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)
147 def _register_attributes(cls):
148 hostname = Attribute("hostname", "Hostname of the machine",
149 flags = Flags.ExecReadOnly)
151 username = Attribute("username", "Local account username",
152 flags = Flags.Credential)
154 port = Attribute("port", "SSH port", flags = Flags.ExecReadOnly)
156 home = Attribute("home",
157 "Experiment home directory to store all experiment related files",
158 flags = Flags.ExecReadOnly)
160 identity = Attribute("identity", "SSH identity file",
161 flags = Flags.Credential)
163 server_key = Attribute("serverKey", "Server public key",
164 flags = Flags.ExecReadOnly)
166 clean_home = Attribute("cleanHome", "Remove all nepi files and directories "
167 " from node home folder before starting experiment",
168 flags = Flags.ExecReadOnly)
170 clean_experiment = Attribute("cleanExperiment", "Remove all files and directories "
171 " from a previous same experiment, before the new experiment starts",
172 flags = Flags.ExecReadOnly)
174 clean_processes = Attribute("cleanProcesses",
175 "Kill all running processes before starting experiment",
176 flags = Flags.ExecReadOnly)
178 tear_down = Attribute("tearDown", "Bash script to be executed before " + \
179 "releasing the resource",
180 flags = Flags.ExecReadOnly)
182 cls._register_attribute(hostname)
183 cls._register_attribute(username)
184 cls._register_attribute(port)
185 cls._register_attribute(home)
186 cls._register_attribute(identity)
187 cls._register_attribute(server_key)
188 cls._register_attribute(clean_home)
189 cls._register_attribute(clean_experiment)
190 cls._register_attribute(clean_processes)
191 cls._register_attribute(tear_down)
193 def __init__(self, ec, guid):
194 super(LinuxNode, self).__init__(ec, guid)
196 # home directory at Linux host
199 # lock to avoid concurrency issues on methods used by applications
200 self._lock = threading.Lock()
202 def log_message(self, msg):
203 return " guid %d - host %s - %s " % (self.guid,
204 self.get("hostname"), msg)
208 home = self.get("home") or ""
209 if not home.startswith("/"):
210 home = os.path.join(self._home_dir, home)
215 return os.path.join(self.home_dir, "nepi-usr")
219 return os.path.join(self.usr_dir, "lib")
223 return os.path.join(self.usr_dir, "bin")
227 return os.path.join(self.usr_dir, "src")
231 return os.path.join(self.usr_dir, "share")
235 return os.path.join(self.home_dir, "nepi-exp")
239 return os.path.join(self.exp_dir, self.ec.exp_id)
243 return os.path.join(self.exp_home, "node-%d" % self.guid)
247 return os.path.join(self.node_home, self.ec.run_id)
254 if (not self.get("hostname") or not self.get("username")):
255 msg = "Can't resolve OS, insufficient data "
257 raise RuntimeError, msg
259 (out, err), proc = self.execute("cat /etc/issue", with_lock = True)
261 if err and proc.poll():
262 msg = "Error detecting OS "
263 self.error(msg, out, err)
264 raise RuntimeError, "%s - %s - %s" %( msg, out, err )
266 if out.find("Fedora release 8") == 0:
267 self._os = OSType.FEDORA_8
268 elif out.find("Fedora release 12") == 0:
269 self._os = OSType.FEDORA_12
270 elif out.find("Fedora release 14") == 0:
271 self._os = OSType.FEDORA_14
272 elif out.find("Debian") == 0:
273 self._os = OSType.DEBIAN
274 elif out.find("Ubuntu") ==0:
275 self._os = OSType.UBUNTU
277 msg = "Unsupported OS"
279 raise RuntimeError, "%s - %s " %( msg, out )
285 return self.os in [OSType.DEBIAN, OSType.UBUNTU]
289 return self.os in [OSType.FEDORA_12, OSType.FEDORA_14, OSType.FEDORA_8,
294 return self.get("hostname") in ['localhost', '127.0.0.7', '::1']
297 # check if host is alive
298 if not self.is_alive():
301 msg = "Deploy failed. Unresponsive node %s" % self.get("hostname")
303 raise RuntimeError, msg
307 if self.get("cleanProcesses"):
308 self.clean_processes()
310 if self.get("cleanHome"):
313 if self.get("cleanExperiment"):
314 self.clean_experiment()
316 # Create shared directory structure
317 self.mkdir(self.lib_dir)
318 self.mkdir(self.bin_dir)
319 self.mkdir(self.src_dir)
320 self.mkdir(self.share_dir)
322 # Create experiment node home directory
323 self.mkdir(self.node_home)
325 super(LinuxNode, self).provision()
328 if self.state == ResourceState.NEW:
333 self._state = ResourceState.FAILED
336 # Node needs to wait until all associated interfaces are
337 # ready before it can finalize deployment
338 from nepi.resources.linux.interface import LinuxInterface
339 ifaces = self.get_connected(LinuxInterface.rtype())
341 if iface.state < ResourceState.READY:
342 self.ec.schedule(reschedule_delay, self.deploy)
345 super(LinuxNode, self).deploy()
348 tear_down = self.get("tearDown")
350 self.execute(tear_down)
352 self.clean_processes()
354 super(LinuxNode, self).release()
356 def valid_connection(self, guid):
360 def clean_processes(self, killer = False):
361 self.info("Cleaning up processes")
365 cmd = ("sudo -S killall python tcpdump || /bin/true ; " +
366 "sudo -S killall python tcpdump || /bin/true ; " +
367 "sudo -S kill $(ps -N -T -o pid --no-heading | grep -v $PPID | sort) || /bin/true ; " +
368 "sudo -S killall -u root || /bin/true ; " +
369 "sudo -S killall -u root || /bin/true ; ")
372 cmd = ("sudo -S killall tcpdump || /bin/true ; " +
373 "sudo -S killall tcpdump || /bin/true ; " +
374 "sudo -S killall -u %s || /bin/true ; " % self.get("username") +
375 "sudo -S killall -u %s || /bin/true ; " % self.get("username"))
378 (out, err), proc = self.execute(cmd, retry = 1, with_lock = True)
380 def clean_home(self):
381 """ Cleans all NEPI related folders in the Linux host
383 self.info("Cleaning up home")
385 cmd = "cd %s ; find . -maxdepth 1 \( -name 'nepi-usr' -o -name 'nepi-exp' \) -execdir rm -rf {} + " % (
388 return self.execute(cmd, with_lock = True)
390 def clean_experiment(self):
391 """ Cleans all experiment related files in the Linux host.
392 It preserves NEPI files and folders that have a multi experiment
395 self.info("Cleaning up experiment files")
397 cmd = "cd %s ; find . -maxdepth 1 -name '%s' -execdir rm -rf {} + " % (
401 return self.execute(cmd, with_lock = True)
403 def execute(self, command,
411 err_on_timeout = True,
412 connect_timeout = 30,
413 strict_host_checking = False,
418 """ Notice that this invocation will block until the
419 execution finishes. If this is not the desired behavior,
420 use 'run' instead."""
423 (out, err), proc = execfuncs.lexec(command,
431 (out, err), proc = sshfuncs.rexec(
433 host = self.get("hostname"),
434 user = self.get("username"),
435 port = self.get("port"),
439 identity = self.get("identity"),
440 server_key = self.get("serverKey"),
443 forward_x11 = forward_x11,
446 err_on_timeout = err_on_timeout,
447 connect_timeout = connect_timeout,
448 persistent = persistent,
450 strict_host_checking = strict_host_checking
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 return (out, err), proc
477 def run(self, command, home,
486 self.debug("Running command '%s'" % command)
489 (out, err), proc = execfuncs.lspawn(command, pidfile,
494 create_home = create_home,
499 (out, err), proc = sshfuncs.rspawn(
503 create_home = create_home,
504 stdin = stdin if stdin is not None else '/dev/null',
505 stdout = stdout if stdout else '/dev/null',
506 stderr = stderr if stderr else '/dev/null',
508 host = self.get("hostname"),
509 user = self.get("username"),
510 port = self.get("port"),
512 identity = self.get("identity"),
513 server_key = self.get("serverKey"),
517 return (out, err), proc
519 def getpid(self, home, pidfile = "pidfile"):
521 pidtuple = execfuncs.lgetpid(os.path.join(home, pidfile))
524 pidtuple = sshfuncs.rgetpid(
525 os.path.join(home, pidfile),
526 host = self.get("hostname"),
527 user = self.get("username"),
528 port = self.get("port"),
530 identity = self.get("identity"),
531 server_key = self.get("serverKey")
536 def status(self, pid, ppid):
538 status = execfuncs.lstatus(pid, ppid)
541 status = sshfuncs.rstatus(
543 host = self.get("hostname"),
544 user = self.get("username"),
545 port = self.get("port"),
547 identity = self.get("identity"),
548 server_key = self.get("serverKey")
553 def kill(self, pid, ppid, sudo = False):
556 status = self.status(pid, ppid)
558 if status == sshfuncs.ProcStatus.RUNNING:
560 (out, err), proc = execfuncs.lkill(pid, ppid, sudo)
563 (out, err), proc = sshfuncs.rkill(
565 host = self.get("hostname"),
566 user = self.get("username"),
567 port = self.get("port"),
570 identity = self.get("identity"),
571 server_key = self.get("serverKey")
574 return (out, err), proc
576 def copy(self, src, dst):
578 (out, err), proc = execfuncs.lcopy(source, dest,
580 strict_host_checking = False)
583 (out, err), proc = sshfuncs.rcopy(
585 port = self.get("port"),
586 identity = self.get("identity"),
587 server_key = self.get("serverKey"),
589 strict_host_checking = False)
591 return (out, err), proc
594 def upload(self, src, dst, text = False, overwrite = True):
595 """ Copy content to destination
597 src content to copy. Can be a local file, directory or a list of files
599 dst destination path on the remote host (remote is always self.host)
601 text src is text input, it must be stored into a temp file before uploading
603 # If source is a string input
605 if text and not os.path.isfile(src):
606 # src is text input that should be uploaded as file
607 # create a temporal file with the content to upload
608 f = tempfile.NamedTemporaryFile(delete=False)
613 # If dst files should not be overwritten, check that the files do not
615 if overwrite == False:
616 src = self.filter_existing_files(src, dst)
618 return ("", ""), None
620 if not self.localhost:
621 # Build destination as <user>@<server>:<path>
622 dst = "%s@%s:%s" % (self.get("username"), self.get("hostname"), dst)
624 result = self.copy(src, dst)
632 def download(self, src, dst):
633 if not self.localhost:
634 # Build destination as <user>@<server>:<path>
635 src = "%s@%s:%s" % (self.get("username"), self.get("hostname"), src)
636 return self.copy(src, dst)
638 def install_packages(self, packages, home, run_home = None):
639 """ Install packages in the Linux host.
641 'home' is the directory to upload the package installation script.
642 'run_home' is the directory from where to execute the script.
646 command = rpmfuncs.install_packages_command(self.os, packages)
648 command = debfuncs.install_packages_command(self.os, packages)
650 msg = "Error installing packages ( OS not known ) "
651 self.error(msg, self.os)
652 raise RuntimeError, msg
654 run_home = run_home or home
656 (out, err), proc = self.run_and_wait(command, run_home,
657 shfile = os.path.join(home, "instpkg.sh"),
658 pidfile = "instpkg_pidfile",
659 ecodefile = "instpkg_exitcode",
660 stdout = "instpkg_stdout",
661 stderr = "instpkg_stderr",
663 raise_on_error = True)
665 return (out, err), proc
667 def remove_packages(self, packages, home, run_home = None):
668 """ Uninstall packages from the Linux host.
670 'home' is the directory to upload the package un-installation script.
671 'run_home' is the directory from where to execute the script.
674 command = rpmfuncs.remove_packages_command(self.os, packages)
676 command = debfuncs.remove_packages_command(self.os, packages)
678 msg = "Error removing packages ( OS not known ) "
680 raise RuntimeError, msg
682 run_home = run_home or home
684 (out, err), proc = self.run_and_wait(command, run_home,
685 shfile = os.path.join(home, "rmpkg.sh"),
686 pidfile = "rmpkg_pidfile",
687 ecodefile = "rmpkg_exitcode",
688 stdout = "rmpkg_stdout",
689 stderr = "rmpkg_stderr",
691 raise_on_error = True)
693 return (out, err), proc
695 def mkdir(self, path, clean = False):
699 return self.execute("mkdir -p %s" % path, with_lock = True)
701 def rmdir(self, path):
702 return self.execute("rm -rf %s" % path, with_lock = True)
704 def run_and_wait(self, command, home,
709 ecodefile = "exitcode",
715 raise_on_error = False):
717 Uploads the 'command' to a bash script in the host.
718 Then runs the script detached in background in the host, and
719 busy-waites until the script finishes executing.
722 if not shfile.startswith("/"):
723 shfile = os.path.join(home, shfile)
725 self.upload_command(command,
727 ecodefile = ecodefile,
729 overwrite = overwrite)
731 command = "bash %s" % shfile
732 # run command in background in remote host
733 (out, err), proc = self.run(command, home,
741 # check no errors occurred
743 msg = " Failed to run command '%s' " % command
744 self.error(msg, out, err)
746 raise RuntimeError, msg
748 # Wait for pid file to be generated
749 pid, ppid = self.wait_pid(
752 raise_on_error = raise_on_error)
754 # wait until command finishes to execute
755 self.wait_run(pid, ppid)
757 (eout, err), proc = self.check_errors(home,
758 ecodefile = ecodefile,
761 # Out is what was written in the stderr file
763 msg = " Failed to run command '%s' " % command
764 self.error(msg, eout, err)
767 raise RuntimeError, msg
769 (out, oerr), proc = self.check_output(home, stdout)
771 return (out, err), proc
773 def exitcode(self, home, ecodefile = "exitcode"):
775 Get the exit code of an application.
776 Returns an integer value with the exit code
778 (out, err), proc = self.check_output(home, ecodefile)
780 # Succeeded to open file, return exit code in the file
783 return int(out.strip())
785 # Error in the content of the file!
786 return ExitCode.CORRUPTFILE
788 # No such file or directory
789 if proc.returncode == 1:
790 return ExitCode.FILENOTFOUND
792 # Other error from 'cat'
793 return ExitCode.ERROR
795 def upload_command(self, command,
797 ecodefile = "exitcode",
800 """ Saves the command as a bash script file in the remote host, and
801 forces to save the exit code of the command execution to the ecodefile
804 if not (command.strip().endswith(";") or command.strip().endswith("&")):
807 # The exit code of the command will be stored in ecodefile
808 command = " { %(command)s } ; echo $? > %(ecodefile)s ;" % {
810 'ecodefile': ecodefile,
814 environ = self.format_environment(env)
816 # Add environ to command
817 command = environ + command
819 return self.upload(command, shfile, text = True, overwrite = overwrite)
821 def format_environment(self, env, inline = False):
822 """ Formats the environment variables for a command to be executed
823 either as an inline command
824 (i.e. export PYTHONPATH=src/..; export LALAL= ..;python script.py) or
825 as a bash script (i.e. export PYTHONPATH=src/.. \n export LALA=.. \n)
827 if not env: return ""
829 # Remove extra white spaces
830 env = re.sub(r'\s+', ' ', env.strip())
832 sep = ";" if inline else "\n"
833 return sep.join(map(lambda e: " export %s" % e, env.split(" "))) + sep
835 def check_errors(self, home,
836 ecodefile = "exitcode",
838 """ Checks whether errors occurred while running a command.
839 It first checks the exit code for the command, and only if the
840 exit code is an error one it returns the error output.
846 # get exit code saved in the 'exitcode' file
847 ecode = self.exitcode(home, ecodefile)
849 if ecode in [ ExitCode.CORRUPTFILE, ExitCode.ERROR ]:
850 err = "Error retrieving exit code status from file %s/%s" % (home, ecodefile)
851 elif ecode > 0 or ecode == ExitCode.FILENOTFOUND:
852 # The process returned an error code or didn't exist.
853 # Check standard error.
854 (err, eerr), proc = self.check_output(home, stderr)
856 # If the stderr file was not found, assume nothing bad happened,
857 # and just ignore the error.
858 # (cat returns 1 for error "No such file or directory")
859 if ecode == ExitCode.FILENOTFOUND and proc.poll() == 1:
862 return ("", err), proc
864 def wait_pid(self, home, pidfile = "pidfile", raise_on_error = False):
865 """ Waits until the pid file for the command is generated,
866 and returns the pid and ppid of the process """
871 pidtuple = self.getpid(home = home, pidfile = pidfile)
880 msg = " Failed to get pid for pidfile %s/%s " % (
885 raise RuntimeError, msg
889 def wait_run(self, pid, ppid, trial = 0):
890 """ wait for a remote process to finish execution """
894 status = self.status(pid, ppid)
896 if status is ProcStatus.FINISHED:
898 elif status is not ProcStatus.RUNNING:
901 # If it takes more than 20 seconds to start, then
902 # asume something went wrong
906 # The app is running, just wait...
909 def check_output(self, home, filename):
910 """ Retrives content of file """
911 (out, err), proc = self.execute("cat %s" %
912 os.path.join(home, filename), retry = 1, with_lock = True)
913 return (out, err), proc
916 """ Checks if host is responsive
923 (out, err), proc = self.execute("echo 'ALIVE'",
927 trace = traceback.format_exc()
928 msg = "Unresponsive host %s " % err
929 self.error(msg, out, trace)
932 if out.strip() == "ALIVE":
935 msg = "Unresponsive host "
936 self.error(msg, out, err)
940 """ Retrieves host home directory
942 (out, err), proc = self.execute("echo ${HOME}", retry = 5,
946 msg = "Imposible to retrieve HOME directory"
947 self.error(msg, out, err)
948 raise RuntimeError, msg
950 self._home_dir = out.strip()
952 def filter_existing_files(self, src, dst):
953 """ Removes files that already exist in the Linux host from src list
955 # construct a dictionary with { dst: src }
956 dests = dict(map(lambda x: ( os.path.join(dst, os.path.basename(x) ), x ),
957 src.strip().split(" ") ) ) if src.strip().find(" ") != -1 else dict({dst: src})
960 for d in dests.keys():
961 command.append(" [ -f %(dst)s ] && echo '%(dst)s' " % {'dst' : d} )
963 command = ";".join(command)
965 (out, err), proc = self.execute(command, retry = 1, with_lock = True)
967 for d in dests.keys():
974 return " ".join(dests.values())