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 "semi-colon 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 "semi-colon 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 "semi-colon 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)
286 for line in out.strip().split("\n"):
287 parts = line.strip().split(" ")
288 procs[parts[0]] = parts[1]
289 pickle.dump(procs, open("/tmp/save.proc", "wb"))
291 # create run dir for application
292 self.node.mkdir(self.run_home)
294 # List of all the provision methods to invoke
301 self.upload_binaries,
303 self.upload_libraries,
308 # install dependencies
309 self.install_dependencies,
317 # Since provisioning takes a long time, before
318 # each step we check that the EC is still
321 self.debug("Interrupting provisioning. EC says 'ABORT")
328 # upload deploy script
329 deploy_command = ";".join(command)
330 self.execute_deploy_command(deploy_command)
332 # upload start script
333 self.upload_start_command()
335 self.info("Provisioning finished")
337 super(LinuxApplication, self).do_provision()
339 def upload_start_command(self, overwrite = False):
340 # Upload command to remote bash script
341 # - only if command can be executed in background and detached
342 command = self.get("command")
344 if command and not self.in_foreground:
345 self.info("Uploading command '%s'" % command)
347 # replace application specific paths in the command
348 command = self.replace_paths(command)
350 # replace application specific paths in the environment
351 env = self.get("env")
352 env = env and self.replace_paths(env)
354 shfile = os.path.join(self.app_home, "start.sh")
356 self.node.upload_command(command,
359 overwrite = overwrite)
361 def execute_deploy_command(self, command, prefix="deploy"):
363 # Upload the command to a bash script and run it
364 # in background ( but wait until the command has
365 # finished to continue )
366 shfile = os.path.join(self.app_home, "%s.sh" % prefix)
367 self.node.run_and_wait(command, self.run_home,
370 pidfile = "%s_pidfile" % prefix,
371 ecodefile = "%s_exitcode" % prefix,
372 stdout = "%s_stdout" % prefix,
373 stderr = "%s_stderr" % prefix)
375 def upload_sources(self, sources = None, src_dir = None):
377 sources = self.get("sources")
382 src_dir = self.node.src_dir
385 self.info("Uploading sources ")
387 sources = map(str.strip, sources.split(";"))
389 # Separate sources that should be downloaded from
390 # the web, from sources that should be uploaded from
393 for source in list(sources):
394 if source.startswith("http") or source.startswith("https"):
395 # remove the hhtp source from the sources list
396 sources.remove(source)
398 command.append( " ( "
399 # Check if the source already exists
400 " ls %(src_dir)s/%(basename)s "
402 # If source doesn't exist, download it and check
403 # that it it downloaded ok
404 " wget -c --directory-prefix=%(src_dir)s %(source)s && "
405 " ls %(src_dir)s/%(basename)s "
407 "basename": os.path.basename(source),
412 command = " && ".join(command)
414 # replace application specific paths in the command
415 command = self.replace_paths(command)
418 sources = ';'.join(sources)
419 self.node.upload(sources, src_dir, overwrite = False)
423 def upload_files(self, files = None):
425 files = self.get("files")
428 self.info("Uploading files %s " % files)
429 self.node.upload(files, self.node.share_dir, overwrite = False)
431 def upload_libraries(self, libs = None):
433 libs = self.get("libs")
436 self.info("Uploading libraries %s " % libaries)
437 self.node.upload(libs, self.node.lib_dir, overwrite = False)
439 def upload_binaries(self, bins = None):
441 bins = self.get("bins")
444 self.info("Uploading binaries %s " % binaries)
445 self.node.upload(bins, self.node.bin_dir, overwrite = False)
447 def upload_code(self, code = None):
449 code = self.get("code")
452 self.info("Uploading code")
454 dst = os.path.join(self.app_home, "code")
455 self.node.upload(code, dst, overwrite = False, text = True)
457 def upload_stdin(self, stdin = None):
459 stdin = self.get("stdin")
462 # create dir for sources
463 self.info("Uploading stdin")
465 # upload stdin file to ${SHARE_DIR} directory
466 if os.path.isfile(stdin):
467 basename = os.path.basename(stdin)
468 dst = os.path.join(self.node.share_dir, basename)
470 dst = os.path.join(self.app_home, "stdin")
472 self.node.upload(stdin, dst, overwrite = False, text = True)
474 # create "stdin" symlink on ${APP_HOME} directory
475 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
476 "app_home": self.app_home,
481 def install_dependencies(self, depends = None):
483 depends = self.get("depends")
486 self.info("Installing dependencies %s" % depends)
487 return self.node.install_packages_command(depends)
489 def build(self, build = None):
491 build = self.get("build")
494 self.info("Building sources ")
496 # replace application specific paths in the command
497 return self.replace_paths(build)
499 def install(self, install = None):
501 install = self.get("install")
504 self.info("Installing sources ")
506 # replace application specific paths in the command
507 return self.replace_paths(install)
510 # Wait until node is associated and deployed
512 if not node or node.state < ResourceState.READY:
513 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
514 self.ec.schedule(reschedule_delay, self.deploy)
516 command = self.get("command") or ""
517 self.info("Deploying command '%s' " % command)
521 super(LinuxApplication, self).do_deploy()
524 command = self.get("command")
526 self.info("Starting command '%s'" % command)
529 # If no command was given (i.e. Application was used for dependency
530 # installation), then the application is directly marked as STOPPED
531 super(LinuxApplication, self).set_stopped()
533 if self.in_foreground:
534 self._run_in_foreground()
536 self._run_in_background()
538 super(LinuxApplication, self).do_start()
540 def _run_in_foreground(self):
541 command = self.get("command")
542 sudo = self.get("sudo") or False
543 x11 = self.get("forwardX11")
544 env = self.get("env")
546 # Command will be launched in foreground and attached to the
547 # terminal using the node 'execute' in non blocking mode.
549 # We save the reference to the process in self._proc
550 # to be able to kill the process from the stop method.
551 # We also set blocking = False, since we don't want the
552 # thread to block until the execution finishes.
553 (out, err), self._proc = self.execute_command(command,
559 if self._proc.poll():
560 self.error(msg, out, err)
561 raise RuntimeError, msg
563 def _run_in_background(self):
564 command = self.get("command")
565 env = self.get("env")
566 sudo = self.get("sudo") or False
570 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
573 # Command will be run as a daemon in baground and detached from any
575 # The command to run was previously uploaded to a bash script
576 # during deployment, now we launch the remote script using 'run'
577 # method from the node.
578 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
579 (out, err), proc = self.node.run(cmd, self.run_home,
585 # check if execution errors occurred
586 msg = " Failed to start command '%s' " % command
589 self.error(msg, out, err)
590 raise RuntimeError, msg
592 # Wait for pid file to be generated
593 pid, ppid = self.node.wait_pid(self.run_home)
594 if pid: self._pid = int(pid)
595 if ppid: self._ppid = int(ppid)
597 # If the process is not running, check for error information
598 # on the remote machine
599 if not self.pid or not self.ppid:
600 (out, err), proc = self.node.check_errors(self.run_home,
603 # Out is what was written in the stderr file
605 msg = " Failed to start command '%s' " % command
606 self.error(msg, out, err)
607 raise RuntimeError, msg
610 """ Stops application execution
612 command = self.get('command') or ''
614 if self.state == ResourceState.STARTED:
616 self.info("Stopping command '%s' " % command)
618 # If the command is running in foreground (it was launched using
619 # the node 'execute' method), then we use the handler to the Popen
620 # process to kill it. Else we send a kill signal using the pid and ppid
621 # retrieved after running the command with the node 'run' method
625 # Only try to kill the process if the pid and ppid
627 if self.pid and self.ppid:
628 (out, err), proc = self.node.kill(self.pid, self.ppid,
629 sudo = self._sudo_kill)
631 # TODO: check if execution errors occurred
632 if (proc and proc.poll()) or err:
633 msg = " Failed to STOP command '%s' " % self.get("command")
634 self.error(msg, out, err)
636 super(LinuxApplication, self).do_stop()
638 def do_release(self):
639 self.info("Releasing resource")
643 tear_down = self.get("tearDown")
645 self.node.execute(tear_down)
647 hard_release = self.get("hardRelease")
649 self.node.rmdir(self.app_home)
651 super(LinuxApplication, self).do_release()
655 """ Returns the state of the application
657 if self._state == ResourceState.STARTED:
658 if self.in_foreground:
659 # Check if the process we used to execute the command
660 # is still running ...
661 retcode = self._proc.poll()
663 # retcode == None -> running
664 # retcode > 0 -> error
665 # retcode == 0 -> finished
668 msg = " Failed to execute command '%s'" % self.get("command")
669 err = self._proc.stderr.read()
670 self.error(msg, out, err)
676 # We need to query the status of the command we launched in
677 # background. In order to avoid overwhelming the remote host and
678 # the local processor with too many ssh queries, the state is only
679 # requested every 'state_check_delay' seconds.
680 state_check_delay = 0.5
681 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
682 if self.pid and self.ppid:
683 # Make sure the process is still running in background
684 status = self.node.status(self.pid, self.ppid)
686 if status == ProcStatus.FINISHED:
687 # If the program finished, check if execution
689 (out, err), proc = self.node.check_errors(
693 msg = "Failed to execute command '%s'" % \
695 self.error(msg, out, err)
700 self._last_state_check = tnow()
704 def execute_command(self, command,
712 environ = self.node.format_environment(env, inline = True)
713 command = environ + command
714 command = self.replace_paths(command)
716 return self.node.execute(command,
718 forward_x11 = forward_x11,
721 def replace_paths(self, command):
723 Replace all special path tags with shell-escaped actual paths.
726 .replace("${USR}", self.node.usr_dir)
727 .replace("${LIB}", self.node.lib_dir)
728 .replace("${BIN}", self.node.bin_dir)
729 .replace("${SRC}", self.node.src_dir)
730 .replace("${SHARE}", self.node.share_dir)
731 .replace("${EXP}", self.node.exp_dir)
732 .replace("${EXP_HOME}", self.node.exp_home)
733 .replace("${APP_HOME}", self.app_home)
734 .replace("${RUN_HOME}", self.run_home)
735 .replace("${NODE_HOME}", self.node.node_home)
736 .replace("${HOME}", self.node.home_dir)
739 def valid_connection(self, guid):