def grid_reload_from_id(gridjob):
logger.info("Reloading the resources from oargrid job %s", gridjob)
gridjob = int(gridjob)
nodes = ex5.get_oargrid_job_nodes(gridjob)
job_sites = ex5.get_oargrid_job_oar_jobs(gridjob)
vlans = []
subnets = []
for (job_id, site) in job_sites:
vlan_ids = ex5.get_oar_job_kavlan(job_id, site)
vlans.extend([{
"site": site,
"vlan_id": vlan_id
} for vlan_id in vlan_ids])
# NOTE(msimonin): this currently returned only one subnet
# even if several are reserved
# We'll need to patch execo the same way it has been patched for vlans
ipmac, info = ex5.get_oar_job_subnets(job_id, site)
if not ipmac:
logger.debug("No subnet information found for this job")
continue
subnet = {
"site": site,
"ipmac": ipmac,
}
subnet.update(info)
# Mandatory key when it comes to concretize resources
subnet.update({"network": info["ip_prefix"]})
subnets.append(subnet)
return nodes, vlans, subnets
def concretize_resources(resources, gridjob, reservation_type):
if reservation_type == "oar":
nodes = ex5.get_oar_job_nodes(gridjob)
else:
nodes = ex5.get_oargrid_job_nodes(gridjob)
concretize_nodes(resources, nodes)
if reservation_type == "oar":
# This block is in charge of detecting the site of the oar reservation
site_candidates = []
for network_description in resources.get("machines", []):
cluster = network_description.get("cluster")
site_candidates += [ex5.get_cluster_site(cluster)]
for network_description in resources.get("networks", []):
site_candidates += [network_description.get("site", "unknown")]
if len(set(site_candidates)) == 1:
site = site_candidates[0]
else:
raise "Could not detect the g5k site of the oarjob %s" % gridjob
job_sites = [(gridjob, site)]
else:
job_sites = ex5.get_oargrid_job_oar_jobs(gridjob)
vlans = []
for (job_id, site) in job_sites:
vlan_ids = ex5.get_oar_job_kavlan(job_id, site)
vlans.extend([{
"site": site,
"vlan_id": vlan_id
} for vlan_id in vlan_ids])
concretize_networks(resources, vlans)
def _get_jobs_and_vlans(self, conf):
"""Get the hosts from an existing job (if any) or from a new job.
This will perform a reservation if necessary."""
provider_conf = conf['provider']
# Look if there is a running job or make a new reservation
gridjob, _ = EX5.planning.get_job_by_name(provider_conf['name'])
if gridjob is None:
gridjob = self._make_reservation(conf)
else:
logging.info("Using running oargrid job %s" % gridjob)
# Wait for the job to start
EX5.wait_oargrid_job_start(gridjob)
nodes = sorted(EX5.get_oargrid_job_nodes(gridjob),
key=lambda n: n.address)
# Checking the number of nodes given
# the disribution policy
self._check_nodes(nodes=nodes,
resources=conf['resources'],
mode=provider_conf['role_distribution'])
# vlans information
job_sites = EX5.get_oargrid_job_oar_jobs(gridjob)
jobs = []
vlans = []
for (job_id, site) in job_sites:
jobs.append((site, job_id))
vlan_id = EX5.get_oar_job_kavlan(job_id, site)
if vlan_id is not None:
vlans.append((site, EX5.get_oar_job_kavlan(job_id, site)))
return (jobs, vlans, nodes)
def grid_reload_from_id(gridjob):
logger.info("Reloading the resources from oargrid job %s", gridjob)
gridjob = int(gridjob)
nodes = ex5.get_oargrid_job_nodes(gridjob)
job_sites = ex5.get_oargrid_job_oar_jobs(gridjob)
vlans = []
subnets = []
for (job_id, site) in job_sites:
vlans, subnets = get_network_info_from_job_id(job_id, site, vlans,
subnets)
return nodes, vlans, subnets
def _get_job(self):
"""Get the hosts from an existing job (if any) or from a new job.
This will perform a reservation if necessary."""
# Look if there is a running job or make a new reservation
self.gridjob, _ = EX5.planning.get_job_by_name(self.config['name'])
if self.gridjob is None:
self._make_reservation()
else:
logging.info("Using running oargrid job %s" % self.gridjob)
# Wait for the job to start
EX5.wait_oargrid_job_start(self.gridjob)
# # XXX Still useful?
# attempts = 0
# self.nodes = None
# while self.nodes is None and attempts < MAX_ATTEMPTS:
# self.nodes = sorted(EX5.get_oargrid_job_nodes(self.gridjob),
# key = lambda n: n.address)
# attempts += 1
self.nodes = sorted(EX5.get_oargrid_job_nodes(self.gridjob),
key=lambda n: n.address)
# # XXX check already done into `_deploy`.
self._check_nodes(nodes=self.nodes,
resources=self.config['resources'],
mode=self.config['role_distribution'])
# XXX(Ad_rien_) Start_date is never used, deadcode? - August
# 11th 2016
self.start_date = None
job_info = EX5.get_oargrid_job_info(self.gridjob)
if 'start_date' in job_info:
self.start_date = job_info['start_date']
# filling some information about the jobs here
self.user = None
job_info = EX5.get_oargrid_job_info(self.gridjob)
if 'user' in job_info:
self.user = job_info['user']
# vlans information
job_sites = EX5.get_oargrid_job_oar_jobs(self.gridjob)
self.jobs = []
self.vlans = []
for (job_id, site) in job_sites:
self.jobs.append((site, job_id))
vlan_id = EX5.get_oar_job_kavlan(job_id, site)
if vlan_id is not None:
self.vlans.append((site, EX5.get_oar_job_kavlan(job_id, site)))
def concretize_resources(resources, gridjob):
nodes = ex5.get_oargrid_job_nodes(gridjob)
concretize_nodes(resources, nodes)
job_sites = ex5.get_oargrid_job_oar_jobs(gridjob)
vlans = []
for (job_id, site) in job_sites:
vlan_ids = ex5.get_oar_job_kavlan(job_id, site)
vlans.extend([{
"site": site,
"vlan_id": vlan_id} for vlan_id in vlan_ids])
concretize_networks(resources, vlans)
def get_job(self):
"""Get the hosts from an existing job (if any) or from a new job.
This will perform a reservation if necessary."""
# Look if there is a running job or make a new reservation
self.gridjob, _ = EX5.planning.get_job_by_name(self.config['name'])
if self.gridjob is None:
self._make_reservation()
else:
logger.info("Using running oargrid job %s" % style.emph(self.gridjob))
# Wait for the job to start
EX5.wait_oargrid_job_start(self.gridjob)
attempts = 0
self.nodes = None
while self.nodes is None and attempts < MAX_ATTEMPTS:
self.nodes = sorted(EX5.get_oargrid_job_nodes(self.gridjob),
key = lambda n: n.address)
attempts += 1
check_nodes(
nodes = self.nodes,
resources = self.config['resources'],
mode = self.config['role_distribution'])
# TODO - Start_date is never used, deadcode ? Ad_rien_ - August 11th 2016
self.start_date = None
job_info = EX5.get_oargrid_job_info(self.gridjob)
if 'start_date' in job_info:
self.start_date = job_info['start_date']
## filling some information about the jobs here
self.user = None
job_info = EX5.get_oargrid_job_info(self.gridjob)
if 'user' in job_info:
self.user = job_info['user']
## vlans information
job_sites = EX5.get_oargrid_job_oar_jobs(self.gridjob)
self.jobs = []
self.vlans = []
for (job_id, site) in job_sites:
self.jobs.append((site, job_id))
vlan_id = EX5.get_oar_job_kavlan(job_id, site)
if vlan_id is not None:
self.vlans.append((site, EX5.get_oar_job_kavlan(job_id, site)))
return self.gridjob
def get_oargrid_job_vm5k_resources(oargrid_job_id):
"""Retrieve the hosts list and (ip, mac) list by sites from an
oargrid_job_id and return the resources dict needed by vm5k_deployment,
with kavlan-global if used in the oargrid job """
oargrid_job_id = int(oargrid_job_id)
logger.info('Waiting job start')
wait_oargrid_job_start(oargrid_job_id)
resources = get_oar_job_vm5k_resources([(oar_job_id, site)
for oar_job_id, site in
get_oargrid_job_oar_jobs(oargrid_job_id)])
kavlan_global = None
for site, res in resources.iteritems():
if res['kavlan'] >= 10:
kavlan_global = {'kavlan': res['kavlan'],
'ip_mac': resources[site]['ip_mac'],
'site': site}
break
if kavlan_global:
resources['global'] = kavlan_global
return resources
def get_job_by_name(job_name, sites=None):
""" """
logger.detail('Looking for a job named %s', style.emph(job_name))
if not sites:
sites = get_g5k_sites()
oargrid_jobs = get_current_oargrid_jobs()
if len(oargrid_jobs) > 0:
for g_job in oargrid_jobs:
for job in get_oargrid_job_oar_jobs(g_job):
info = get_oar_job_info(job[0], job[1])
if info['name'] == job_name:
logger.info('Oargridjob %s found !', style.emph(g_job))
return g_job, None
running_jobs = get_current_oar_jobs(sites)
for job in running_jobs:
info = get_oar_job_info(job[0], job[1])
if info['name'] == job_name:
logger.info('Job %s found on site %s !', style.emph(job[0]),
style.host(job[1]))
return job
return None, None
def get_oargrid_job_vm5k_resources(oargrid_job_id):
"""Retrieve the hosts list and (ip, mac) list by sites from an
oargrid_job_id and return the resources dict needed by vm5k_deployment,
with kavlan-global if used in the oargrid job """
oargrid_job_id = int(oargrid_job_id)
logger.info('Waiting job start')
wait_oargrid_job_start(oargrid_job_id)
resources = get_oar_job_vm5k_resources([(oar_job_id, site)
for oar_job_id, site in
get_oargrid_job_oar_jobs(oargrid_job_id)])
kavlan_global = None
for site, res in resources.iteritems():
if res['kavlan'] >= 10:
kavlan_global = {'kavlan': res['kavlan'],
'ip_mac': resources[site]['ip_mac'],
'site': site}
break
if kavlan_global:
resources['global'] = kavlan_global
return resources
def main():
copy_outputs('config.log', 'config.log')
args = parser.parse_args()
whoami = os.getlogin()
logger.info('whoami: %s', whoami)
jobids = []
if args.grid_job_id == None:
jobids = args.job_ids
else:
grid_job_id = int(args.grid_job_id[0])
jobids = ["%s:%d" % (site, job_id) for job_id, site in get_oargrid_job_oar_jobs(grid_job_id)]
logger.info('Using jobs %s', style.emph(' '.join(jobids)))
sites = [j.strip().split(':')[0] for j in jobids]
frontends = [str('frontend.'+s) for s in sites]
oar_ids = [j.strip().split(':')[1] for j in jobids]
jobids_list=[(int(j.strip().split(':')[1]),str(j.strip().split(':')[0])) for j in jobids]
# print sites
# print oar_ids
# print jobids_list
# print frontends
logger.info("Get list of associated nodes")
nodes = [ job_nodes for job in jobids_list for job_nodes in get_oar_job_nodes(*job) ]
# logger.info('%s', hosts_list(nodes))
logger.info('%s', nodes)
logger.info("Deploying %i nodes" % (len(nodes),))
deployed, undeployed = deploy(Deployment(nodes, env_name = "jessie-x64-nfs"))
logger.info("%i deployed, %i undeployed" % (len(deployed), len(undeployed)))
## Configure Host OSes
logger.info('Finalize node customization')
# use root to connect on the host
default_connection_params['user'] = '******'
## Copy local .ssh to remote nodes:
logger.info('Copy ssh entries into root of each node')
Put(nodes, ['/home/'+whoami+'/.ssh/id_rsa','/home/'+whoami+'/.ssh/id_rsa.pub'],'.ssh/.').run()
## Install missing packages
logger.info('| - Install Packages')
install_packages = TaktukRemote('export DEBIAN_MASTER=noninteractive ; export https_proxy="https://proxy:3128"; apt-get -o Acquire::Check-Valid-Until=false update && apt-get install -y --force-yes python-pip lynx openjdk-8-jdk uuid-runtime cpufrequtils kanif -o Acquire::Check-Valid-Until=false -o Dpkgtions::="--force-confdef" -o Dpkgtions::="--force-confold" ; pip install tabview', nodes).run()
## Fix ulimit and related stuffs
logger.info('| - set limit related stuffs')
cmd = 'ulimit -c unlimited; sysctl -w vm.max_map_count=331072 ; echo 120000 > /proc/sys/kernel/threads-max ; echo 600000 > /proc/sys/vm/max_map_count ; echo 200000 > /proc/sys/kernel/pid_max'
TaktukRemote(cmd, nodes).run()
## Copy the DHT-EXP hierarchy to the remote site
logger.info('Copy sloth and injector files on each NFS server involved in the experiment')
TaktukRemote('mkdir -p ~/SLOTH-EXP-TMP/', frontends, connection_params={'user': str(whoami)}).run()
TaktukPut(frontends, ['./SLOTH_HOME' ],'./SLOTH-EXP-TMP/.', connection_params={'user': str(whoami)}).run()
TaktukPut(frontends, ['./INJECTOR_HOME' ], './SLOTH-EXP-TMP/.', connection_params={'user': str(whoami)}).run()
## Prepare the address file for the sloth peers (please remind that the last node is dedicated for the injector
logger.info('Prepare the peers list')
f1 = open('./hosts.info', 'w')
f2 = open('./peers.info', 'w')
logger.info('enumerate %s', enumerate(nodes[:-1]))
i = 0
for j, node in enumerate(nodes[:-1]):
f1.write("%s\n" % (node.address))
for cores in range(get_host_attributes(node)['architecture']['smt_size']):
f2.write("%s:%d:%d\n" % (node.address, 3000 + i, 8000 + i))
i = i + 1
f1.close()
f2.close()
f = open('./service_node.info', 'w')
f.write("%s" % nodes[-1].address)
f.close()
logger.info("Nodes are now ready, you should launch ./runExperiment.sh ... from the lyon frontend")
logger.info("The list of sloth peers is in ./peers.info")
logger.info("The service node is in ./service_node.info")
logger.info("The injector will run on %s" % nodes[-1].address)
logger.info("The usual(max) command should be : ./runExperiment.sh in_vivo %d %s/peers.info %s" % (i,os.getcwd(),nodes[-1].address))
