ntasks=1 --cpus-per-task={nr_cores} --jobid={slurm_job_id} --job-name={job_name} docker-run-slurm.sh --rm -u {uid} -v /data:/data --net=host {docker_env} lofar-pipeline:${LOFAR_TAG} {command}
diff --git a/MAC/Services/src/PipelineControl.py b/MAC/Services/src/PipelineControl.py
index 7c780af556f..b79095bc2a2 100755
--- a/MAC/Services/src/PipelineControl.py
+++ b/MAC/Services/src/PipelineControl.py
@@ -493,7 +493,7 @@ runcmd {setStatus_active}
wget -O - -q "http://ganglia.control.lofar/ganglia/api/events.php?action=add&start_time=now&summary=Pipeline {obsid} ACTIVE&host_regex="
# run the pipeline
-runcmd docker run --rm --net=host \
+runcmd docker-run-slurm.sh --rm --net=host \
-e LOFARENV={lofarenv} \
-u $UID -e USER=$USER \
-e HOME=$HOME -v $HOME/.ssh:$HOME/.ssh:ro \
diff --git a/RTCP/Cobalt/GPUProc/src/scripts/runObservation.sh b/RTCP/Cobalt/GPUProc/src/scripts/runObservation.sh
index 77fef760786..2333b3230c1 100755
--- a/RTCP/Cobalt/GPUProc/src/scripts/runObservation.sh
+++ b/RTCP/Cobalt/GPUProc/src/scripts/runObservation.sh
@@ -344,7 +344,7 @@ if $DOCKER; then
# "echo 950000 > /sys/fs/cgroup/cpu/cpu.rt_runtime_us" is needed to configure the real-time scheduler for this cgroup
# --cpu-shares=24576 (=24*1024) provides this process with as much share of the CPU as 24 other containers (note: CEP4 has 24 cores/node)
- OUTPUTPROC_CMDLINE="docker run --rm --cpu-shares=24576 --cap-add=sys_nice --cap-add=sys_admin -u `id -u $SSH_USER_NAME` --net=host -v $GLOBALFS_DIR:$GLOBALFS_DIR lofar-outputproc:$TAG bash -c \"sudo echo 950000 > /sys/fs/cgroup/cpu/cpu.rt_runtime_us; $OUTPUTPROC_CMDLINE\""
+ OUTPUTPROC_CMDLINE="docker-run-slurm.sh --rm --cpu-shares=24576 --cap-add=sys_nice --cap-add=sys_admin -u `id -u $SSH_USER_NAME` --net=host -v $GLOBALFS_DIR:$GLOBALFS_DIR lofar-outputproc:$TAG bash -c \"sudo echo 950000 > /sys/fs/cgroup/cpu/cpu.rt_runtime_us; $OUTPUTPROC_CMDLINE\""
fi
echo "[outputProc] command line = $OUTPUTPROC_CMDLINE"
--
GitLab