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-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-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. ",
93 flags = Flags.ExecReadOnly)
94 forward_x11 = Attribute("forwardX11", "Enables X11 forwarding for SSH connections",
95 flags = Flags.ExecReadOnly)
96 env = Attribute("env", "Environment variables string for command execution",
97 flags = Flags.ExecReadOnly)
98 sudo = Attribute("sudo", "Run with root privileges",
99 flags = Flags.ExecReadOnly)
100 depends = Attribute("depends",
101 "Space-separated list of packages required to run the application",
102 flags = Flags.ExecReadOnly)
103 sources = Attribute("sources",
104 "Space-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.ExecReadOnly)
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.ExecReadOnly)
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.ExecReadOnly)
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.ExecReadOnly)
127 code = Attribute("code",
128 "Plain text source code to be uploaded to the ${APP_HOME} directory. ",
129 flags = Flags.ExecReadOnly)
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.ReadOnly)
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.ReadOnly)
143 stdin = Attribute("stdin", "Standard input for the 'command'",
144 flags = Flags.ExecReadOnly)
145 tear_down = Attribute("tearDown", "Command to be executed just before "
146 "releasing the resource",
147 flags = Flags.ReadOnly)
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(self, name, attr = TraceAttr.ALL, block = 512, offset = 0):
231 self.info("Retrieving '%s' trace %s " % (name, attr))
233 path = os.path.join(self.run_home, name)
235 command = "(test -f %s && echo 'success') || echo 'error'" % path
236 (out, err), proc = self.node.execute(command)
238 if (err and proc.poll()) or out.find("error") != -1:
239 msg = " Couldn't find trace %s " % name
240 self.error(msg, out, err)
243 if attr == TraceAttr.PATH:
246 if attr == TraceAttr.ALL:
247 (out, err), proc = self.node.check_output(self.run_home, name)
250 msg = " Couldn't read trace %s " % name
251 self.error(msg, out, err)
256 if attr == TraceAttr.STREAM:
257 cmd = "dd if=%s bs=%d count=1 skip=%d" % (path, block, offset)
258 elif attr == TraceAttr.SIZE:
259 cmd = "stat -c%%s %s " % path
261 (out, err), proc = self.node.execute(cmd)
264 msg = " Couldn't find trace %s " % name
265 self.error(msg, out, err)
268 if attr == TraceAttr.SIZE:
269 out = int(out.strip())
273 def do_provision(self):
274 # take a snapshot of the system if user is root
275 # to assure cleanProcess kill every nepi process
276 if self.node.get("username") == 'root':
279 ps_aux = "ps aux |awk '{print $2,$11}'"
280 (out, err), proc = self.node.execute(ps_aux)
281 for line in out.strip().split("\n"):
282 parts = line.strip().split(" ")
283 procs[parts[0]] = parts[1]
284 pickle.dump(procs, open("/tmp/save.proc", "wb"))
286 # create run dir for application
287 self.node.mkdir(self.run_home)
289 # List of all the provision methods to invoke
296 self.upload_binaries,
298 self.upload_libraries,
303 # install dependencies
304 self.install_dependencies,
312 # Since provisioning takes a long time, before
313 # each step we check that the EC is still
316 self.debug("Interrupting provisioning. EC says 'ABORT")
323 # upload deploy script
324 deploy_command = ";".join(command)
325 self.execute_deploy_command(deploy_command)
327 # upload start script
328 self.upload_start_command()
330 self.info("Provisioning finished")
332 super(LinuxApplication, self).do_provision()
334 def upload_start_command(self, overwrite = False):
335 # Upload command to remote bash script
336 # - only if command can be executed in background and detached
337 command = self.get("command")
339 if command and not self.in_foreground:
340 self.info("Uploading command '%s'" % command)
342 # replace application specific paths in the command
343 command = self.replace_paths(command)
345 # replace application specific paths in the environment
346 env = self.get("env")
347 env = env and self.replace_paths(env)
349 shfile = os.path.join(self.app_home, "start.sh")
351 self.node.upload_command(command,
354 overwrite = overwrite)
356 def execute_deploy_command(self, command):
358 # Upload the command to a bash script and run it
359 # in background ( but wait until the command has
360 # finished to continue )
361 shfile = os.path.join(self.app_home, "deploy.sh")
362 self.node.run_and_wait(command, self.run_home,
365 pidfile = "deploy_pidfile",
366 ecodefile = "deploy_exitcode",
367 stdout = "deploy_stdout",
368 stderr = "deploy_stderr")
370 def upload_sources(self):
371 sources = self.get("sources")
376 self.info("Uploading sources ")
378 sources = sources.split(' ')
380 # Separate sources that should be downloaded from
381 # the web, from sources that should be uploaded from
384 for source in list(sources):
385 if source.startswith("http") or source.startswith("https"):
386 # remove the hhtp source from the sources list
387 sources.remove(source)
389 command.append( " ( "
390 # Check if the source already exists
391 " ls ${SRC}/%(basename)s "
393 # If source doesn't exist, download it and check
394 # that it it downloaded ok
395 " wget -c --directory-prefix=${SRC} %(source)s && "
396 " ls ${SRC}/%(basename)s "
398 "basename": os.path.basename(source),
402 command = " && ".join(command)
404 # replace application specific paths in the command
405 command = self.replace_paths(command)
408 sources = ' '.join(sources)
409 self.node.upload(sources, self.node.src_dir, overwrite = False)
413 def upload_files(self):
414 files = self.get("files")
417 self.info("Uploading files %s " % files)
418 self.node.upload(files, self.node.share_dir, overwrite = False)
420 def upload_libraries(self):
421 libs = self.get("libs")
424 self.info("Uploading libraries %s " % libaries)
425 self.node.upload(libs, self.node.lib_dir, overwrite = False)
427 def upload_binaries(self):
428 bins = self.get("bins")
431 self.info("Uploading binaries %s " % binaries)
432 self.node.upload(bins, self.node.bin_dir, overwrite = False)
434 def upload_code(self):
435 code = self.get("code")
438 self.info("Uploading code")
440 dst = os.path.join(self.app_home, "code")
441 self.node.upload(code, dst, overwrite = False, text = True)
443 def upload_stdin(self):
444 stdin = self.get("stdin")
446 # create dir for sources
447 self.info("Uploading stdin")
449 # upload stdin file to ${SHARE_DIR} directory
450 basename = os.path.basename(stdin)
451 dst = os.path.join(self.node.share_dir, basename)
452 self.node.upload(stdin, dst, overwrite = False, text = True)
454 # create "stdin" symlink on ${APP_HOME} directory
455 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
456 "app_home": self.app_home,
461 def install_dependencies(self):
462 depends = self.get("depends")
464 self.info("Installing dependencies %s" % depends)
465 return self.node.install_packages_command(depends)
468 build = self.get("build")
471 self.info("Building sources ")
473 # replace application specific paths in the command
474 return self.replace_paths(build)
477 install = self.get("install")
480 self.info("Installing sources ")
482 # replace application specific paths in the command
483 return self.replace_paths(install)
486 # Wait until node is associated and deployed
488 if not node or node.state < ResourceState.READY:
489 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
490 self.ec.schedule(reschedule_delay, self.deploy)
492 command = self.get("command") or ""
493 self.info("Deploying command '%s' " % command)
497 super(LinuxApplication, self).do_deploy()
500 command = self.get("command")
502 self.info("Starting command '%s'" % command)
505 # If no command was given (i.e. Application was used for dependency
506 # installation), then the application is directly marked as STOPPED
507 super(LinuxApplication, self).set_stopped()
509 if self.in_foreground:
510 self._run_in_foreground()
512 self._run_in_background()
514 super(LinuxApplication, self).do_start()
516 def _run_in_foreground(self):
517 command = self.get("command")
518 sudo = self.get("sudo") or False
519 x11 = self.get("forwardX11")
520 env = self.get("env")
522 # For a command being executed in foreground, if there is stdin,
523 # it is expected to be text string not a file or pipe
524 stdin = self.get("stdin") or None
526 # Command will be launched in foreground and attached to the
527 # terminal using the node 'execute' in non blocking mode.
529 # We save the reference to the process in self._proc
530 # to be able to kill the process from the stop method.
531 # We also set blocking = False, since we don't want the
532 # thread to block until the execution finishes.
533 (out, err), self._proc = self.execute_command(command,
540 if self._proc.poll():
541 self.error(msg, out, err)
542 raise RuntimeError, msg
544 def _run_in_background(self):
545 command = self.get("command")
546 env = self.get("env")
547 sudo = self.get("sudo") or False
551 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
554 # Command will be run as a daemon in baground and detached from any
556 # The command to run was previously uploaded to a bash script
557 # during deployment, now we launch the remote script using 'run'
558 # method from the node.
559 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
560 (out, err), proc = self.node.run(cmd, self.run_home,
566 # check if execution errors occurred
567 msg = " Failed to start command '%s' " % command
570 self.error(msg, out, err)
571 raise RuntimeError, msg
573 # Wait for pid file to be generated
574 pid, ppid = self.node.wait_pid(self.run_home)
575 if pid: self._pid = int(pid)
576 if ppid: self._ppid = int(ppid)
578 # If the process is not running, check for error information
579 # on the remote machine
580 if not self.pid or not self.ppid:
581 (out, err), proc = self.node.check_errors(self.run_home,
584 # Out is what was written in the stderr file
586 msg = " Failed to start command '%s' " % command
587 self.error(msg, out, err)
588 raise RuntimeError, msg
591 """ Stops application execution
593 command = self.get('command') or ''
595 if self.state == ResourceState.STARTED:
597 self.info("Stopping command '%s' " % command)
599 # If the command is running in foreground (it was launched using
600 # the node 'execute' method), then we use the handler to the Popen
601 # process to kill it. Else we send a kill signal using the pid and ppid
602 # retrieved after running the command with the node 'run' method
606 # Only try to kill the process if the pid and ppid
608 if self.pid and self.ppid:
609 (out, err), proc = self.node.kill(self.pid, self.ppid,
610 sudo = self._sudo_kill)
612 # TODO: check if execution errors occurred
613 if proc.poll() or err:
614 msg = " Failed to STOP command '%s' " % self.get("command")
615 self.error(msg, out, err)
617 super(LinuxApplication, self).do_stop()
619 def do_release(self):
620 self.info("Releasing resource")
622 tear_down = self.get("tearDown")
624 self.node.execute(tear_down)
628 super(LinuxApplication, self).do_release()
632 """ Returns the state of the application
634 if self._state == ResourceState.STARTED:
635 if self.in_foreground:
636 # Check if the process we used to execute the command
637 # is still running ...
638 retcode = self._proc.poll()
640 # retcode == None -> running
641 # retcode > 0 -> error
642 # retcode == 0 -> finished
645 msg = " Failed to execute command '%s'" % self.get("command")
646 err = self._proc.stderr.read()
647 self.error(msg, out, err)
653 # We need to query the status of the command we launched in
654 # background. In order to avoid overwhelming the remote host and
655 # the local processor with too many ssh queries, the state is only
656 # requested every 'state_check_delay' seconds.
657 state_check_delay = 0.5
658 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
659 if self.pid and self.ppid:
660 # Make sure the process is still running in background
661 status = self.node.status(self.pid, self.ppid)
663 if status == ProcStatus.FINISHED:
664 # If the program finished, check if execution
666 (out, err), proc = self.node.check_errors(
670 msg = "Failed to execute command '%s'" % \
672 self.error(msg, out, err)
677 self._last_state_check = tnow()
681 def execute_command(self, command,
690 environ = self.node.format_environment(env, inline = True)
691 command = environ + command
692 command = self.replace_paths(command)
694 return self.node.execute(command,
697 forward_x11 = forward_x11,
700 def replace_paths(self, command):
702 Replace all special path tags with shell-escaped actual paths.
705 .replace("${USR}", self.node.usr_dir)
706 .replace("${LIB}", self.node.lib_dir)
707 .replace("${BIN}", self.node.bin_dir)
708 .replace("${SRC}", self.node.src_dir)
709 .replace("${SHARE}", self.node.share_dir)
710 .replace("${EXP}", self.node.exp_dir)
711 .replace("${EXP_HOME}", self.node.exp_home)
712 .replace("${APP_HOME}", self.app_home)
713 .replace("${RUN_HOME}", self.run_home)
714 .replace("${NODE_HOME}", self.node.node_home)
715 .replace("${HOME}", self.node.home_dir)
718 def valid_connection(self, guid):