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, ResourceState, \
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!!
34 class LinuxApplication(ResourceManager):
36 .. class:: Class Args :
38 :param ec: The Experiment controller
39 :type ec: ExperimentController
40 :param guid: guid of the RM
45 A LinuxApplication RM represents a process that can be executed in
46 a remote Linux host using SSH.
48 The LinuxApplication RM takes care of uploadin sources and any files
49 needed to run the experiment, to the remote host.
50 It also allows to provide source compilation (build) and installation
51 instructions, and takes care of automating the sources build and
52 installation tasks for the user.
54 It is important to note that files uploaded to the remote host have
55 two possible scopes: single-experiment or multi-experiment.
56 Single experiment files are those that will not be re-used by other
57 experiments. Multi-experiment files are those that will.
58 Sources and shared files are always made available to all experiments.
62 The directory structure used by LinuxApplication RM at the Linux
63 host is the following:
65 ${HOME}/nepi-usr --> Base directory for multi-experiment files
67 ${LIB} |- /lib --> Base directory for libraries
68 ${BIN} |- /bin --> Base directory for binary files
69 ${SRC} |- /src --> Base directory for sources
70 ${SHARE} |- /share --> Base directory for other files
72 ${HOME}/nepi-exp --> Base directory for single-experiment files
74 ${EXP_HOME} |- /<exp-id> --> Base directory for experiment exp-id
76 ${APP_HOME} |- /<app-guid> --> Base directory for application
77 | specific files (e.g. command.sh, input)
79 ${RUN_HOME} |- /<run-id> --> Base directory for run specific
83 _rtype = "LinuxApplication"
86 def _register_attributes(cls):
87 command = Attribute("command", "Command to execute at application start. "
88 "Note that commands will be executed in the ${RUN_HOME} directory, "
89 "make sure to take this into account when using relative paths. ",
90 flags = Flags.ExecReadOnly)
91 forward_x11 = Attribute("forwardX11", "Enables X11 forwarding for SSH connections",
92 flags = Flags.ExecReadOnly)
93 env = Attribute("env", "Environment variables string for command execution",
94 flags = Flags.ExecReadOnly)
95 sudo = Attribute("sudo", "Run with root privileges",
96 flags = Flags.ExecReadOnly)
97 depends = Attribute("depends",
98 "Space-separated list of packages required to run the application",
99 flags = Flags.ExecReadOnly)
100 sources = Attribute("sources",
101 "Space-separated list of regular files to be uploaded to ${SRC} "
102 "directory prior to building. Archives won't be expanded automatically. "
103 "Sources are globally available for all experiments unless "
104 "cleanHome is set to True (This will delete all sources). ",
105 flags = Flags.ExecReadOnly)
106 files = Attribute("files",
107 "Space-separated list of regular miscellaneous files to be uploaded "
108 "to ${SHARE} directory. "
109 "Files are globally available for all experiments unless "
110 "cleanHome is set to True (This will delete all files). ",
111 flags = Flags.ExecReadOnly)
112 libs = Attribute("libs",
113 "Space-separated list of libraries (e.g. .so files) to be uploaded "
114 "to ${LIB} directory. "
115 "Libraries are globally available for all experiments unless "
116 "cleanHome is set to True (This will delete all files). ",
117 flags = Flags.ExecReadOnly)
118 bins = Attribute("bins",
119 "Space-separated list of binary files to be uploaded "
120 "to ${BIN} directory. "
121 "Binaries are globally available for all experiments unless "
122 "cleanHome is set to True (This will delete all files). ",
123 flags = Flags.ExecReadOnly)
124 code = Attribute("code",
125 "Plain text source code to be uploaded to the ${APP_HOME} directory. ",
126 flags = Flags.ExecReadOnly)
127 build = Attribute("build",
128 "Build commands to execute after deploying the sources. "
129 "Sources are uploaded to the ${SRC} directory and code "
130 "is uploaded to the ${APP_HOME} directory. \n"
131 "Usage example: tar xzf ${SRC}/my-app.tgz && cd my-app && "
132 "./configure && make && make clean.\n"
133 "Make sure to make the build commands return with a nonzero exit "
135 flags = Flags.ReadOnly)
136 install = Attribute("install",
137 "Commands to transfer built files to their final destinations. "
138 "Install commands are executed after build commands. ",
139 flags = Flags.ReadOnly)
140 stdin = Attribute("stdin", "Standard input for the 'command'",
141 flags = Flags.ExecReadOnly)
142 tear_down = Attribute("tearDown", "Command to be executed just before "
143 "releasing the resource",
144 flags = Flags.ReadOnly)
146 cls._register_attribute(command)
147 cls._register_attribute(forward_x11)
148 cls._register_attribute(env)
149 cls._register_attribute(sudo)
150 cls._register_attribute(depends)
151 cls._register_attribute(sources)
152 cls._register_attribute(code)
153 cls._register_attribute(files)
154 cls._register_attribute(bins)
155 cls._register_attribute(libs)
156 cls._register_attribute(build)
157 cls._register_attribute(install)
158 cls._register_attribute(stdin)
159 cls._register_attribute(tear_down)
162 def _register_traces(cls):
163 stdout = Trace("stdout", "Standard output stream")
164 stderr = Trace("stderr", "Standard error stream")
166 cls._register_trace(stdout)
167 cls._register_trace(stderr)
169 def __init__(self, ec, guid):
170 super(LinuxApplication, self).__init__(ec, guid)
173 self._home = "app-%s" % self.guid
174 self._in_foreground = False
176 # keep a reference to the running process handler when
177 # the command is not executed as remote daemon in background
180 # timestamp of last state check of the application
181 self._last_state_check = tnow()
183 def log_message(self, msg):
184 return " guid %d - host %s - %s " % (self.guid,
185 self.node.get("hostname"), msg)
189 node = self.get_connected(LinuxNode.rtype())
190 if node: return node[0]
195 return os.path.join(self.node.exp_home, self._home)
199 return os.path.join(self.app_home, self.ec.run_id)
210 def in_foreground(self):
211 """ Returns True if the command needs to be executed in foreground.
212 This means that command will be executed using 'execute' instead of
213 'run' ('run' executes a command in background and detached from the
216 When using X11 forwarding option, the command can not run in background
217 and detached from a terminal, since we need to keep the terminal attached
220 return self.get("forwardX11") or self._in_foreground
222 def trace(self, name, attr = TraceAttr.ALL, block = 512, offset = 0):
223 self.info("Retrieving '%s' trace %s " % (name, attr))
225 path = os.path.join(self.run_home, name)
227 command = "(test -f %s && echo 'success') || echo 'error'" % path
228 (out, err), proc = self.node.execute(command)
230 if (err and proc.poll()) or out.find("error") != -1:
231 msg = " Couldn't find trace %s " % name
232 self.error(msg, out, err)
235 if attr == TraceAttr.PATH:
238 if attr == TraceAttr.ALL:
239 (out, err), proc = self.node.check_output(self.run_home, name)
241 if err and proc.poll():
242 msg = " Couldn't read trace %s " % name
243 self.error(msg, out, err)
248 if attr == TraceAttr.STREAM:
249 cmd = "dd if=%s bs=%d count=1 skip=%d" % (path, block, offset)
250 elif attr == TraceAttr.SIZE:
251 cmd = "stat -c%%s %s " % path
253 (out, err), proc = self.node.execute(cmd)
255 if err and proc.poll():
256 msg = " Couldn't find trace %s " % name
257 self.error(msg, out, err)
260 if attr == TraceAttr.SIZE:
261 out = int(out.strip())
266 # create run dir for application
267 self.node.mkdir(self.run_home)
269 # List of all the provision methods to invoke
276 self.upload_binaries,
278 self.upload_libraries,
283 # install dependencies
284 self.install_dependencies,
292 # Since provisioning takes a long time, before
293 # each step we check that the EC is still
296 raise RuntimeError, "EC finished"
302 # upload deploy script
303 deploy_command = ";".join(command)
304 self.execute_deploy_command(deploy_command)
306 # upload start script
307 self.upload_start_command()
309 self.info("Provisioning finished")
311 super(LinuxApplication, self).provision()
313 def upload_start_command(self):
314 # Upload command to remote bash script
315 # - only if command can be executed in background and detached
316 command = self.get("command")
318 if command and not self.in_foreground:
319 self.info("Uploading command '%s'" % command)
321 # replace application specific paths in the command
322 command = self.replace_paths(command)
324 # replace application specific paths in the environment
325 env = self.get("env")
326 env = env and self.replace_paths(env)
328 shfile = os.path.join(self.app_home, "start.sh")
330 self.node.upload_command(command,
334 def execute_deploy_command(self, command):
336 # Upload the command to a bash script and run it
337 # in background ( but wait until the command has
338 # finished to continue )
339 shfile = os.path.join(self.app_home, "deploy.sh")
340 self.node.run_and_wait(command, self.run_home,
343 pidfile = "deploy_pidfile",
344 ecodefile = "deploy_exitcode",
345 stdout = "deploy_stdout",
346 stderr = "deploy_stderr")
348 def upload_sources(self):
349 sources = self.get("sources")
354 self.info("Uploading sources ")
356 sources = sources.split(' ')
358 # Separate sources that should be downloaded from
359 # the web, from sources that should be uploaded from
362 for source in list(sources):
363 if source.startswith("http") or source.startswith("https"):
364 # remove the hhtp source from the sources list
365 sources.remove(source)
367 command.append( " ( "
368 # Check if the source already exists
369 " ls ${SRC}/%(basename)s "
371 # If source doesn't exist, download it and check
372 # that it it downloaded ok
373 " wget -c --directory-prefix=${SRC} %(source)s && "
374 " ls ${SRC}/%(basename)s "
376 "basename": os.path.basename(source),
380 command = " && ".join(command)
382 # replace application specific paths in the command
383 command = self.replace_paths(command)
386 sources = ' '.join(sources)
387 self.node.upload(sources, self.node.src_dir, overwrite = False)
391 def upload_files(self):
392 files = self.get("files")
395 self.info("Uploading files %s " % files)
396 self.node.upload(files, self.node.share_dir, overwrite = False)
398 def upload_libraries(self):
399 libs = self.get("libs")
402 self.info("Uploading libraries %s " % libaries)
403 self.node.upload(libs, self.node.lib_dir, overwrite = False)
405 def upload_binaries(self):
406 bins = self.get("bins")
409 self.info("Uploading binaries %s " % binaries)
410 self.node.upload(bins, self.node.bin_dir, overwrite = False)
412 def upload_code(self):
413 code = self.get("code")
416 self.info("Uploading code")
418 dst = os.path.join(self.app_home, "code")
419 self.node.upload(code, dst, overwrite = False, text = True)
421 def upload_stdin(self):
422 stdin = self.get("stdin")
424 # create dir for sources
425 self.info("Uploading stdin")
427 # upload stdin file to ${SHARE_DIR} directory
428 basename = os.path.basename(stdin)
429 dst = os.path.join(self.node.share_dir, basename)
430 self.node.upload(stdin, dst, overwrite = False, text = True)
432 # create "stdin" symlink on ${APP_HOME} directory
433 command = "( cd %s ; ln -s %s stdin )" % ( self.app_home, dst)
437 def install_dependencies(self):
438 depends = self.get("depends")
440 self.info("Installing dependencies %s" % depends)
441 self.node.install_packages(depends, self.app_home, self.run_home)
444 build = self.get("build")
447 self.info("Building sources ")
449 # replace application specific paths in the command
450 return self.replace_paths(build)
453 install = self.get("install")
456 self.info("Installing sources ")
458 # replace application specific paths in the command
459 return self.replace_paths(install)
462 # Wait until node is associated and deployed
464 if not node or node.state < ResourceState.READY:
465 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
466 self.ec.schedule(reschedule_delay, self.deploy)
469 command = self.get("command") or ""
470 self.info("Deploying command '%s' " % command)
477 super(LinuxApplication, self).deploy()
480 command = self.get("command")
482 self.info("Starting command '%s'" % command)
485 # If no command was given (i.e. Application was used for dependency
486 # installation), then the application is directly marked as FINISHED
487 self._state = ResourceState.FINISHED
490 if self.in_foreground:
491 self._start_in_foreground()
493 self._start_in_background()
495 super(LinuxApplication, self).start()
497 def _start_in_foreground(self):
498 command = self.get("command")
499 sudo = self.get("sudo") or False
500 x11 = self.get("forwardX11")
502 # For a command being executed in foreground, if there is stdin,
503 # it is expected to be text string not a file or pipe
504 stdin = self.get("stdin") or None
506 # Command will be launched in foreground and attached to the
507 # terminal using the node 'execute' in non blocking mode.
510 env = self.get("env")
511 environ = self.node.format_environment(env, inline = True)
512 command = environ + command
513 command = self.replace_paths(command)
515 # We save the reference to the process in self._proc
516 # to be able to kill the process from the stop method.
517 # We also set blocking = False, since we don't want the
518 # thread to block until the execution finishes.
519 (out, err), self._proc = self.node.execute(command,
525 if self._proc.poll():
527 self.error(msg, out, err)
528 raise RuntimeError, msg
530 def _start_in_background(self):
531 command = self.get("command")
532 env = self.get("env")
533 sudo = self.get("sudo") or False
537 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
540 # Command will be run as a daemon in baground and detached from any
542 # The command to run was previously uploaded to a bash script
543 # during deployment, now we launch the remote script using 'run'
544 # method from the node.
545 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
546 (out, err), proc = self.node.run(cmd, self.run_home,
552 # check if execution errors occurred
553 msg = " Failed to start command '%s' " % command
557 self.error(msg, out, err)
558 raise RuntimeError, msg
560 # Wait for pid file to be generated
561 pid, ppid = self.node.wait_pid(self.run_home)
562 if pid: self._pid = int(pid)
563 if ppid: self._ppid = int(ppid)
565 # If the process is not running, check for error information
566 # on the remote machine
567 if not self.pid or not self.ppid:
568 (out, err), proc = self.node.check_errors(self.run_home,
571 # Out is what was written in the stderr file
574 msg = " Failed to start command '%s' " % command
575 self.error(msg, out, err)
576 raise RuntimeError, msg
579 """ Stops application execution
581 command = self.get('command') or ''
583 if self.state == ResourceState.STARTED:
586 self.info("Stopping command '%s'" % command)
588 # If the command is running in foreground (it was launched using
589 # the node 'execute' method), then we use the handler to the Popen
590 # process to kill it. Else we send a kill signal using the pid and ppid
591 # retrieved after running the command with the node 'run' method
596 # Only try to kill the process if the pid and ppid
598 if self.pid and self.ppid:
599 (out, err), proc = self.node.kill(self.pid, self.ppid)
602 # check if execution errors occurred
603 msg = " Failed to STOP command '%s' " % self.get("command")
604 self.error(msg, out, err)
609 super(LinuxApplication, self).stop()
612 self.info("Releasing resource")
614 tear_down = self.get("tearDown")
616 self.node.execute(tear_down)
620 if self.state == ResourceState.STOPPED:
621 super(LinuxApplication, self).release()
625 """ Returns the state of the application
627 if self._state == ResourceState.STARTED:
628 if self.in_foreground:
629 # Check if the process we used to execute the command
630 # is still running ...
631 retcode = self._proc.poll()
633 # retcode == None -> running
634 # retcode > 0 -> error
635 # retcode == 0 -> finished
638 msg = " Failed to execute command '%s'" % self.get("command")
639 err = self._proc.stderr.read()
640 self.error(msg, out, err)
643 self._state = ResourceState.FINISHED
646 # We need to query the status of the command we launched in
647 # background. In oredr to avoid overwhelming the remote host and
648 # the local processor with too many ssh queries, the state is only
649 # requested every 'state_check_delay' seconds.
650 state_check_delay = 0.5
651 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
652 # check if execution errors occurred
653 (out, err), proc = self.node.check_errors(self.run_home)
656 msg = " Failed to execute command '%s'" % self.get("command")
657 self.error(msg, out, err)
660 elif self.pid and self.ppid:
661 # No execution errors occurred. Make sure the background
662 # process with the recorded pid is still running.
663 status = self.node.status(self.pid, self.ppid)
665 if status == ProcStatus.FINISHED:
666 self._state = ResourceState.FINISHED
668 self._last_state_check = tnow()
672 def replace_paths(self, command):
674 Replace all special path tags with shell-escaped actual paths.
677 .replace("${USR}", self.node.usr_dir)
678 .replace("${LIB}", self.node.lib_dir)
679 .replace("${BIN}", self.node.bin_dir)
680 .replace("${SRC}", self.node.src_dir)
681 .replace("${SHARE}", self.node.share_dir)
682 .replace("${EXP}", self.node.exp_dir)
683 .replace("${EXP_HOME}", self.node.exp_home)
684 .replace("${APP_HOME}", self.app_home)
685 .replace("${RUN_HOME}", self.run_home)
686 .replace("${NODE_HOME}", self.node.node_home)
687 .replace("${HOME}", self.node.home_dir)
690 def valid_connection(self, guid):