From f24d3003be4c29af3acdda6ed8beb9014e84c1f9 Mon Sep 17 00:00:00 2001
From: Stefan Froehlich <s.froehlich@fz-juelich.de>
Date: Mon, 21 Sep 2015 11:39:03 +0000
Subject: [PATCH] Task #8440: added srun command as it is on jureca right now.
left out the jureca fix for hostnames.
---
.../framework/lofarpipe/support/remotecommand.py | 15 +++++++++++++++
1 file changed, 15 insertions(+)
diff --git a/CEP/Pipeline/framework/lofarpipe/support/remotecommand.py b/CEP/Pipeline/framework/lofarpipe/support/remotecommand.py
index a6c0439fb94..2713ffe96f5 100644
--- a/CEP/Pipeline/framework/lofarpipe/support/remotecommand.py
+++ b/CEP/Pipeline/framework/lofarpipe/support/remotecommand.py
@@ -92,9 +92,24 @@ def run_remote_command(config, logger, host, command, env, arguments = None):
return run_via_mpiexec(logger, command, arguments, host)
elif method == "cep_mpi":
return run_via_mpiexec_cep(logger, command, arguments, host)
+ elif method == "slurm_srun_cep3":
+ return run_via_slurm_srun_cep3(logger, command, arguments, host)
else:
return run_via_ssh(logger, host, command, env, arguments)
+def run_via_slurm_srun_cep3(logger, command, arguments, host):
+ for arg in arguments:
+ command = command + " " + str(arg)
+ commandstring = ["srun","-N 1","--cpu_bind=map_cpu:none","-w",host, "/bin/sh", "-c", "hostname && " + command]
+ # we have a bug that crashes jobs when too many get startet at the same time
+ # temporary NOT 100% reliable workaround
+ #from random import randint
+ #time.sleep(randint(0,10))
+ ##########################
+ process = spawn_process(commandstring, logger)
+ process.kill = lambda : os.kill(process.pid, signal.SIGKILL)
+ return process
+
def run_via_mpirun(logger, host, command, environment, arguments):
"""
Dispatch a remote command via mpirun.
--
GitLab