#
# Author: Alina Quereilhac <alina.quereilhac@inria.fr>
-from nepi.execution.attribute import Attribute, Flags
-from nepi.execution.resource import ResourceManager, clsinit, ResourceState, \
- reschedule_delay
+from nepi.execution.attribute import Attribute, Flags, Types
+from nepi.execution.resource import ResourceManager, clsinit_copy, \
+ ResourceState, reschedule_delay
from nepi.resources.linux import rpmfuncs, debfuncs
from nepi.util import sshfuncs, execfuncs
from nepi.util.sshfuncs import ProcStatus
UBUNTU = "ubuntu"
DEBIAN = "debian"
-@clsinit
+@clsinit_copy
class LinuxNode(ResourceManager):
"""
.. class:: Class Args :
clean_home = Attribute("cleanHome", "Remove all nepi files and directories "
" from node home folder before starting experiment",
+ type = Types.Bool,
+ default = False,
flags = Flags.ExecReadOnly)
clean_experiment = Attribute("cleanExperiment", "Remove all files and directories "
" from a previous same experiment, before the new experiment starts",
+ type = Types.Bool,
+ default = False,
flags = Flags.ExecReadOnly)
clean_processes = Attribute("cleanProcesses",
"Kill all running processes before starting experiment",
+ type = Types.Bool,
+ default = False,
flags = Flags.ExecReadOnly)
tear_down = Attribute("tearDown", "Bash script to be executed before " + \
"releasing the resource",
flags = Flags.ExecReadOnly)
+ gateway_user = Attribute("gatewayUser", "Gateway account username",
+ flags = Flags.ExecReadOnly)
+
+ gateway = Attribute("gateway", "Hostname of the gateway machine",
+ flags = Flags.ExecReadOnly)
+
cls._register_attribute(hostname)
cls._register_attribute(username)
cls._register_attribute(port)
cls._register_attribute(clean_experiment)
cls._register_attribute(clean_processes)
cls._register_attribute(tear_down)
+ cls._register_attribute(gateway_user)
+ cls._register_attribute(gateway)
def __init__(self, ec, guid):
super(LinuxNode, self).__init__(ec, guid)
self._os = None
# home directory at Linux host
self._home_dir = ""
-
+
# lock to prevent concurrent applications on the same node,
# to execute commands at the same time. There are potential
# concurrency issues when using SSH to a same host from
self._os = OSType.FEDORA_12
elif out.find("Fedora release 14") == 0:
self._os = OSType.FEDORA_14
+ elif out.find("Fedora release") == 0:
+ self._os = OSType.FEDORA
elif out.find("Debian") == 0:
self._os = OSType.DEBIAN
elif out.find("Ubuntu") ==0:
# To work arround this, repeat the operation N times or
# until the result is not empty string
out = ""
- retrydelay = 1.0
- for i in xrange(2):
- try:
- (out, err), proc = self.execute("cat /etc/issue",
- retry = 5,
- with_lock = True,
- blocking = True)
-
- if out.strip() != "":
- return out
- except:
- trace = traceback.format_exc()
- msg = "Error detecting OS: %s " % trace
- self.error(msg, out, err)
- return False
-
- time.sleep(min(30.0, retrydelay))
- retrydelay *= 1.5
-
+ try:
+ (out, err), proc = self.execute("cat /etc/issue",
+ with_lock = True,
+ blocking = True)
+ except:
+ trace = traceback.format_exc()
+ msg = "Error detecting OS: %s " % trace
+ self.error(msg, out, err)
+
+ return out
@property
def use_deb(self):
def localhost(self):
return self.get("hostname") in ['localhost', '127.0.0.7', '::1']
- def provision(self):
+ def do_provision(self):
# check if host is alive
if not self.is_alive():
- self.fail()
-
msg = "Deploy failed. Unresponsive node %s" % self.get("hostname")
self.error(msg)
raise RuntimeError, msg
# Create experiment node home directory
self.mkdir(self.node_home)
- super(LinuxNode, self).provision()
+ super(LinuxNode, self).do_provision()
- def deploy(self):
+ def do_deploy(self):
if self.state == ResourceState.NEW:
- try:
- self.discover()
- self.provision()
- except:
- self.fail()
- raise
+ self.info("Deploying node")
+ self.do_discover()
+ self.do_provision()
# Node needs to wait until all associated interfaces are
# ready before it can finalize deployment
from nepi.resources.linux.interface import LinuxInterface
- ifaces = self.get_connected(LinuxInterface.rtype())
+ ifaces = self.get_connected(LinuxInterface.get_rtype())
for iface in ifaces:
if iface.state < ResourceState.READY:
self.ec.schedule(reschedule_delay, self.deploy)
return
- super(LinuxNode, self).deploy()
+ super(LinuxNode, self).do_deploy()
- def release(self):
- # Node needs to wait until all associated RMs are released
- # to be released
+ def do_release(self):
rms = self.get_connected()
for rm in rms:
- if rm.state < ResourceState.STOPPED:
+ # Node needs to wait until all associated RMs are released
+ # before it can be released
+ if rm.state != ResourceState.RELEASED:
self.ec.schedule(reschedule_delay, self.release)
return
self.clean_processes()
- super(LinuxNode, self).release()
+ super(LinuxNode, self).do_release()
def valid_connection(self, guid):
# TODO: Validate!
return True
- def clean_processes(self, killer = False):
+ def clean_processes(self):
self.info("Cleaning up processes")
-
- if killer:
- # Hardcore kill
- cmd = ("sudo -S killall python tcpdump || /bin/true ; " +
- "sudo -S killall python tcpdump || /bin/true ; " +
- "sudo -S kill $(ps -N -T -o pid --no-heading | grep -v $PPID | sort) || /bin/true ; " +
- "sudo -S killall -u root || /bin/true ; " +
- "sudo -S killall -u root || /bin/true ; ")
- else:
- # Be gentler...
+
+ if self.get("username") != 'root':
cmd = ("sudo -S killall tcpdump || /bin/true ; " +
- "sudo -S killall tcpdump || /bin/true ; " +
- "sudo -S killall -u %s || /bin/true ; " % self.get("username") +
+ "sudo -S kill $(ps aux | grep '[n]epi' | awk '{print $2}') || /bin/true ; " +
"sudo -S killall -u %s || /bin/true ; " % self.get("username"))
+ else:
+ if self.state >= ResourceState.READY:
+ import pickle
+ pids = pickle.load(open("/tmp/save.proc", "rb"))
+ pids_temp = dict()
+ ps_aux = "ps aux |awk '{print $2,$11}'"
+ (out, err), proc = self.execute(ps_aux)
+ for line in out.strip().split("\n"):
+ parts = line.strip().split(" ")
+ pids_temp[parts[0]] = parts[1]
+ kill_pids = set(pids_temp.items()) - set(pids.items())
+ kill_pids = ' '.join(dict(kill_pids).keys())
+
+ cmd = ("killall tcpdump || /bin/true ; " +
+ "kill $(ps aux | grep '[n]epi' | awk '{print $2}') || /bin/true ; " +
+ "kill %s || /bin/true ; " % kill_pids)
+ else:
+ cmd = ("killall tcpdump || /bin/true ; " +
+ "kill $(ps aux | grep '[n]epi' | awk '{print $2}') || /bin/true ; ")
out = err = ""
- (out, err), proc = self.execute(cmd, retry = 1, with_lock = True)
-
+ (out, err), proc = self.execute(cmd, retry = 1, with_lock = True)
+
def clean_home(self):
""" Cleans all NEPI related folders in the Linux host
"""
if self.localhost:
(out, err), proc = execfuncs.lexec(command,
- user = user,
+ user = self.get("username"), # still problem with localhost
sudo = sudo,
stdin = stdin,
env = env)
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
sudo = sudo,
stdin = stdin,
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
sudo = sudo,
stdin = stdin,
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
identity = self.get("identity"),
server_key = self.get("serverKey"),
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
identity = self.get("identity"),
server_key = self.get("serverKey")
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
identity = self.get("identity"),
server_key = self.get("serverKey")
host = self.get("hostname"),
user = self.get("username"),
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
agent = True,
sudo = sudo,
identity = self.get("identity"),
(out, err), proc = sshfuncs.rcopy(
src, dst,
port = self.get("port"),
+ gwuser = self.get("gatewayUser"),
+ gw = self.get("gateway"),
identity = self.get("identity"),
server_key = self.get("serverKey"),
recursive = True,
return (out, err), proc
-
def upload(self, src, dst, text = False, overwrite = True):
""" Copy content to destination
return True
out = err = ""
+ msg = "Unresponsive host. Wrong answer. "
+
# The underlying SSH layer will sometimes return an empty
# output (even if the command was executed without errors).
# To work arround this, repeat the operation N times or
# until the result is not empty string
- retrydelay = 1.0
- for i in xrange(2):
- try:
- (out, err), proc = self.execute("echo 'ALIVE'",
- retry = 5,
- blocking = True,
- with_lock = True)
-
- if out.find("ALIVE") > -1:
- return True
- except:
- trace = traceback.format_exc()
- msg = "Unresponsive host. Error reaching host: %s " % trace
- self.error(msg, out, err)
- return False
-
- time.sleep(min(30.0, retrydelay))
- retrydelay *= 1.5
+ try:
+ (out, err), proc = self.execute("echo 'ALIVE'",
+ blocking = True,
+ with_lock = True)
+
+ if out.find("ALIVE") > -1:
+ return True
+ except:
+ trace = traceback.format_exc()
+ msg = "Unresponsive host. Error reaching host: %s " % trace
- if out.find("ALIVE") > -1:
- return True
- else:
- msg = "Unresponsive host. Wrong answer. "
- self.error(msg, out, err)
- return False
+ self.error(msg, out, err)
+ return False
def find_home(self):
""" Retrieves host home directory
# output (even if the command was executed without errors).
# To work arround this, repeat the operation N times or
# until the result is not empty string
- retrydelay = 1.0
- for i in xrange(2):
- try:
- (out, err), proc = self.execute("echo ${HOME}",
- retry = 5,
- blocking = True,
- with_lock = True)
-
- if out.strip() != "":
- self._home_dir = out.strip()
- break
- except:
- trace = traceback.format_exc()
- msg = "Impossible to retrieve HOME directory" % trace
- self.error(msg, out, err)
- return False
-
- time.sleep(min(30.0, retrydelay))
- retrydelay *= 1.5
+ msg = "Impossible to retrieve HOME directory"
+ try:
+ (out, err), proc = self.execute("echo ${HOME}",
+ blocking = True,
+ with_lock = True)
+
+ if out.strip() != "":
+ self._home_dir = out.strip()
+ except:
+ trace = traceback.format_exc()
+ msg = "Impossible to retrieve HOME directory %s" % trace
if not self._home_dir:
- msg = "Impossible to retrieve HOME directory"
- self.error(msg, out, err)
+ self.error(msg)
raise RuntimeError, msg
def filter_existing_files(self, src, dst):
git://git.onelab.eu / nepi.git / blobdiff