| 1 : |
agomez |
10 |
#!/usr/bin/python
|
| 2 : |
|
|
#
|
| 3 : |
|
|
# BonFIRE Virtual Clusters on Federated Clouds Demonstration Kit
|
| 4 : |
|
|
#
|
| 5 : |
|
|
# Copyright (c) Fundacion Centro Tecnologico de Supercomputacion de Galicia 2012
|
| 6 : |
|
|
#
|
| 7 : |
agomez |
14 |
# License Apache Software
|
| 8 : |
agomez |
10 |
#
|
| 9 : |
|
|
# The research leading to these results has received funding from
|
| 10 : |
|
|
# the European Community's Seventh Framework Programme (FP7/2007-2013)
|
| 11 : |
|
|
# under agreement number 257386
|
| 12 : |
|
|
#
|
| 13 : |
|
|
# This software is provided with ABSOLUTELY NO WARRANTY
|
| 14 : |
|
|
#
|
| 15 : |
|
|
import sys
|
| 16 : |
|
|
import time
|
| 17 : |
|
|
|
| 18 : |
|
|
from logger import log,configure
|
| 19 : |
|
|
|
| 20 : |
|
|
UPDATER_PERIOD = 30
|
| 21 : |
|
|
|
| 22 : |
|
|
if len(sys.argv) >1:
|
| 23 : |
|
|
configure(logfile = sys.argv[1], debug=True, console=False)
|
| 24 : |
|
|
else:
|
| 25 : |
|
|
configure(debug=True, console=True)
|
| 26 : |
|
|
|
| 27 : |
|
|
try:
|
| 28 : |
|
|
import bonfire
|
| 29 : |
|
|
import ogs
|
| 30 : |
|
|
import hostsfile
|
| 31 : |
|
|
|
| 32 : |
|
|
default_bonfire = "/etc/default/bonfire"
|
| 33 : |
|
|
bonfire.read_defaults(default_bonfire)
|
| 34 : |
|
|
hosts_file = "/etc/hosts"
|
| 35 : |
|
|
|
| 36 : |
|
|
log.info("Bonfire URI: %s" % bonfire.uri)
|
| 37 : |
|
|
log.info("Credentials: %s:%s" % (bonfire.user,bonfire.password,))
|
| 38 : |
|
|
log.info("Experiment: %s" % bonfire.experiment_id)
|
| 39 : |
|
|
|
| 40 : |
|
|
log.debug("Set OGS all.q queue to use /bin/bash")
|
| 41 : |
|
|
while(True):
|
| 42 : |
|
|
try:
|
| 43 : |
|
|
ogs.modify_queue("all.q",{"shell":"/bin/bash"})
|
| 44 : |
|
|
ogs.modify_queue("all.q",{"rerun":"TRUE"})
|
| 45 : |
|
|
ogs.modify_global({"max_unheard":"00:02:00","reschedule_unknown":"00:01:00"})
|
| 46 : |
|
|
break
|
| 47 : |
|
|
except Exception as excpt:
|
| 48 : |
|
|
log.exception(excpt)
|
| 49 : |
|
|
time.sleep(10)
|
| 50 : |
|
|
|
| 51 : |
|
|
while(True):
|
| 52 : |
|
|
try:
|
| 53 : |
|
|
log.info("Get experiment")
|
| 54 : |
|
|
experiment = bonfire.get_experiment(bonfire.experiment_id)
|
| 55 : |
|
|
|
| 56 : |
|
|
try:
|
| 57 : |
|
|
#Update /etc/hosts
|
| 58 : |
|
|
log.info("Hosts file")
|
| 59 : |
|
|
hosts = hostsfile.load(hosts_file)
|
| 60 : |
|
|
new_hosts = [(compute['ip'],compute['hostname']) for compute in experiment['computes'] if "client" in compute['hostname'] and not "DONE" in compute['state'] and not "FAILED" in compute ['state'] and not hosts.has_key(compute['ip'])]
|
| 61 : |
|
|
log.debug(new_hosts)
|
| 62 : |
|
|
for ip,hostname in new_hosts:
|
| 63 : |
|
|
#Add to hosts list file
|
| 64 : |
|
|
hostsfile.append(hosts_file,ip,hostname)
|
| 65 : |
|
|
except Exception as excpt:
|
| 66 : |
|
|
log.exception(excpt)
|
| 67 : |
|
|
|
| 68 : |
|
|
try:
|
| 69 : |
|
|
#Update OGS
|
| 70 : |
|
|
log.info("Update OGS")
|
| 71 : |
|
|
hosts = ogs.get_execution_hosts()
|
| 72 : |
|
|
new_hosts = [(compute['ip'],compute['hostname'],float(compute['cpu'])) for compute in experiment['computes'] if "client" in compute['hostname'] and not "DONE" in compute ['state'] and not "FAILED" in compute ['state'] and not compute['hostname'] in hosts]
|
| 73 : |
|
|
log.debug(new_hosts)
|
| 74 : |
|
|
for ip,hostname,cpu in new_hosts:
|
| 75 : |
|
|
#Add to OGS
|
| 76 : |
|
|
ogs.new_host(hostname,cpu)
|
| 77 : |
|
|
except Exception as excpt:
|
| 78 : |
|
|
log.exception(excpt)
|
| 79 : |
|
|
|
| 80 : |
|
|
try:
|
| 81 : |
|
|
#Remove invalid hosts
|
| 82 : |
|
|
log.info("Remove invalid hosts")
|
| 83 : |
|
|
done_hostnames = [compute['hostname'] for compute in experiment['computes'] if "client" in compute['hostname'] and (("DONE" in compute['state']) or ("FAILED" in compute['state']))]
|
| 84 : |
|
|
log.debug(done_hostnames)
|
| 85 : |
|
|
for hostname in done_hostnames:
|
| 86 : |
|
|
#Remove from OGS
|
| 87 : |
|
|
ogs.remove_host(hostname)
|
| 88 : |
|
|
except Exception as excpt:
|
| 89 : |
|
|
log.exception(excpt)
|
| 90 : |
|
|
except Exception as excpt:
|
| 91 : |
|
|
log.exception(excpt)
|
| 92 : |
|
|
time.sleep(UPDATER_PERIOD)
|
| 93 : |
|
|
except Exception as excpt:
|
| 94 : |
|
|
log.exception(excpt)
|
Parent Directory
Revision Log