2 # NEPI, a framework to manage network experiments
3 # Copyright (C) 2013 INRIA
5 # This program is free software: you can redistribute it and/or modify
6 # it under the terms of the GNU General Public License as published by
7 # the Free Software Foundation, either version 3 of the License, or
8 # (at your option) any later version.
10 # This program is distributed in the hope that it will be useful,
11 # but WITHOUT ANY WARRANTY; without even the implied warranty of
12 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 # GNU General Public License for more details.
15 # You should have received a copy of the GNU General Public License
16 # along with this program. If not, see <http://www.gnu.org/licenses/>.
18 # Author: Alina Quereilhac <alina.quereilhac@inria.fr>
20 from nepi.execution.attribute import Attribute, Flags, Types
21 from nepi.execution.resource import ResourceManager, clsinit_copy, \
22 ResourceState, reschedule_delay
23 from nepi.resources.linux import rpmfuncs, debfuncs
24 from nepi.util import sshfuncs, execfuncs
25 from nepi.util.sshfuncs import ProcStatus
36 # TODO: Unify delays!!
37 # TODO: Validate outcome of uploads!!
41 Error codes that the rexitcode function can return if unable to
42 check the exit code of a spawned process
51 Supported flavors of Linux OS
61 class LinuxNode(ResourceManager):
63 .. class:: Class Args :
65 :param ec: The Experiment controller
66 :type ec: ExperimentController
67 :param guid: guid of the RM
72 There are different ways in which commands can be executed using the
73 LinuxNode interface (i.e. 'execute' - blocking and non blocking, 'run',
78 * 'execute' (blocking mode) :
80 HOW IT WORKS: 'execute', forks a process and run the
81 command, synchronously, attached to the terminal, in
83 The execute method will block until the command returns
84 the result on 'out', 'err' (so until it finishes executing).
86 USAGE: short-lived commands that must be executed attached
87 to a terminal and in foreground, for which it IS necessary
88 to block until the command has finished (e.g. if you want
89 to run 'ls' or 'cat').
91 * 'execute' (NON blocking mode - blocking = False) :
93 HOW IT WORKS: Same as before, except that execute method
94 will return immediately (even if command still running).
96 USAGE: long-lived commands that must be executed attached
97 to a terminal and in foreground, but for which it is not
98 necessary to block until the command has finished. (e.g.
99 start an application using X11 forwarding)
103 HOW IT WORKS: Connects to the host ( using SSH if remote)
104 and launches the command in background, detached from any
105 terminal (daemonized), and returns. The command continues to
106 run remotely, but since it is detached from the terminal,
107 its pipes (stdin, stdout, stderr) can't be redirected to the
108 console (as normal non detached processes would), and so they
109 are explicitly redirected to files. The pidfile is created as
110 part of the process of launching the command. The pidfile
111 holds the pid and ppid of the process forked in background,
112 so later on it is possible to check whether the command is still
115 USAGE: long-lived commands that can run detached in background,
116 for which it is NOT necessary to block (wait) until the command
117 has finished. (e.g. start an application that is not using X11
118 forwarding. It can run detached and remotely in background)
122 HOW IT WORKS: Similar to 'run' except that it 'blocks' until
123 the command has finished execution. It also checks whether
124 errors occurred during runtime by reading the exitcode file,
125 which contains the exit code of the command that was run
126 (checking stderr only is not always reliable since many
127 commands throw debugging info to stderr and the only way to
128 automatically know whether an error really happened is to
129 check the process exit code).
131 Another difference with respect to 'run', is that instead
132 of directly executing the command as a bash command line,
133 it uploads the command to a bash script and runs the script.
134 This allows to use the bash script to debug errors, since
135 it remains at the remote host and can be run manually to
138 USAGE: medium-lived commands that can run detached in
139 background, for which it IS necessary to block (wait) until
140 the command has finished. (e.g. Package installation,
141 source compilation, file download, etc)
145 _help = "Controls Linux host machines ( either localhost or a host " \
146 "that can be accessed using a SSH key)"
147 _backend_type = "linux"
150 def _register_attributes(cls):
151 hostname = Attribute("hostname", "Hostname of the machine",
152 flags = Flags.ExecReadOnly)
154 username = Attribute("username", "Local account username",
155 flags = Flags.Credential)
157 port = Attribute("port", "SSH port", flags = Flags.ExecReadOnly)
159 home = Attribute("home",
160 "Experiment home directory to store all experiment related files",
161 flags = Flags.ExecReadOnly)
163 identity = Attribute("identity", "SSH identity file",
164 flags = Flags.Credential)
166 server_key = Attribute("serverKey", "Server public key",
167 flags = Flags.ExecReadOnly)
169 clean_home = Attribute("cleanHome", "Remove all nepi files and directories "
170 " from node home folder before starting experiment",
173 flags = Flags.ExecReadOnly)
175 clean_experiment = Attribute("cleanExperiment", "Remove all files and directories "
176 " from a previous same experiment, before the new experiment starts",
179 flags = Flags.ExecReadOnly)
181 clean_processes = Attribute("cleanProcesses",
182 "Kill all running processes before starting experiment",
185 flags = Flags.ExecReadOnly)
187 tear_down = Attribute("tearDown", "Bash script to be executed before " + \
188 "releasing the resource",
189 flags = Flags.ExecReadOnly)
191 cls._register_attribute(hostname)
192 cls._register_attribute(username)
193 cls._register_attribute(port)
194 cls._register_attribute(home)
195 cls._register_attribute(identity)
196 cls._register_attribute(server_key)
197 cls._register_attribute(clean_home)
198 cls._register_attribute(clean_experiment)
199 cls._register_attribute(clean_processes)
200 cls._register_attribute(tear_down)
202 def __init__(self, ec, guid):
203 super(LinuxNode, self).__init__(ec, guid)
205 # home directory at Linux host
208 # lock to prevent concurrent applications on the same node,
209 # to execute commands at the same time. There are potential
210 # concurrency issues when using SSH to a same host from
211 # multiple threads. There are also possible operational
212 # issues, e.g. an application querying the existence
213 # of a file or folder prior to its creation, and another
214 # application creating the same file or folder in between.
215 self._node_lock = threading.Lock()
217 def log_message(self, msg):
218 return " guid %d - host %s - %s " % (self.guid,
219 self.get("hostname"), msg)
223 home = self.get("home") or ""
224 if not home.startswith("/"):
225 home = os.path.join(self._home_dir, home)
230 return os.path.join(self.home_dir, "nepi-usr")
234 return os.path.join(self.usr_dir, "lib")
238 return os.path.join(self.usr_dir, "bin")
242 return os.path.join(self.usr_dir, "src")
246 return os.path.join(self.usr_dir, "share")
250 return os.path.join(self.home_dir, "nepi-exp")
254 return os.path.join(self.exp_dir, self.ec.exp_id)
258 return os.path.join(self.exp_home, "node-%d" % self.guid)
262 return os.path.join(self.node_home, self.ec.run_id)
269 if (not self.get("hostname") or not self.get("username")):
270 msg = "Can't resolve OS, insufficient data "
272 raise RuntimeError, msg
276 if out.find("Fedora release 8") == 0:
277 self._os = OSType.FEDORA_8
278 elif out.find("Fedora release 12") == 0:
279 self._os = OSType.FEDORA_12
280 elif out.find("Fedora release 14") == 0:
281 self._os = OSType.FEDORA_14
282 elif out.find("Fedora release") == 0:
283 self._os = OSType.FEDORA
284 elif out.find("Debian") == 0:
285 self._os = OSType.DEBIAN
286 elif out.find("Ubuntu") ==0:
287 self._os = OSType.UBUNTU
289 msg = "Unsupported OS"
291 raise RuntimeError, "%s - %s " %( msg, out )
296 # The underlying SSH layer will sometimes return an empty
297 # output (even if the command was executed without errors).
298 # To work arround this, repeat the operation N times or
299 # until the result is not empty string
302 (out, err), proc = self.execute("cat /etc/issue",
306 trace = traceback.format_exc()
307 msg = "Error detecting OS: %s " % trace
308 self.error(msg, out, err)
314 return self.os in [OSType.DEBIAN, OSType.UBUNTU]
318 return self.os in [OSType.FEDORA_12, OSType.FEDORA_14, OSType.FEDORA_8,
323 return self.get("hostname") in ['localhost', '127.0.0.7', '::1']
325 def do_provision(self):
326 # check if host is alive
327 if not self.is_alive():
328 msg = "Deploy failed. Unresponsive node %s" % self.get("hostname")
330 raise RuntimeError, msg
334 if self.get("cleanProcesses"):
335 self.clean_processes()
337 if self.get("cleanHome"):
340 if self.get("cleanExperiment"):
341 self.clean_experiment()
343 # Create shared directory structure
344 self.mkdir(self.lib_dir)
345 self.mkdir(self.bin_dir)
346 self.mkdir(self.src_dir)
347 self.mkdir(self.share_dir)
349 # Create experiment node home directory
350 self.mkdir(self.node_home)
352 super(LinuxNode, self).do_provision()
355 if self.state == ResourceState.NEW:
356 self.info("Deploying node")
360 # Node needs to wait until all associated interfaces are
361 # ready before it can finalize deployment
362 from nepi.resources.linux.interface import LinuxInterface
363 ifaces = self.get_connected(LinuxInterface.get_rtype())
365 if iface.state < ResourceState.READY:
366 self.ec.schedule(reschedule_delay, self.deploy)
369 super(LinuxNode, self).do_deploy()
371 def do_release(self):
372 rms = self.get_connected()
374 # Node needs to wait until all associated RMs are released
375 # before it can be released
376 if rm.state != ResourceState.RELEASED:
377 self.ec.schedule(reschedule_delay, self.release)
380 tear_down = self.get("tearDown")
382 self.execute(tear_down)
384 self.clean_processes()
386 super(LinuxNode, self).do_release()
388 def valid_connection(self, guid):
392 def clean_processes(self):
393 self.info("Cleaning up processes")
395 cmd = ("sudo -S killall tcpdump || /bin/true ; " +
396 "sudo -S kill $(ps aux | grep '[n]epi' | awk '{print $2}') || /bin/true ; " +
397 "sudo -S killall -u %s || /bin/true ; " % self.get("username"))
400 (out, err), proc = self.execute(cmd, retry = 1, with_lock = True)
402 def clean_home(self):
403 """ Cleans all NEPI related folders in the Linux host
405 self.info("Cleaning up home")
407 cmd = "cd %s ; find . -maxdepth 1 \( -name 'nepi-usr' -o -name 'nepi-exp' \) -execdir rm -rf {} + " % (
410 return self.execute(cmd, with_lock = True)
412 def clean_experiment(self):
413 """ Cleans all experiment related files in the Linux host.
414 It preserves NEPI files and folders that have a multi experiment
417 self.info("Cleaning up experiment files")
419 cmd = "cd %s ; find . -maxdepth 1 -name '%s' -execdir rm -rf {} + " % (
423 return self.execute(cmd, with_lock = True)
425 def execute(self, command,
433 err_on_timeout = True,
434 connect_timeout = 30,
435 strict_host_checking = False,
440 """ Notice that this invocation will block until the
441 execution finishes. If this is not the desired behavior,
442 use 'run' instead."""
445 (out, err), proc = execfuncs.lexec(command,
452 with self._node_lock:
453 (out, err), proc = sshfuncs.rexec(
455 host = self.get("hostname"),
456 user = self.get("username"),
457 port = self.get("port"),
461 identity = self.get("identity"),
462 server_key = self.get("serverKey"),
465 forward_x11 = forward_x11,
468 err_on_timeout = err_on_timeout,
469 connect_timeout = connect_timeout,
470 persistent = persistent,
472 strict_host_checking = strict_host_checking
475 (out, err), proc = sshfuncs.rexec(
477 host = self.get("hostname"),
478 user = self.get("username"),
479 port = self.get("port"),
483 identity = self.get("identity"),
484 server_key = self.get("serverKey"),
487 forward_x11 = forward_x11,
490 err_on_timeout = err_on_timeout,
491 connect_timeout = connect_timeout,
492 persistent = persistent,
494 strict_host_checking = strict_host_checking
497 return (out, err), proc
499 def run(self, command, home,
508 self.debug("Running command '%s'" % command)
511 (out, err), proc = execfuncs.lspawn(command, pidfile,
516 create_home = create_home,
520 with self._node_lock:
521 (out, err), proc = sshfuncs.rspawn(
525 create_home = create_home,
526 stdin = stdin if stdin is not None else '/dev/null',
527 stdout = stdout if stdout else '/dev/null',
528 stderr = stderr if stderr else '/dev/null',
530 host = self.get("hostname"),
531 user = self.get("username"),
532 port = self.get("port"),
534 identity = self.get("identity"),
535 server_key = self.get("serverKey"),
539 return (out, err), proc
541 def getpid(self, home, pidfile = "pidfile"):
543 pidtuple = execfuncs.lgetpid(os.path.join(home, pidfile))
545 with self._node_lock:
546 pidtuple = sshfuncs.rgetpid(
547 os.path.join(home, pidfile),
548 host = self.get("hostname"),
549 user = self.get("username"),
550 port = self.get("port"),
552 identity = self.get("identity"),
553 server_key = self.get("serverKey")
558 def status(self, pid, ppid):
560 status = execfuncs.lstatus(pid, ppid)
562 with self._node_lock:
563 status = sshfuncs.rstatus(
565 host = self.get("hostname"),
566 user = self.get("username"),
567 port = self.get("port"),
569 identity = self.get("identity"),
570 server_key = self.get("serverKey")
575 def kill(self, pid, ppid, sudo = False):
578 status = self.status(pid, ppid)
580 if status == sshfuncs.ProcStatus.RUNNING:
582 (out, err), proc = execfuncs.lkill(pid, ppid, sudo)
584 with self._node_lock:
585 (out, err), proc = sshfuncs.rkill(
587 host = self.get("hostname"),
588 user = self.get("username"),
589 port = self.get("port"),
592 identity = self.get("identity"),
593 server_key = self.get("serverKey")
596 return (out, err), proc
598 def copy(self, src, dst):
600 (out, err), proc = execfuncs.lcopy(source, dest,
602 strict_host_checking = False)
604 with self._node_lock:
605 (out, err), proc = sshfuncs.rcopy(
607 port = self.get("port"),
608 identity = self.get("identity"),
609 server_key = self.get("serverKey"),
611 strict_host_checking = False)
613 return (out, err), proc
615 def upload(self, src, dst, text = False, overwrite = True):
616 """ Copy content to destination
618 src content to copy. Can be a local file, directory or a list of files
620 dst destination path on the remote host (remote is always self.host)
622 text src is text input, it must be stored into a temp file before uploading
624 # If source is a string input
626 if text and not os.path.isfile(src):
627 # src is text input that should be uploaded as file
628 # create a temporal file with the content to upload
629 f = tempfile.NamedTemporaryFile(delete=False)
634 # If dst files should not be overwritten, check that the files do not
636 if overwrite == False:
637 src = self.filter_existing_files(src, dst)
639 return ("", ""), None
641 if not self.localhost:
642 # Build destination as <user>@<server>:<path>
643 dst = "%s@%s:%s" % (self.get("username"), self.get("hostname"), dst)
645 result = self.copy(src, dst)
653 def download(self, src, dst):
654 if not self.localhost:
655 # Build destination as <user>@<server>:<path>
656 src = "%s@%s:%s" % (self.get("username"), self.get("hostname"), src)
657 return self.copy(src, dst)
659 def install_packages_command(self, packages):
662 command = rpmfuncs.install_packages_command(self.os, packages)
664 command = debfuncs.install_packages_command(self.os, packages)
666 msg = "Error installing packages ( OS not known ) "
667 self.error(msg, self.os)
668 raise RuntimeError, msg
672 def install_packages(self, packages, home, run_home = None):
673 """ Install packages in the Linux host.
675 'home' is the directory to upload the package installation script.
676 'run_home' is the directory from where to execute the script.
678 command = self.install_packages_command(packages)
680 run_home = run_home or home
682 (out, err), proc = self.run_and_wait(command, run_home,
683 shfile = os.path.join(home, "instpkg.sh"),
684 pidfile = "instpkg_pidfile",
685 ecodefile = "instpkg_exitcode",
686 stdout = "instpkg_stdout",
687 stderr = "instpkg_stderr",
689 raise_on_error = True)
691 return (out, err), proc
693 def remove_packages(self, packages, home, run_home = None):
694 """ Uninstall packages from the Linux host.
696 'home' is the directory to upload the package un-installation script.
697 'run_home' is the directory from where to execute the script.
700 command = rpmfuncs.remove_packages_command(self.os, packages)
702 command = debfuncs.remove_packages_command(self.os, packages)
704 msg = "Error removing packages ( OS not known ) "
706 raise RuntimeError, msg
708 run_home = run_home or home
710 (out, err), proc = self.run_and_wait(command, run_home,
711 shfile = os.path.join(home, "rmpkg.sh"),
712 pidfile = "rmpkg_pidfile",
713 ecodefile = "rmpkg_exitcode",
714 stdout = "rmpkg_stdout",
715 stderr = "rmpkg_stderr",
717 raise_on_error = True)
719 return (out, err), proc
721 def mkdir(self, path, clean = False):
725 return self.execute("mkdir -p %s" % path, with_lock = True)
727 def rmdir(self, path):
728 return self.execute("rm -rf %s" % path, with_lock = True)
730 def run_and_wait(self, command, home,
735 ecodefile = "exitcode",
741 raise_on_error = False):
743 Uploads the 'command' to a bash script in the host.
744 Then runs the script detached in background in the host, and
745 busy-waites until the script finishes executing.
748 if not shfile.startswith("/"):
749 shfile = os.path.join(home, shfile)
751 self.upload_command(command,
753 ecodefile = ecodefile,
755 overwrite = overwrite)
757 command = "bash %s" % shfile
758 # run command in background in remote host
759 (out, err), proc = self.run(command, home,
767 # check no errors occurred
769 msg = " Failed to run command '%s' " % command
770 self.error(msg, out, err)
772 raise RuntimeError, msg
774 # Wait for pid file to be generated
775 pid, ppid = self.wait_pid(
778 raise_on_error = raise_on_error)
780 # wait until command finishes to execute
781 self.wait_run(pid, ppid)
783 (eout, err), proc = self.check_errors(home,
784 ecodefile = ecodefile,
787 # Out is what was written in the stderr file
789 msg = " Failed to run command '%s' " % command
790 self.error(msg, eout, err)
793 raise RuntimeError, msg
795 (out, oerr), proc = self.check_output(home, stdout)
797 return (out, err), proc
799 def exitcode(self, home, ecodefile = "exitcode"):
801 Get the exit code of an application.
802 Returns an integer value with the exit code
804 (out, err), proc = self.check_output(home, ecodefile)
806 # Succeeded to open file, return exit code in the file
809 return int(out.strip())
811 # Error in the content of the file!
812 return ExitCode.CORRUPTFILE
814 # No such file or directory
815 if proc.returncode == 1:
816 return ExitCode.FILENOTFOUND
818 # Other error from 'cat'
819 return ExitCode.ERROR
821 def upload_command(self, command,
823 ecodefile = "exitcode",
826 """ Saves the command as a bash script file in the remote host, and
827 forces to save the exit code of the command execution to the ecodefile
830 if not (command.strip().endswith(";") or command.strip().endswith("&")):
833 # The exit code of the command will be stored in ecodefile
834 command = " { %(command)s } ; echo $? > %(ecodefile)s ;" % {
836 'ecodefile': ecodefile,
840 environ = self.format_environment(env)
842 # Add environ to command
843 command = environ + command
845 return self.upload(command, shfile, text = True, overwrite = overwrite)
847 def format_environment(self, env, inline = False):
848 """ Formats the environment variables for a command to be executed
849 either as an inline command
850 (i.e. export PYTHONPATH=src/..; export LALAL= ..;python script.py) or
851 as a bash script (i.e. export PYTHONPATH=src/.. \n export LALA=.. \n)
853 if not env: return ""
855 # Remove extra white spaces
856 env = re.sub(r'\s+', ' ', env.strip())
858 sep = ";" if inline else "\n"
859 return sep.join(map(lambda e: " export %s" % e, env.split(" "))) + sep
861 def check_errors(self, home,
862 ecodefile = "exitcode",
864 """ Checks whether errors occurred while running a command.
865 It first checks the exit code for the command, and only if the
866 exit code is an error one it returns the error output.
872 # get exit code saved in the 'exitcode' file
873 ecode = self.exitcode(home, ecodefile)
875 if ecode in [ ExitCode.CORRUPTFILE, ExitCode.ERROR ]:
876 err = "Error retrieving exit code status from file %s/%s" % (home, ecodefile)
877 elif ecode > 0 or ecode == ExitCode.FILENOTFOUND:
878 # The process returned an error code or didn't exist.
879 # Check standard error.
880 (err, eerr), proc = self.check_output(home, stderr)
882 # If the stderr file was not found, assume nothing bad happened,
883 # and just ignore the error.
884 # (cat returns 1 for error "No such file or directory")
885 if ecode == ExitCode.FILENOTFOUND and proc.poll() == 1:
888 return ("", err), proc
890 def wait_pid(self, home, pidfile = "pidfile", raise_on_error = False):
891 """ Waits until the pid file for the command is generated,
892 and returns the pid and ppid of the process """
897 pidtuple = self.getpid(home = home, pidfile = pidfile)
906 msg = " Failed to get pid for pidfile %s/%s " % (
911 raise RuntimeError, msg
915 def wait_run(self, pid, ppid, trial = 0):
916 """ wait for a remote process to finish execution """
920 status = self.status(pid, ppid)
922 if status is ProcStatus.FINISHED:
924 elif status is not ProcStatus.RUNNING:
927 # If it takes more than 20 seconds to start, then
928 # asume something went wrong
932 # The app is running, just wait...
935 def check_output(self, home, filename):
936 """ Retrives content of file """
937 (out, err), proc = self.execute("cat %s" %
938 os.path.join(home, filename), retry = 1, with_lock = True)
939 return (out, err), proc
942 """ Checks if host is responsive
948 msg = "Unresponsive host. Wrong answer. "
950 # The underlying SSH layer will sometimes return an empty
951 # output (even if the command was executed without errors).
952 # To work arround this, repeat the operation N times or
953 # until the result is not empty string
955 (out, err), proc = self.execute("echo 'ALIVE'",
959 if out.find("ALIVE") > -1:
962 trace = traceback.format_exc()
963 msg = "Unresponsive host. Error reaching host: %s " % trace
965 self.error(msg, out, err)
969 """ Retrieves host home directory
971 # The underlying SSH layer will sometimes return an empty
972 # output (even if the command was executed without errors).
973 # To work arround this, repeat the operation N times or
974 # until the result is not empty string
975 msg = "Impossible to retrieve HOME directory"
977 (out, err), proc = self.execute("echo ${HOME}",
981 if out.strip() != "":
982 self._home_dir = out.strip()
984 trace = traceback.format_exc()
985 msg = "Impossible to retrieve HOME directory" % trace
987 if not self._home_dir:
988 self.error(msg, out, err)
989 raise RuntimeError, msg
991 def filter_existing_files(self, src, dst):
992 """ Removes files that already exist in the Linux host from src list
994 # construct a dictionary with { dst: src }
995 dests = dict(map(lambda x: ( os.path.join(dst, os.path.basename(x) ), x ),
996 src.strip().split(" ") ) ) if src.strip().find(" ") != -1 else dict({dst: src})
999 for d in dests.keys():
1000 command.append(" [ -f %(dst)s ] && echo '%(dst)s' " % {'dst' : d} )
1002 command = ";".join(command)
1004 (out, err), proc = self.execute(command, retry = 1, with_lock = True)
1006 for d in dests.keys():
1007 if out.find(d) > -1:
1013 return " ".join(dests.values())