X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=src%2Fnepi%2Fresources%2Flinux%2Fnode.py;h=232a9df5824b7a1b7c570f0e198c1d6ff05d37c3;hb=a55b5cf0818ced6ab8dc3e88fad7f6616675e826;hp=b547ce55941bad923c6a46aef5fe2d9be264a172;hpb=09ac796bac9aa2c41c5ad830f404fe128fffb22d;p=nepi.git diff --git a/src/nepi/resources/linux/node.py b/src/nepi/resources/linux/node.py index b547ce55..232a9df5 100644 --- a/src/nepi/resources/linux/node.py +++ b/src/nepi/resources/linux/node.py @@ -28,6 +28,7 @@ import collections import os import random import re +import socket import tempfile import time import threading @@ -149,44 +150,54 @@ class LinuxNode(ResourceManager): @classmethod def _register_attributes(cls): hostname = Attribute("hostname", "Hostname of the machine", - flags = Flags.ExecReadOnly) + flags = Flags.Design) username = Attribute("username", "Local account username", flags = Flags.Credential) - port = Attribute("port", "SSH port", flags = Flags.ExecReadOnly) + port = Attribute("port", "SSH port", flags = Flags.Design) home = Attribute("home", "Experiment home directory to store all experiment related files", - flags = Flags.ExecReadOnly) + flags = Flags.Design) identity = Attribute("identity", "SSH identity file", flags = Flags.Credential) server_key = Attribute("serverKey", "Server public key", - flags = Flags.ExecReadOnly) + flags = Flags.Design) 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) + flags = Flags.Design) 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) + flags = Flags.Design) clean_processes = Attribute("cleanProcesses", "Kill all running processes before starting experiment", type = Types.Bool, default = False, - flags = Flags.ExecReadOnly) + flags = Flags.Design) tear_down = Attribute("tearDown", "Bash script to be executed before " + \ "releasing the resource", - flags = Flags.ExecReadOnly) + flags = Flags.Design) + + gateway_user = Attribute("gatewayUser", "Gateway account username", + flags = Flags.Design) + + gateway = Attribute("gateway", "Hostname of the gateway machine", + flags = Flags.Design) + + ip = Attribute("ip", "Linux host public IP address. " + "Must not be modified by the user unless hostname is 'localhost'", + flags = Flags.Design) cls._register_attribute(hostname) cls._register_attribute(username) @@ -198,13 +209,16 @@ class LinuxNode(ResourceManager): cls._register_attribute(clean_experiment) cls._register_attribute(clean_processes) cls._register_attribute(tear_down) + cls._register_attribute(gateway_user) + cls._register_attribute(gateway) + cls._register_attribute(ip) 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 @@ -225,9 +239,13 @@ class LinuxNode(ResourceManager): home = os.path.join(self._home_dir, home) return home + @property + def nepi_home(self): + return os.path.join(self.home_dir, ".nepi") + @property def usr_dir(self): - return os.path.join(self.home_dir, "nepi-usr") + return os.path.join(self.nepi_home, "nepi-usr") @property def lib_dir(self): @@ -247,7 +265,7 @@ class LinuxNode(ResourceManager): @property def exp_dir(self): - return os.path.join(self.home_dir, "nepi-exp") + return os.path.join(self.nepi_home, "nepi-exp") @property def exp_home(self): @@ -266,7 +284,7 @@ class LinuxNode(ResourceManager): if self._os: return self._os - if (not self.get("hostname") or not self.get("username")): + if not self.localhost and not self.get("username"): msg = "Can't resolve OS, insufficient data " self.error(msg) raise RuntimeError, msg @@ -298,24 +316,16 @@ class LinuxNode(ResourceManager): # 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): @@ -328,7 +338,7 @@ class LinuxNode(ResourceManager): @property def localhost(self): - return self.get("hostname") in ['localhost', '127.0.0.7', '::1'] + return self.get("hostname") in ['localhost', '127.0.0.1', '::1'] def do_provision(self): # check if host is alive @@ -348,14 +358,23 @@ class LinuxNode(ResourceManager): if self.get("cleanExperiment"): self.clean_experiment() - # Create shared directory structure - self.mkdir(self.lib_dir) - self.mkdir(self.bin_dir) - self.mkdir(self.src_dir) - self.mkdir(self.share_dir) + # Create shared directory structure and node home directory + paths = [self.lib_dir, + self.bin_dir, + self.src_dir, + self.share_dir, + self.node_home] + + self.mkdir(paths) - # Create experiment node home directory - self.mkdir(self.node_home) + # Get Public IP address + if not self.get("ip"): + if self.localhost: + ip = socket.gethostbyname(socket.gethostname()) + else: + ip = socket.gethostbyname(self.get("hostname")) + + self.set("ip", ip) super(LinuxNode, self).do_provision() @@ -368,7 +387,7 @@ class LinuxNode(ResourceManager): # 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) @@ -397,24 +416,39 @@ class LinuxNode(ResourceManager): # TODO: Validate! return True - def clean_processes(self, killer = False): + def clean_processes(self): self.info("Cleaning up processes") + + if self.localhost: + return - 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 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) + if len(out) != 0: + 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 ; ") + 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) def clean_home(self): @@ -422,7 +456,7 @@ class LinuxNode(ResourceManager): """ self.info("Cleaning up home") - cmd = "cd %s ; find . -maxdepth 1 \( -name 'nepi-usr' -o -name 'nepi-exp' \) -execdir rm -rf {} + " % ( + cmd = "cd %s ; find . -maxdepth 1 -name \.nepi -execdir rm -rf {} + " % ( self.home_dir ) return self.execute(cmd, with_lock = True) @@ -442,13 +476,10 @@ class LinuxNode(ResourceManager): def execute(self, command, sudo = False, - stdin = None, env = None, tty = False, forward_x11 = False, - timeout = None, retry = 3, - err_on_timeout = True, connect_timeout = 30, strict_host_checking = False, persistent = True, @@ -461,29 +492,31 @@ class LinuxNode(ResourceManager): 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) else: if with_lock: + # If the execute command is blocking, we don't want to keep + # the node lock. This lock is used to avoid race conditions + # when creating the ControlMaster sockets. A more elegant + # solution is needed. with self._node_lock: (out, err), proc = sshfuncs.rexec( command, 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, identity = self.get("identity"), server_key = self.get("serverKey"), env = env, tty = tty, forward_x11 = forward_x11, - timeout = timeout, retry = retry, - err_on_timeout = err_on_timeout, connect_timeout = connect_timeout, persistent = persistent, blocking = blocking, @@ -495,17 +528,16 @@ class LinuxNode(ResourceManager): 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, identity = self.get("identity"), server_key = self.get("serverKey"), env = env, tty = tty, forward_x11 = forward_x11, - timeout = timeout, retry = retry, - err_on_timeout = err_on_timeout, connect_timeout = connect_timeout, persistent = persistent, blocking = blocking, @@ -521,19 +553,19 @@ class LinuxNode(ResourceManager): stdout = 'stdout', stderr = 'stderr', sudo = False, - tty = False): + tty = False, + strict_host_checking = False): self.debug("Running command '%s'" % command) if self.localhost: - (out, err), proc = execfuncs.lspawn(command, pidfile, - stdout = stdout, - stderr = stderr, - stdin = stdin, + (out, err), proc = execfuncs.lspawn(command, pidfile, home = home, create_home = create_home, - sudo = sudo, - user = user) + stdin = stdin or '/dev/null', + stdout = stdout or '/dev/null', + stderr = stderr or '/dev/null', + sudo = sudo) else: with self._node_lock: (out, err), proc = sshfuncs.rspawn( @@ -541,17 +573,20 @@ class LinuxNode(ResourceManager): pidfile = pidfile, home = home, create_home = create_home, - stdin = stdin if stdin is not None else '/dev/null', - stdout = stdout if stdout else '/dev/null', - stderr = stderr if stderr else '/dev/null', + stdin = stdin or '/dev/null', + stdout = stdout or '/dev/null', + stderr = stderr or '/dev/null', sudo = sudo, 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"), - tty = tty + tty = tty, + strict_host_checking = strict_host_checking ) return (out, err), proc @@ -566,9 +601,12 @@ class LinuxNode(ResourceManager): 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") + server_key = self.get("serverKey"), + strict_host_checking = False ) return pidtuple @@ -583,9 +621,12 @@ class LinuxNode(ResourceManager): 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") + server_key = self.get("serverKey"), + strict_host_checking = False ) return status @@ -605,24 +646,28 @@ class LinuxNode(ResourceManager): 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"), - server_key = self.get("serverKey") + server_key = self.get("serverKey"), + strict_host_checking = False ) return (out, err), proc def copy(self, src, dst): if self.localhost: - (out, err), proc = execfuncs.lcopy(source, dest, - recursive = True, - strict_host_checking = False) + (out, err), proc = execfuncs.lcopy(src, dst, + recursive = True) else: with self._node_lock: (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, @@ -630,14 +675,21 @@ class LinuxNode(ResourceManager): return (out, err), proc - def upload(self, src, dst, text = False, overwrite = True): + def upload(self, src, dst, text = False, overwrite = True, + raise_on_error = True): """ Copy content to destination - src content to copy. Can be a local file, directory or a list of files + src string with the content to copy. Can be: + - plain text + - a string with the path to a local file + - a string with a semi-colon separeted list of local files + - a string with a local directory - dst destination path on the remote host (remote is always self.host) + dst string with destination path on the remote host (remote is + always self.host) - text src is text input, it must be stored into a temp file before uploading + text src is text input, it must be stored into a temp file before + uploading """ # If source is a string input f = None @@ -650,29 +702,50 @@ class LinuxNode(ResourceManager): src = f.name # If dst files should not be overwritten, check that the files do not - # exits already + # exits already + if isinstance(src, str): + src = map(str.strip, src.split(";")) + if overwrite == False: src = self.filter_existing_files(src, dst) if not src: - return ("", ""), None + return ("", ""), None if not self.localhost: # Build destination as @: dst = "%s@%s:%s" % (self.get("username"), self.get("hostname"), dst) - result = self.copy(src, dst) + ((out, err), proc) = self.copy(src, dst) # clean up temp file if f: os.remove(f.name) - return result + if err: + msg = " Failed to upload files - src: %s dst: %s" % (";".join(src), dst) + self.error(msg, out, err) + + msg = "%s out: %s err: %s" % (msg, out, err) + if raise_on_error: + raise RuntimeError, msg + + return ((out, err), proc) - def download(self, src, dst): + def download(self, src, dst, raise_on_error = True): if not self.localhost: # Build destination as @: src = "%s@%s:%s" % (self.get("username"), self.get("hostname"), src) - return self.copy(src, dst) + + ((out, err), proc) = self.copy(src, dst) + + if err: + msg = " Failed to download files - src: %s dst: %s" % (";".join(src), dst) + self.error(msg, out, err) + + if raise_on_error: + raise RuntimeError, msg + + return ((out, err), proc) def install_packages_command(self, packages): command = "" @@ -687,7 +760,8 @@ class LinuxNode(ResourceManager): return command - def install_packages(self, packages, home, run_home = None): + def install_packages(self, packages, home, run_home = None, + raise_on_error = True): """ Install packages in the Linux host. 'home' is the directory to upload the package installation script. @@ -704,11 +778,12 @@ class LinuxNode(ResourceManager): stdout = "instpkg_stdout", stderr = "instpkg_stderr", overwrite = False, - raise_on_error = True) + raise_on_error = raise_on_error) return (out, err), proc - def remove_packages(self, packages, home, run_home = None): + def remove_packages(self, packages, home, run_home = None, + raise_on_error = True): """ Uninstall packages from the Linux host. 'home' is the directory to upload the package un-installation script. @@ -732,18 +807,35 @@ class LinuxNode(ResourceManager): stdout = "rmpkg_stdout", stderr = "rmpkg_stderr", overwrite = False, - raise_on_error = True) + raise_on_error = raise_on_error) return (out, err), proc - def mkdir(self, path, clean = False): + def mkdir(self, paths, clean = False): + """ Paths is either a single remote directory path to create, + or a list of directories to create. + """ if clean: - self.rmdir(path) + self.rmdir(paths) + + if isinstance(paths, str): + paths = [paths] - return self.execute("mkdir -p %s" % path, with_lock = True) + cmd = " ; ".join(map(lambda path: "mkdir -p %s" % path, paths)) - def rmdir(self, path): - return self.execute("rm -rf %s" % path, with_lock = True) + return self.execute(cmd, with_lock = True) + + def rmdir(self, paths): + """ Paths is either a single remote directory path to delete, + or a list of directories to delete. + """ + + if isinstance(paths, str): + paths = [paths] + + cmd = " ; ".join(map(lambda path: "rm -rf %s" % path, paths)) + + return self.execute(cmd, with_lock = True) def run_and_wait(self, command, home, shfile = "cmd.sh", @@ -756,7 +848,7 @@ class LinuxNode(ResourceManager): stderr = "stderr", sudo = False, tty = False, - raise_on_error = False): + raise_on_error = True): """ Uploads the 'command' to a bash script in the host. Then runs the script detached in background in the host, and @@ -963,35 +1055,25 @@ class LinuxNode(ResourceManager): 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 @@ -1000,37 +1082,28 @@ class LinuxNode(ResourceManager): # 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): """ Removes files that already exist in the Linux host from src list """ # construct a dictionary with { dst: src } - dests = dict(map(lambda x: ( os.path.join(dst, os.path.basename(x) ), x ), - src.strip().split(" ") ) ) if src.strip().find(" ") != -1 else dict({dst: src}) + dests = dict(map(lambda s: (os.path.join(dst, os.path.basename(s)), s), src)) \ + if len(src) > 1 else dict({dst: src[0]}) command = [] for d in dests.keys(): @@ -1045,7 +1118,7 @@ class LinuxNode(ResourceManager): del dests[d] if not dests: - return "" + return [] - return " ".join(dests.values()) + return dests.values()