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.trace import Trace, TraceAttr
22 from nepi.execution.resource import ResourceManager, clsinit_copy, \
23 ResourceState, reschedule_delay
24 from nepi.resources.linux.node import LinuxNode
25 from nepi.util.sshfuncs import ProcStatus
26 from nepi.util.timefuncs import tnow, tdiffsec
31 # TODO: Resolve wildcards in commands!!
32 # TODO: When a failure occurs during deployment, scp and ssh processes are left running behind!!
35 class LinuxApplication(ResourceManager):
37 .. class:: Class Args :
39 :param ec: The Experiment controller
40 :type ec: ExperimentController
41 :param guid: guid of the RM
46 A LinuxApplication RM represents a process that can be executed in
47 a remote Linux host using SSH.
49 The LinuxApplication RM takes care of uploadin sources and any files
50 needed to run the experiment, to the remote host.
51 It also allows to provide source compilation (build) and installation
52 instructions, and takes care of automating the sources build and
53 installation tasks for the user.
55 It is important to note that files uploaded to the remote host have
56 two possible scopes: single-experiment or multi-experiment.
57 Single experiment files are those that will not be re-used by other
58 experiments. Multi-experiment files are those that will.
59 Sources and shared files are always made available to all experiments.
63 The directory structure used by LinuxApplication RM at the Linux
64 host is the following:
66 ${HOME}/.nepi/nepi-usr --> Base directory for multi-experiment files
68 ${LIB} |- /lib --> Base directory for libraries
69 ${BIN} |- /bin --> Base directory for binary files
70 ${SRC} |- /src --> Base directory for sources
71 ${SHARE} |- /share --> Base directory for other files
73 ${HOME}/.nepi/nepi-exp --> Base directory for single-experiment files
75 ${EXP_HOME} |- /<exp-id> --> Base directory for experiment exp-id
77 ${APP_HOME} |- /<app-guid> --> Base directory for application
78 | specific files (e.g. command.sh, input)
80 ${RUN_HOME} |- /<run-id> --> Base directory for run specific
84 _rtype = "LinuxApplication"
85 _help = "Runs an application on a Linux host with a BASH command "
86 _backend_type = "linux"
89 def _register_attributes(cls):
90 command = Attribute("command", "Command to execute at application start. "
91 "Note that commands will be executed in the ${RUN_HOME} directory, "
92 "make sure to take this into account when using relative paths. ",
94 forward_x11 = Attribute("forwardX11", "Enables X11 forwarding for SSH connections",
96 env = Attribute("env", "Environment variables string for command execution",
98 sudo = Attribute("sudo", "Run with root privileges",
100 depends = Attribute("depends",
101 "Space-separated list of packages required to run the application",
102 flags = Flags.Design)
103 sources = Attribute("sources",
104 "semi-colon separated list of regular files to be uploaded to ${SRC} "
105 "directory prior to building. Archives won't be expanded automatically. "
106 "Sources are globally available for all experiments unless "
107 "cleanHome is set to True (This will delete all sources). ",
108 flags = Flags.Design)
109 files = Attribute("files",
110 "Space-separated list of regular miscellaneous files to be uploaded "
111 "to ${SHARE} directory. "
112 "Files are globally available for all experiments unless "
113 "cleanHome is set to True (This will delete all files). ",
114 flags = Flags.Design)
115 libs = Attribute("libs",
116 "Space-separated list of libraries (e.g. .so files) to be uploaded "
117 "to ${LIB} directory. "
118 "Libraries are globally available for all experiments unless "
119 "cleanHome is set to True (This will delete all files). ",
120 flags = Flags.Design)
121 bins = Attribute("bins",
122 "Space-separated list of binary files to be uploaded "
123 "to ${BIN} directory. "
124 "Binaries are globally available for all experiments unless "
125 "cleanHome is set to True (This will delete all files). ",
126 flags = Flags.Design)
127 code = Attribute("code",
128 "Plain text source code to be uploaded to the ${APP_HOME} directory. ",
129 flags = Flags.Design)
130 build = Attribute("build",
131 "Build commands to execute after deploying the sources. "
132 "Sources are uploaded to the ${SRC} directory and code "
133 "is uploaded to the ${APP_HOME} directory. \n"
134 "Usage example: tar xzf ${SRC}/my-app.tgz && cd my-app && "
135 "./configure && make && make clean.\n"
136 "Make sure to make the build commands return with a nonzero exit "
138 flags = Flags.Design)
139 install = Attribute("install",
140 "Commands to transfer built files to their final destinations. "
141 "Install commands are executed after build commands. ",
142 flags = Flags.Design)
143 stdin = Attribute("stdin", "Standard input for the 'command'",
144 flags = Flags.Design)
145 tear_down = Attribute("tearDown", "Command to be executed just before "
146 "releasing the resource",
147 flags = Flags.Design)
149 cls._register_attribute(command)
150 cls._register_attribute(forward_x11)
151 cls._register_attribute(env)
152 cls._register_attribute(sudo)
153 cls._register_attribute(depends)
154 cls._register_attribute(sources)
155 cls._register_attribute(code)
156 cls._register_attribute(files)
157 cls._register_attribute(bins)
158 cls._register_attribute(libs)
159 cls._register_attribute(build)
160 cls._register_attribute(install)
161 cls._register_attribute(stdin)
162 cls._register_attribute(tear_down)
165 def _register_traces(cls):
166 stdout = Trace("stdout", "Standard output stream", enabled = True)
167 stderr = Trace("stderr", "Standard error stream", enabled = True)
169 cls._register_trace(stdout)
170 cls._register_trace(stderr)
172 def __init__(self, ec, guid):
173 super(LinuxApplication, self).__init__(ec, guid)
176 self._home = "app-%s" % self.guid
177 # whether the command should run in foreground attached
179 self._in_foreground = False
181 # whether to use sudo to kill the application process
182 self._sudo_kill = False
184 # keep a reference to the running process handler when
185 # the command is not executed as remote daemon in background
188 # timestamp of last state check of the application
189 self._last_state_check = tnow()
191 def log_message(self, msg):
192 return " guid %d - host %s - %s " % (self.guid,
193 self.node.get("hostname"), msg)
197 node = self.get_connected(LinuxNode.get_rtype())
198 if node: return node[0]
203 return os.path.join(self.node.exp_home, self._home)
207 return os.path.join(self.app_home, self.ec.run_id)
218 def in_foreground(self):
219 """ Returns True if the command needs to be executed in foreground.
220 This means that command will be executed using 'execute' instead of
221 'run' ('run' executes a command in background and detached from the
224 When using X11 forwarding option, the command can not run in background
225 and detached from a terminal, since we need to keep the terminal attached
228 return self.get("forwardX11") or self._in_foreground
230 def trace_filepath(self, filename):
231 return os.path.join(self.run_home, filename)
233 def trace(self, name, attr = TraceAttr.ALL, block = 512, offset = 0):
234 self.info("Retrieving '%s' trace %s " % (name, attr))
236 path = self.trace_filepath(name)
238 command = "(test -f %s && echo 'success') || echo 'error'" % path
239 (out, err), proc = self.node.execute(command)
241 if (err and proc.poll()) or out.find("error") != -1:
242 msg = " Couldn't find trace %s " % name
243 self.error(msg, out, err)
246 if attr == TraceAttr.PATH:
249 if attr == TraceAttr.ALL:
250 (out, err), proc = self.node.check_output(self.run_home, name)
253 msg = " Couldn't read trace %s " % name
254 self.error(msg, out, err)
259 if attr == TraceAttr.STREAM:
260 cmd = "dd if=%s bs=%d count=1 skip=%d" % (path, block, offset)
261 elif attr == TraceAttr.SIZE:
262 cmd = "stat -c%%s %s " % path
264 (out, err), proc = self.node.execute(cmd)
267 msg = " Couldn't find trace %s " % name
268 self.error(msg, out, err)
271 if attr == TraceAttr.SIZE:
272 out = int(out.strip())
276 def do_provision(self):
277 # take a snapshot of the system if user is root
278 # to ensure that cleanProcess will not kill
279 # pre-existent processes
280 if self.node.get("username") == 'root':
283 ps_aux = "ps aux |awk '{print $2,$11}'"
284 (out, err), proc = self.node.execute(ps_aux)
285 for line in out.strip().split("\n"):
286 parts = line.strip().split(" ")
287 procs[parts[0]] = parts[1]
288 pickle.dump(procs, open("/tmp/save.proc", "wb"))
290 # create run dir for application
291 self.node.mkdir(self.run_home)
293 # List of all the provision methods to invoke
300 self.upload_binaries,
302 self.upload_libraries,
307 # install dependencies
308 self.install_dependencies,
316 # Since provisioning takes a long time, before
317 # each step we check that the EC is still
320 self.debug("Interrupting provisioning. EC says 'ABORT")
327 # upload deploy script
328 deploy_command = ";".join(command)
329 self.execute_deploy_command(deploy_command)
331 # upload start script
332 self.upload_start_command()
334 self.info("Provisioning finished")
336 super(LinuxApplication, self).do_provision()
338 def upload_start_command(self, overwrite = False):
339 # Upload command to remote bash script
340 # - only if command can be executed in background and detached
341 command = self.get("command")
343 if command and not self.in_foreground:
344 self.info("Uploading command '%s'" % command)
346 # replace application specific paths in the command
347 command = self.replace_paths(command)
349 # replace application specific paths in the environment
350 env = self.get("env")
351 env = env and self.replace_paths(env)
353 shfile = os.path.join(self.app_home, "start.sh")
355 self.node.upload_command(command,
358 overwrite = overwrite)
360 def execute_deploy_command(self, command, prefix="deploy"):
362 # Upload the command to a bash script and run it
363 # in background ( but wait until the command has
364 # finished to continue )
365 shfile = os.path.join(self.app_home, "%s.sh" % prefix)
366 self.node.run_and_wait(command, self.run_home,
369 pidfile = "%s_pidfile" % prefix,
370 ecodefile = "%s_exitcode" % prefix,
371 stdout = "%s_stdout" % prefix,
372 stderr = "%s_stderr" % prefix)
374 def upload_sources(self, sources = None, src_dir = None):
376 sources = self.get("sources")
381 src_dir = self.node.src_dir
384 self.info("Uploading sources ")
386 sources = map(str.strip, sources.split(";"))
388 # Separate sources that should be downloaded from
389 # the web, from sources that should be uploaded from
392 for source in list(sources):
393 if source.startswith("http") or source.startswith("https"):
394 # remove the hhtp source from the sources list
395 sources.remove(source)
397 command.append( " ( "
398 # Check if the source already exists
399 " ls %(src_dir)s/%(basename)s "
401 # If source doesn't exist, download it and check
402 # that it it downloaded ok
403 " wget -c --directory-prefix=%(src_dir)s %(source)s && "
404 " ls %(src_dir)s/%(basename)s "
406 "basename": os.path.basename(source),
411 command = " && ".join(command)
413 # replace application specific paths in the command
414 command = self.replace_paths(command)
417 sources = ';'.join(sources)
418 self.node.upload(sources, src_dir, overwrite = False)
422 def upload_files(self, files = None):
424 files = self.get("files")
427 self.info("Uploading files %s " % files)
428 self.node.upload(files, self.node.share_dir, overwrite = False)
430 def upload_libraries(self, libs = None):
432 libs = self.get("libs")
435 self.info("Uploading libraries %s " % libaries)
436 self.node.upload(libs, self.node.lib_dir, overwrite = False)
438 def upload_binaries(self, bins = None):
440 bins = self.get("bins")
443 self.info("Uploading binaries %s " % binaries)
444 self.node.upload(bins, self.node.bin_dir, overwrite = False)
446 def upload_code(self, code = None):
448 code = self.get("code")
451 self.info("Uploading code")
453 dst = os.path.join(self.app_home, "code")
454 self.node.upload(code, dst, overwrite = False, text = True)
456 def upload_stdin(self, stdin = None):
458 stdin = self.get("stdin")
461 # create dir for sources
462 self.info("Uploading stdin")
464 # upload stdin file to ${SHARE_DIR} directory
465 if os.path.isfile(stdin):
466 basename = os.path.basename(stdin)
467 dst = os.path.join(self.node.share_dir, basename)
469 dst = os.path.join(self.app_home, "stdin")
471 self.node.upload(stdin, dst, overwrite = False, text = True)
473 # create "stdin" symlink on ${APP_HOME} directory
474 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
475 "app_home": self.app_home,
480 def install_dependencies(self, depends = None):
482 depends = self.get("depends")
485 self.info("Installing dependencies %s" % depends)
486 return self.node.install_packages_command(depends)
488 def build(self, build = None):
490 build = self.get("build")
493 self.info("Building sources ")
495 # replace application specific paths in the command
496 return self.replace_paths(build)
498 def install(self, install = None):
500 install = self.get("install")
503 self.info("Installing sources ")
505 # replace application specific paths in the command
506 return self.replace_paths(install)
509 # Wait until node is associated and deployed
511 if not node or node.state < ResourceState.READY:
512 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
513 self.ec.schedule(reschedule_delay, self.deploy)
515 command = self.get("command") or ""
516 self.info("Deploying command '%s' " % command)
520 super(LinuxApplication, self).do_deploy()
523 command = self.get("command")
525 self.info("Starting command '%s'" % command)
528 # If no command was given (i.e. Application was used for dependency
529 # installation), then the application is directly marked as STOPPED
530 super(LinuxApplication, self).set_stopped()
532 if self.in_foreground:
533 self._run_in_foreground()
535 self._run_in_background()
537 super(LinuxApplication, self).do_start()
539 def _run_in_foreground(self):
540 command = self.get("command")
541 sudo = self.get("sudo") or False
542 x11 = self.get("forwardX11")
543 env = self.get("env")
545 # Command will be launched in foreground and attached to the
546 # terminal using the node 'execute' in non blocking mode.
548 # We save the reference to the process in self._proc
549 # to be able to kill the process from the stop method.
550 # We also set blocking = False, since we don't want the
551 # thread to block until the execution finishes.
552 (out, err), self._proc = self.execute_command(command,
558 if self._proc.poll():
559 self.error(msg, out, err)
560 raise RuntimeError, msg
562 def _run_in_background(self):
563 command = self.get("command")
564 env = self.get("env")
565 sudo = self.get("sudo") or False
569 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
572 # Command will be run as a daemon in baground and detached from any
574 # The command to run was previously uploaded to a bash script
575 # during deployment, now we launch the remote script using 'run'
576 # method from the node.
577 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
578 (out, err), proc = self.node.run(cmd, self.run_home,
584 # check if execution errors occurred
585 msg = " Failed to start command '%s' " % command
588 self.error(msg, out, err)
589 raise RuntimeError, msg
591 # Wait for pid file to be generated
592 pid, ppid = self.node.wait_pid(self.run_home)
593 if pid: self._pid = int(pid)
594 if ppid: self._ppid = int(ppid)
596 # If the process is not running, check for error information
597 # on the remote machine
598 if not self.pid or not self.ppid:
599 (out, err), proc = self.node.check_errors(self.run_home,
602 # Out is what was written in the stderr file
604 msg = " Failed to start command '%s' " % command
605 self.error(msg, out, err)
606 raise RuntimeError, msg
609 """ Stops application execution
611 command = self.get('command') or ''
613 if self.state == ResourceState.STARTED:
615 self.info("Stopping command '%s' " % command)
617 # If the command is running in foreground (it was launched using
618 # the node 'execute' method), then we use the handler to the Popen
619 # process to kill it. Else we send a kill signal using the pid and ppid
620 # retrieved after running the command with the node 'run' method
624 # Only try to kill the process if the pid and ppid
626 if self.pid and self.ppid:
627 (out, err), proc = self.node.kill(self.pid, self.ppid,
628 sudo = self._sudo_kill)
630 # TODO: check if execution errors occurred
631 if (proc and proc.poll()) or err:
632 msg = " Failed to STOP command '%s' " % self.get("command")
633 self.error(msg, out, err)
635 super(LinuxApplication, self).do_stop()
637 def do_release(self):
638 self.info("Releasing resource")
640 tear_down = self.get("tearDown")
642 self.node.execute(tear_down)
646 super(LinuxApplication, self).do_release()
650 """ Returns the state of the application
652 if self._state == ResourceState.STARTED:
653 if self.in_foreground:
654 # Check if the process we used to execute the command
655 # is still running ...
656 retcode = self._proc.poll()
658 # retcode == None -> running
659 # retcode > 0 -> error
660 # retcode == 0 -> finished
663 msg = " Failed to execute command '%s'" % self.get("command")
664 err = self._proc.stderr.read()
665 self.error(msg, out, err)
671 # We need to query the status of the command we launched in
672 # background. In order to avoid overwhelming the remote host and
673 # the local processor with too many ssh queries, the state is only
674 # requested every 'state_check_delay' seconds.
675 state_check_delay = 0.5
676 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
677 if self.pid and self.ppid:
678 # Make sure the process is still running in background
679 status = self.node.status(self.pid, self.ppid)
681 if status == ProcStatus.FINISHED:
682 # If the program finished, check if execution
684 (out, err), proc = self.node.check_errors(
688 msg = "Failed to execute command '%s'" % \
690 self.error(msg, out, err)
695 self._last_state_check = tnow()
699 def execute_command(self, command,
707 environ = self.node.format_environment(env, inline = True)
708 command = environ + command
709 command = self.replace_paths(command)
711 return self.node.execute(command,
713 forward_x11 = forward_x11,
716 def replace_paths(self, command):
718 Replace all special path tags with shell-escaped actual paths.
721 .replace("${USR}", self.node.usr_dir)
722 .replace("${LIB}", self.node.lib_dir)
723 .replace("${BIN}", self.node.bin_dir)
724 .replace("${SRC}", self.node.src_dir)
725 .replace("${SHARE}", self.node.share_dir)
726 .replace("${EXP}", self.node.exp_dir)
727 .replace("${EXP_HOME}", self.node.exp_home)
728 .replace("${APP_HOME}", self.app_home)
729 .replace("${RUN_HOME}", self.run_home)
730 .replace("${NODE_HOME}", self.node.node_home)
731 .replace("${HOME}", self.node.home_dir)
734 def valid_connection(self, guid):