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 assure cleanProcess kill every nepi process
279 if self.node.get("username") == 'root':
282 ps_aux = "ps aux |awk '{print $2,$11}'"
283 (out, err), proc = self.node.execute(ps_aux)
284 for line in out.strip().split("\n"):
285 parts = line.strip().split(" ")
286 procs[parts[0]] = parts[1]
287 pickle.dump(procs, open("/tmp/save.proc", "wb"))
289 # create run dir for application
290 self.node.mkdir(self.run_home)
292 # List of all the provision methods to invoke
299 self.upload_binaries,
301 self.upload_libraries,
306 # install dependencies
307 self.install_dependencies,
315 # Since provisioning takes a long time, before
316 # each step we check that the EC is still
319 self.debug("Interrupting provisioning. EC says 'ABORT")
326 # upload deploy script
327 deploy_command = ";".join(command)
328 self.execute_deploy_command(deploy_command)
330 # upload start script
331 self.upload_start_command()
333 self.info("Provisioning finished")
335 super(LinuxApplication, self).do_provision()
337 def upload_start_command(self, overwrite = False):
338 # Upload command to remote bash script
339 # - only if command can be executed in background and detached
340 command = self.get("command")
342 if command and not self.in_foreground:
343 self.info("Uploading command '%s'" % command)
345 # replace application specific paths in the command
346 command = self.replace_paths(command)
348 # replace application specific paths in the environment
349 env = self.get("env")
350 env = env and self.replace_paths(env)
352 shfile = os.path.join(self.app_home, "start.sh")
354 self.node.upload_command(command,
357 overwrite = overwrite)
359 def execute_deploy_command(self, command):
361 # Upload the command to a bash script and run it
362 # in background ( but wait until the command has
363 # finished to continue )
364 shfile = os.path.join(self.app_home, "deploy.sh")
365 self.node.run_and_wait(command, self.run_home,
368 pidfile = "deploy_pidfile",
369 ecodefile = "deploy_exitcode",
370 stdout = "deploy_stdout",
371 stderr = "deploy_stderr")
373 def upload_sources(self, src_dir = None):
374 sources = self.get("sources")
379 src_dir = self.node.src_dir
382 self.info("Uploading sources ")
384 sources = map(str.strip, sources.split(";"))
386 # Separate sources that should be downloaded from
387 # the web, from sources that should be uploaded from
390 for source in list(sources):
391 if source.startswith("http") or source.startswith("https"):
392 # remove the hhtp source from the sources list
393 sources.remove(source)
395 command.append( " ( "
396 # Check if the source already exists
397 " ls %(src_dir)s/%(basename)s "
399 # If source doesn't exist, download it and check
400 # that it it downloaded ok
401 " wget -c --directory-prefix=%(src_dir)s %(source)s && "
402 " ls %(src_dir)s/%(basename)s "
404 "basename": os.path.basename(source),
409 command = " && ".join(command)
411 # replace application specific paths in the command
412 command = self.replace_paths(command)
415 sources = ';'.join(sources)
416 self.node.upload(sources, src_dir, overwrite = False)
420 def upload_files(self):
421 files = self.get("files")
424 self.info("Uploading files %s " % files)
425 self.node.upload(files, self.node.share_dir, overwrite = False)
427 def upload_libraries(self):
428 libs = self.get("libs")
431 self.info("Uploading libraries %s " % libaries)
432 self.node.upload(libs, self.node.lib_dir, overwrite = False)
434 def upload_binaries(self):
435 bins = self.get("bins")
438 self.info("Uploading binaries %s " % binaries)
439 self.node.upload(bins, self.node.bin_dir, overwrite = False)
441 def upload_code(self):
442 code = self.get("code")
445 self.info("Uploading code")
447 dst = os.path.join(self.app_home, "code")
448 self.node.upload(code, dst, overwrite = False, text = True)
450 def upload_stdin(self):
451 stdin = self.get("stdin")
453 # create dir for sources
454 self.info("Uploading stdin")
456 # upload stdin file to ${SHARE_DIR} directory
457 basename = os.path.basename(stdin)
458 dst = os.path.join(self.node.share_dir, basename)
459 self.node.upload(stdin, dst, overwrite = False, text = True)
461 # create "stdin" symlink on ${APP_HOME} directory
462 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
463 "app_home": self.app_home,
468 def install_dependencies(self):
469 depends = self.get("depends")
471 self.info("Installing dependencies %s" % depends)
472 return self.node.install_packages_command(depends)
475 build = self.get("build")
478 self.info("Building sources ")
480 # replace application specific paths in the command
481 return self.replace_paths(build)
484 install = self.get("install")
487 self.info("Installing sources ")
489 # replace application specific paths in the command
490 return self.replace_paths(install)
493 # Wait until node is associated and deployed
495 if not node or node.state < ResourceState.READY:
496 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
497 self.ec.schedule(reschedule_delay, self.deploy)
499 command = self.get("command") or ""
500 self.info("Deploying command '%s' " % command)
504 super(LinuxApplication, self).do_deploy()
507 command = self.get("command")
509 self.info("Starting command '%s'" % command)
512 # If no command was given (i.e. Application was used for dependency
513 # installation), then the application is directly marked as STOPPED
514 super(LinuxApplication, self).set_stopped()
516 if self.in_foreground:
517 self._run_in_foreground()
519 self._run_in_background()
521 super(LinuxApplication, self).do_start()
523 def _run_in_foreground(self):
524 command = self.get("command")
525 sudo = self.get("sudo") or False
526 x11 = self.get("forwardX11")
527 env = self.get("env")
529 # Command will be launched in foreground and attached to the
530 # terminal using the node 'execute' in non blocking mode.
532 # We save the reference to the process in self._proc
533 # to be able to kill the process from the stop method.
534 # We also set blocking = False, since we don't want the
535 # thread to block until the execution finishes.
536 (out, err), self._proc = self.execute_command(command,
542 if self._proc.poll():
543 self.error(msg, out, err)
544 raise RuntimeError, msg
546 def _run_in_background(self):
547 command = self.get("command")
548 env = self.get("env")
549 sudo = self.get("sudo") or False
553 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
556 # Command will be run as a daemon in baground and detached from any
558 # The command to run was previously uploaded to a bash script
559 # during deployment, now we launch the remote script using 'run'
560 # method from the node.
561 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
562 (out, err), proc = self.node.run(cmd, self.run_home,
568 # check if execution errors occurred
569 msg = " Failed to start command '%s' " % command
572 self.error(msg, out, err)
573 raise RuntimeError, msg
575 # Wait for pid file to be generated
576 pid, ppid = self.node.wait_pid(self.run_home)
577 if pid: self._pid = int(pid)
578 if ppid: self._ppid = int(ppid)
580 # If the process is not running, check for error information
581 # on the remote machine
582 if not self.pid or not self.ppid:
583 (out, err), proc = self.node.check_errors(self.run_home,
586 # Out is what was written in the stderr file
588 msg = " Failed to start command '%s' " % command
589 self.error(msg, out, err)
590 raise RuntimeError, msg
593 """ Stops application execution
595 command = self.get('command') or ''
597 if self.state == ResourceState.STARTED:
599 self.info("Stopping command '%s' " % command)
601 # If the command is running in foreground (it was launched using
602 # the node 'execute' method), then we use the handler to the Popen
603 # process to kill it. Else we send a kill signal using the pid and ppid
604 # retrieved after running the command with the node 'run' method
608 # Only try to kill the process if the pid and ppid
610 if self.pid and self.ppid:
611 (out, err), proc = self.node.kill(self.pid, self.ppid,
612 sudo = self._sudo_kill)
614 # TODO: check if execution errors occurred
615 if (proc and proc.poll()) or err:
616 msg = " Failed to STOP command '%s' " % self.get("command")
617 self.error(msg, out, err)
619 super(LinuxApplication, self).do_stop()
621 def do_release(self):
622 self.info("Releasing resource")
624 tear_down = self.get("tearDown")
626 self.node.execute(tear_down)
630 super(LinuxApplication, self).do_release()
634 """ Returns the state of the application
636 if self._state == ResourceState.STARTED:
637 if self.in_foreground:
638 # Check if the process we used to execute the command
639 # is still running ...
640 retcode = self._proc.poll()
642 # retcode == None -> running
643 # retcode > 0 -> error
644 # retcode == 0 -> finished
647 msg = " Failed to execute command '%s'" % self.get("command")
648 err = self._proc.stderr.read()
649 self.error(msg, out, err)
655 # We need to query the status of the command we launched in
656 # background. In order to avoid overwhelming the remote host and
657 # the local processor with too many ssh queries, the state is only
658 # requested every 'state_check_delay' seconds.
659 state_check_delay = 0.5
660 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
661 if self.pid and self.ppid:
662 # Make sure the process is still running in background
663 status = self.node.status(self.pid, self.ppid)
665 if status == ProcStatus.FINISHED:
666 # If the program finished, check if execution
668 (out, err), proc = self.node.check_errors(
672 msg = "Failed to execute command '%s'" % \
674 self.error(msg, out, err)
679 self._last_state_check = tnow()
683 def execute_command(self, command,
691 environ = self.node.format_environment(env, inline = True)
692 command = environ + command
693 command = self.replace_paths(command)
695 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):