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
21 from nepi.execution.resource import ResourceManager, clsinit, ResourceState, \
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",
171 flags = Flags.ExecReadOnly)
173 clean_experiment = Attribute("cleanExperiment", "Remove all files and directories "
174 " from a previous same experiment, before the new experiment starts",
175 flags = Flags.ExecReadOnly)
177 clean_processes = Attribute("cleanProcesses",
178 "Kill all running processes before starting experiment",
179 flags = Flags.ExecReadOnly)
181 tear_down = Attribute("tearDown", "Bash script to be executed before " + \
182 "releasing the resource",
183 flags = Flags.ExecReadOnly)
185 cls._register_attribute(hostname)
186 cls._register_attribute(username)
187 cls._register_attribute(port)
188 cls._register_attribute(home)
189 cls._register_attribute(identity)
190 cls._register_attribute(server_key)
191 cls._register_attribute(clean_home)
192 cls._register_attribute(clean_experiment)
193 cls._register_attribute(clean_processes)
194 cls._register_attribute(tear_down)
196 def __init__(self, ec, guid):
197 super(LinuxNode, self).__init__(ec, guid)
199 # home directory at Linux host
202 # lock to prevent concurrent applications on the same node,
203 # to execute commands at the same time. There are potential
204 # concurrency issues when using SSH to a same host from
205 # multiple threads. There are also possible operational
206 # issues, e.g. an application querying the existence
207 # of a file or folder prior to its creation, and another
208 # application creating the same file or folder in between.
209 self._node_lock = threading.Lock()
211 def log_message(self, msg):
212 return " guid %d - host %s - %s " % (self.guid,
213 self.get("hostname"), msg)
217 home = self.get("home") or ""
218 if not home.startswith("/"):
219 home = os.path.join(self._home_dir, home)
224 return os.path.join(self.home_dir, "nepi-usr")
228 return os.path.join(self.usr_dir, "lib")
232 return os.path.join(self.usr_dir, "bin")
236 return os.path.join(self.usr_dir, "src")
240 return os.path.join(self.usr_dir, "share")
244 return os.path.join(self.home_dir, "nepi-exp")
248 return os.path.join(self.exp_dir, self.ec.exp_id)
252 return os.path.join(self.exp_home, "node-%d" % self.guid)
256 return os.path.join(self.node_home, self.ec.run_id)
263 if (not self.get("hostname") or not self.get("username")):
264 msg = "Can't resolve OS, insufficient data "
266 raise RuntimeError, msg
270 if out.find("Fedora release 8") == 0:
271 self._os = OSType.FEDORA_8
272 elif out.find("Fedora release 12") == 0:
273 self._os = OSType.FEDORA_12
274 elif out.find("Fedora release 14") == 0:
275 self._os = OSType.FEDORA_14
276 elif out.find("Debian") == 0:
277 self._os = OSType.DEBIAN
278 elif out.find("Ubuntu") ==0:
279 self._os = OSType.UBUNTU
281 msg = "Unsupported OS"
283 raise RuntimeError, "%s - %s " %( msg, out )
288 # The underlying SSH layer will sometimes return an empty
289 # output (even if the command was executed without errors).
290 # To work arround this, repeat the operation N times or
291 # until the result is not empty string
296 (out, err), proc = self.execute("cat /etc/issue",
301 if out.strip() != "":
304 trace = traceback.format_exc()
305 msg = "Error detecting OS: %s " % trace
306 self.error(msg, out, err)
309 time.sleep(min(30.0, retrydelay))
315 return self.os in [OSType.DEBIAN, OSType.UBUNTU]
319 return self.os in [OSType.FEDORA_12, OSType.FEDORA_14, OSType.FEDORA_8,
324 return self.get("hostname") in ['localhost', '127.0.0.7', '::1']
327 # check if host is alive
328 if not self.is_alive():
330 msg = "Deploy failed. Unresponsive node %s" % self.get("hostname")
332 raise RuntimeError, msg
336 if self.get("cleanProcesses"):
337 self.clean_processes()
339 if self.get("cleanHome"):
342 if self.get("cleanExperiment"):
343 self.clean_experiment()
345 # Create shared directory structure
346 self.mkdir(self.lib_dir)
347 self.mkdir(self.bin_dir)
348 self.mkdir(self.src_dir)
349 self.mkdir(self.share_dir)
351 # Create experiment node home directory
352 self.mkdir(self.node_home)
354 super(LinuxNode, self).provision()
357 if self.state == ResourceState.NEW:
365 # Node needs to wait until all associated interfaces are
366 # ready before it can finalize deployment
367 from nepi.resources.linux.interface import LinuxInterface
368 ifaces = self.get_connected(LinuxInterface.rtype())
370 if iface.state < ResourceState.READY:
371 self.ec.schedule(reschedule_delay, self.deploy)
374 super(LinuxNode, self).deploy()
377 # Node needs to wait until all associated RMs are released
379 rms = self.get_connected()
381 if rm.state < ResourceState.STOPPED:
382 self.ec.schedule(reschedule_delay, self.release)
385 tear_down = self.get("tearDown")
387 self.execute(tear_down)
389 self.clean_processes()
391 super(LinuxNode, self).release()
393 def valid_connection(self, guid):
397 def clean_processes(self, killer = False):
398 self.info("Cleaning up processes")
402 cmd = ("sudo -S killall python tcpdump || /bin/true ; " +
403 "sudo -S killall python tcpdump || /bin/true ; " +
404 "sudo -S kill $(ps -N -T -o pid --no-heading | grep -v $PPID | sort) || /bin/true ; " +
405 "sudo -S killall -u root || /bin/true ; " +
406 "sudo -S killall -u root || /bin/true ; ")
409 cmd = ("sudo -S killall tcpdump || /bin/true ; " +
410 "sudo -S killall tcpdump || /bin/true ; " +
411 "sudo -S killall -u %s || /bin/true ; " % self.get("username") +
412 "sudo -S killall -u %s || /bin/true ; " % self.get("username"))
415 (out, err), proc = self.execute(cmd, retry = 1, with_lock = True)
417 def clean_home(self):
418 """ Cleans all NEPI related folders in the Linux host
420 self.info("Cleaning up home")
422 cmd = "cd %s ; find . -maxdepth 1 \( -name 'nepi-usr' -o -name 'nepi-exp' \) -execdir rm -rf {} + " % (
425 return self.execute(cmd, with_lock = True)
427 def clean_experiment(self):
428 """ Cleans all experiment related files in the Linux host.
429 It preserves NEPI files and folders that have a multi experiment
432 self.info("Cleaning up experiment files")
434 cmd = "cd %s ; find . -maxdepth 1 -name '%s' -execdir rm -rf {} + " % (
438 return self.execute(cmd, with_lock = True)
440 def execute(self, command,
448 err_on_timeout = True,
449 connect_timeout = 30,
450 strict_host_checking = False,
455 """ Notice that this invocation will block until the
456 execution finishes. If this is not the desired behavior,
457 use 'run' instead."""
460 (out, err), proc = execfuncs.lexec(command,
467 with self._node_lock:
468 (out, err), proc = sshfuncs.rexec(
470 host = self.get("hostname"),
471 user = self.get("username"),
472 port = self.get("port"),
476 identity = self.get("identity"),
477 server_key = self.get("serverKey"),
480 forward_x11 = forward_x11,
483 err_on_timeout = err_on_timeout,
484 connect_timeout = connect_timeout,
485 persistent = persistent,
487 strict_host_checking = strict_host_checking
490 (out, err), proc = sshfuncs.rexec(
492 host = self.get("hostname"),
493 user = self.get("username"),
494 port = self.get("port"),
498 identity = self.get("identity"),
499 server_key = self.get("serverKey"),
502 forward_x11 = forward_x11,
505 err_on_timeout = err_on_timeout,
506 connect_timeout = connect_timeout,
507 persistent = persistent,
509 strict_host_checking = strict_host_checking
512 return (out, err), proc
514 def run(self, command, home,
523 self.debug("Running command '%s'" % command)
526 (out, err), proc = execfuncs.lspawn(command, pidfile,
531 create_home = create_home,
535 with self._node_lock:
536 (out, err), proc = sshfuncs.rspawn(
540 create_home = create_home,
541 stdin = stdin if stdin is not None else '/dev/null',
542 stdout = stdout if stdout else '/dev/null',
543 stderr = stderr if stderr else '/dev/null',
545 host = self.get("hostname"),
546 user = self.get("username"),
547 port = self.get("port"),
549 identity = self.get("identity"),
550 server_key = self.get("serverKey"),
554 return (out, err), proc
556 def getpid(self, home, pidfile = "pidfile"):
558 pidtuple = execfuncs.lgetpid(os.path.join(home, pidfile))
560 with self._node_lock:
561 pidtuple = sshfuncs.rgetpid(
562 os.path.join(home, pidfile),
563 host = self.get("hostname"),
564 user = self.get("username"),
565 port = self.get("port"),
567 identity = self.get("identity"),
568 server_key = self.get("serverKey")
573 def status(self, pid, ppid):
575 status = execfuncs.lstatus(pid, ppid)
577 with self._node_lock:
578 status = sshfuncs.rstatus(
580 host = self.get("hostname"),
581 user = self.get("username"),
582 port = self.get("port"),
584 identity = self.get("identity"),
585 server_key = self.get("serverKey")
590 def kill(self, pid, ppid, sudo = False):
593 status = self.status(pid, ppid)
595 if status == sshfuncs.ProcStatus.RUNNING:
597 (out, err), proc = execfuncs.lkill(pid, ppid, sudo)
599 with self._node_lock:
600 (out, err), proc = sshfuncs.rkill(
602 host = self.get("hostname"),
603 user = self.get("username"),
604 port = self.get("port"),
607 identity = self.get("identity"),
608 server_key = self.get("serverKey")
611 return (out, err), proc
613 def copy(self, src, dst):
615 (out, err), proc = execfuncs.lcopy(source, dest,
617 strict_host_checking = False)
619 with self._node_lock:
620 (out, err), proc = sshfuncs.rcopy(
622 port = self.get("port"),
623 identity = self.get("identity"),
624 server_key = self.get("serverKey"),
626 strict_host_checking = False)
628 return (out, err), proc
631 def upload(self, src, dst, text = False, overwrite = True):
632 """ Copy content to destination
634 src content to copy. Can be a local file, directory or a list of files
636 dst destination path on the remote host (remote is always self.host)
638 text src is text input, it must be stored into a temp file before uploading
640 # If source is a string input
642 if text and not os.path.isfile(src):
643 # src is text input that should be uploaded as file
644 # create a temporal file with the content to upload
645 f = tempfile.NamedTemporaryFile(delete=False)
650 # If dst files should not be overwritten, check that the files do not
652 if overwrite == False:
653 src = self.filter_existing_files(src, dst)
655 return ("", ""), None
657 if not self.localhost:
658 # Build destination as <user>@<server>:<path>
659 dst = "%s@%s:%s" % (self.get("username"), self.get("hostname"), dst)
661 result = self.copy(src, dst)
669 def download(self, src, dst):
670 if not self.localhost:
671 # Build destination as <user>@<server>:<path>
672 src = "%s@%s:%s" % (self.get("username"), self.get("hostname"), src)
673 return self.copy(src, dst)
675 def install_packages_command(self, packages):
678 command = rpmfuncs.install_packages_command(self.os, packages)
680 command = debfuncs.install_packages_command(self.os, packages)
682 msg = "Error installing packages ( OS not known ) "
683 self.error(msg, self.os)
684 raise RuntimeError, msg
688 def install_packages(self, packages, home, run_home = None):
689 """ Install packages in the Linux host.
691 'home' is the directory to upload the package installation script.
692 'run_home' is the directory from where to execute the script.
694 command = self.install_packages_command(packages)
696 run_home = run_home or home
698 (out, err), proc = self.run_and_wait(command, run_home,
699 shfile = os.path.join(home, "instpkg.sh"),
700 pidfile = "instpkg_pidfile",
701 ecodefile = "instpkg_exitcode",
702 stdout = "instpkg_stdout",
703 stderr = "instpkg_stderr",
705 raise_on_error = True)
707 return (out, err), proc
709 def remove_packages(self, packages, home, run_home = None):
710 """ Uninstall packages from the Linux host.
712 'home' is the directory to upload the package un-installation script.
713 'run_home' is the directory from where to execute the script.
716 command = rpmfuncs.remove_packages_command(self.os, packages)
718 command = debfuncs.remove_packages_command(self.os, packages)
720 msg = "Error removing packages ( OS not known ) "
722 raise RuntimeError, msg
724 run_home = run_home or home
726 (out, err), proc = self.run_and_wait(command, run_home,
727 shfile = os.path.join(home, "rmpkg.sh"),
728 pidfile = "rmpkg_pidfile",
729 ecodefile = "rmpkg_exitcode",
730 stdout = "rmpkg_stdout",
731 stderr = "rmpkg_stderr",
733 raise_on_error = True)
735 return (out, err), proc
737 def mkdir(self, path, clean = False):
741 return self.execute("mkdir -p %s" % path, with_lock = True)
743 def rmdir(self, path):
744 return self.execute("rm -rf %s" % path, with_lock = True)
746 def run_and_wait(self, command, home,
751 ecodefile = "exitcode",
757 raise_on_error = False):
759 Uploads the 'command' to a bash script in the host.
760 Then runs the script detached in background in the host, and
761 busy-waites until the script finishes executing.
764 if not shfile.startswith("/"):
765 shfile = os.path.join(home, shfile)
767 self.upload_command(command,
769 ecodefile = ecodefile,
771 overwrite = overwrite)
773 command = "bash %s" % shfile
774 # run command in background in remote host
775 (out, err), proc = self.run(command, home,
783 # check no errors occurred
785 msg = " Failed to run command '%s' " % command
786 self.error(msg, out, err)
788 raise RuntimeError, msg
790 # Wait for pid file to be generated
791 pid, ppid = self.wait_pid(
794 raise_on_error = raise_on_error)
796 # wait until command finishes to execute
797 self.wait_run(pid, ppid)
799 (eout, err), proc = self.check_errors(home,
800 ecodefile = ecodefile,
803 # Out is what was written in the stderr file
805 msg = " Failed to run command '%s' " % command
806 self.error(msg, eout, err)
809 raise RuntimeError, msg
811 (out, oerr), proc = self.check_output(home, stdout)
813 return (out, err), proc
815 def exitcode(self, home, ecodefile = "exitcode"):
817 Get the exit code of an application.
818 Returns an integer value with the exit code
820 (out, err), proc = self.check_output(home, ecodefile)
822 # Succeeded to open file, return exit code in the file
825 return int(out.strip())
827 # Error in the content of the file!
828 return ExitCode.CORRUPTFILE
830 # No such file or directory
831 if proc.returncode == 1:
832 return ExitCode.FILENOTFOUND
834 # Other error from 'cat'
835 return ExitCode.ERROR
837 def upload_command(self, command,
839 ecodefile = "exitcode",
842 """ Saves the command as a bash script file in the remote host, and
843 forces to save the exit code of the command execution to the ecodefile
846 if not (command.strip().endswith(";") or command.strip().endswith("&")):
849 # The exit code of the command will be stored in ecodefile
850 command = " { %(command)s } ; echo $? > %(ecodefile)s ;" % {
852 'ecodefile': ecodefile,
856 environ = self.format_environment(env)
858 # Add environ to command
859 command = environ + command
861 return self.upload(command, shfile, text = True, overwrite = overwrite)
863 def format_environment(self, env, inline = False):
864 """ Formats the environment variables for a command to be executed
865 either as an inline command
866 (i.e. export PYTHONPATH=src/..; export LALAL= ..;python script.py) or
867 as a bash script (i.e. export PYTHONPATH=src/.. \n export LALA=.. \n)
869 if not env: return ""
871 # Remove extra white spaces
872 env = re.sub(r'\s+', ' ', env.strip())
874 sep = ";" if inline else "\n"
875 return sep.join(map(lambda e: " export %s" % e, env.split(" "))) + sep
877 def check_errors(self, home,
878 ecodefile = "exitcode",
880 """ Checks whether errors occurred while running a command.
881 It first checks the exit code for the command, and only if the
882 exit code is an error one it returns the error output.
888 # get exit code saved in the 'exitcode' file
889 ecode = self.exitcode(home, ecodefile)
891 if ecode in [ ExitCode.CORRUPTFILE, ExitCode.ERROR ]:
892 err = "Error retrieving exit code status from file %s/%s" % (home, ecodefile)
893 elif ecode > 0 or ecode == ExitCode.FILENOTFOUND:
894 # The process returned an error code or didn't exist.
895 # Check standard error.
896 (err, eerr), proc = self.check_output(home, stderr)
898 # If the stderr file was not found, assume nothing bad happened,
899 # and just ignore the error.
900 # (cat returns 1 for error "No such file or directory")
901 if ecode == ExitCode.FILENOTFOUND and proc.poll() == 1:
904 return ("", err), proc
906 def wait_pid(self, home, pidfile = "pidfile", raise_on_error = False):
907 """ Waits until the pid file for the command is generated,
908 and returns the pid and ppid of the process """
913 pidtuple = self.getpid(home = home, pidfile = pidfile)
922 msg = " Failed to get pid for pidfile %s/%s " % (
927 raise RuntimeError, msg
931 def wait_run(self, pid, ppid, trial = 0):
932 """ wait for a remote process to finish execution """
936 status = self.status(pid, ppid)
938 if status is ProcStatus.FINISHED:
940 elif status is not ProcStatus.RUNNING:
943 # If it takes more than 20 seconds to start, then
944 # asume something went wrong
948 # The app is running, just wait...
951 def check_output(self, home, filename):
952 """ Retrives content of file """
953 (out, err), proc = self.execute("cat %s" %
954 os.path.join(home, filename), retry = 1, with_lock = True)
955 return (out, err), proc
958 """ Checks if host is responsive
964 # The underlying SSH layer will sometimes return an empty
965 # output (even if the command was executed without errors).
966 # To work arround this, repeat the operation N times or
967 # until the result is not empty string
971 (out, err), proc = self.execute("echo 'ALIVE'",
976 if out.find("ALIVE") > -1:
979 trace = traceback.format_exc()
980 msg = "Unresponsive host. Error reaching host: %s " % trace
981 self.error(msg, out, err)
984 time.sleep(min(30.0, retrydelay))
987 if out.find("ALIVE") > -1:
990 msg = "Unresponsive host. Wrong answer. "
991 self.error(msg, out, err)
995 """ Retrieves host home directory
997 # The underlying SSH layer will sometimes return an empty
998 # output (even if the command was executed without errors).
999 # To work arround this, repeat the operation N times or
1000 # until the result is not empty string
1002 for i in xrange(10):
1004 (out, err), proc = self.execute("echo ${HOME}",
1009 if out.strip() != "":
1010 self._home_dir = out.strip()
1013 trace = traceback.format_exc()
1014 msg = "Impossible to retrieve HOME directory" % trace
1015 self.error(msg, out, err)
1018 time.sleep(min(30.0, retrydelay))
1021 if not self._home_dir:
1022 msg = "Impossible to retrieve HOME directory"
1023 self.error(msg, out, err)
1024 raise RuntimeError, msg
1026 def filter_existing_files(self, src, dst):
1027 """ Removes files that already exist in the Linux host from src list
1029 # construct a dictionary with { dst: src }
1030 dests = dict(map(lambda x: ( os.path.join(dst, os.path.basename(x) ), x ),
1031 src.strip().split(" ") ) ) if src.strip().find(" ") != -1 else dict({dst: src})
1034 for d in dests.keys():
1035 command.append(" [ -f %(dst)s ] && echo '%(dst)s' " % {'dst' : d} )
1037 command = ";".join(command)
1039 (out, err), proc = self.execute(command, retry = 1, with_lock = True)
1041 for d in dests.keys():
1042 if out.find(d) > -1:
1048 return " ".join(dests.values())