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 # whether the command should run in foreground attached
176 self._in_foreground = False
178 # whether to use sudo to kill the application process
179 self._sudo_kill = False
181 # keep a reference to the running process handler when
182 # the command is not executed as remote daemon in background
185 # timestamp of last state check of the application
186 self._last_state_check = tnow()
188 def log_message(self, msg):
189 return " guid %d - host %s - %s " % (self.guid,
190 self.node.get("hostname"), msg)
194 node = self.get_connected(LinuxNode.rtype())
195 if node: return node[0]
200 return os.path.join(self.node.exp_home, self._home)
204 return os.path.join(self.app_home, self.ec.run_id)
215 def in_foreground(self):
216 """ Returns True if the command needs to be executed in foreground.
217 This means that command will be executed using 'execute' instead of
218 'run' ('run' executes a command in background and detached from the
221 When using X11 forwarding option, the command can not run in background
222 and detached from a terminal, since we need to keep the terminal attached
225 return self.get("forwardX11") or self._in_foreground
227 def trace(self, name, attr = TraceAttr.ALL, block = 512, offset = 0):
228 self.info("Retrieving '%s' trace %s " % (name, attr))
230 path = os.path.join(self.run_home, name)
232 command = "(test -f %s && echo 'success') || echo 'error'" % path
233 (out, err), proc = self.node.execute(command)
235 if (err and proc.poll()) or out.find("error") != -1:
236 msg = " Couldn't find trace %s " % name
237 self.error(msg, out, err)
240 if attr == TraceAttr.PATH:
243 if attr == TraceAttr.ALL:
244 (out, err), proc = self.node.check_output(self.run_home, name)
247 msg = " Couldn't read trace %s " % name
248 self.error(msg, out, err)
253 if attr == TraceAttr.STREAM:
254 cmd = "dd if=%s bs=%d count=1 skip=%d" % (path, block, offset)
255 elif attr == TraceAttr.SIZE:
256 cmd = "stat -c%%s %s " % path
258 (out, err), proc = self.node.execute(cmd)
261 msg = " Couldn't find trace %s " % name
262 self.error(msg, out, err)
265 if attr == TraceAttr.SIZE:
266 out = int(out.strip())
271 # create run dir for application
272 self.node.mkdir(self.run_home)
274 # List of all the provision methods to invoke
281 self.upload_binaries,
283 self.upload_libraries,
288 # install dependencies
289 self.install_dependencies,
297 # Since provisioning takes a long time, before
298 # each step we check that the EC is still
301 raise RuntimeError, "EC finished"
307 # upload deploy script
308 deploy_command = ";".join(command)
309 self.execute_deploy_command(deploy_command)
311 # upload start script
312 self.upload_start_command()
314 self.info("Provisioning finished")
316 super(LinuxApplication, self).provision()
318 def upload_start_command(self):
319 # Upload command to remote bash script
320 # - only if command can be executed in background and detached
321 command = self.get("command")
323 if command and not self.in_foreground:
324 self.info("Uploading command '%s'" % command)
326 # replace application specific paths in the command
327 command = self.replace_paths(command)
329 # replace application specific paths in the environment
330 env = self.get("env")
331 env = env and self.replace_paths(env)
333 shfile = os.path.join(self.app_home, "start.sh")
335 self.node.upload_command(command,
339 def execute_deploy_command(self, command):
341 # Upload the command to a bash script and run it
342 # in background ( but wait until the command has
343 # finished to continue )
344 shfile = os.path.join(self.app_home, "deploy.sh")
345 self.node.run_and_wait(command, self.run_home,
348 pidfile = "deploy_pidfile",
349 ecodefile = "deploy_exitcode",
350 stdout = "deploy_stdout",
351 stderr = "deploy_stderr")
353 def upload_sources(self):
354 sources = self.get("sources")
359 self.info("Uploading sources ")
361 sources = sources.split(' ')
363 # Separate sources that should be downloaded from
364 # the web, from sources that should be uploaded from
367 for source in list(sources):
368 if source.startswith("http") or source.startswith("https"):
369 # remove the hhtp source from the sources list
370 sources.remove(source)
372 command.append( " ( "
373 # Check if the source already exists
374 " ls ${SRC}/%(basename)s "
376 # If source doesn't exist, download it and check
377 # that it it downloaded ok
378 " wget -c --directory-prefix=${SRC} %(source)s && "
379 " ls ${SRC}/%(basename)s "
381 "basename": os.path.basename(source),
385 command = " && ".join(command)
387 # replace application specific paths in the command
388 command = self.replace_paths(command)
391 sources = ' '.join(sources)
392 self.node.upload(sources, self.node.src_dir, overwrite = False)
396 def upload_files(self):
397 files = self.get("files")
400 self.info("Uploading files %s " % files)
401 self.node.upload(files, self.node.share_dir, overwrite = False)
403 def upload_libraries(self):
404 libs = self.get("libs")
407 self.info("Uploading libraries %s " % libaries)
408 self.node.upload(libs, self.node.lib_dir, overwrite = False)
410 def upload_binaries(self):
411 bins = self.get("bins")
414 self.info("Uploading binaries %s " % binaries)
415 self.node.upload(bins, self.node.bin_dir, overwrite = False)
417 def upload_code(self):
418 code = self.get("code")
421 self.info("Uploading code")
423 dst = os.path.join(self.app_home, "code")
424 self.node.upload(code, dst, overwrite = False, text = True)
426 def upload_stdin(self):
427 stdin = self.get("stdin")
429 # create dir for sources
430 self.info("Uploading stdin")
432 # upload stdin file to ${SHARE_DIR} directory
433 basename = os.path.basename(stdin)
434 dst = os.path.join(self.node.share_dir, basename)
435 self.node.upload(stdin, dst, overwrite = False, text = True)
437 # create "stdin" symlink on ${APP_HOME} directory
438 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
439 "app_home": self.app_home,
444 def install_dependencies(self):
445 depends = self.get("depends")
447 self.info("Installing dependencies %s" % depends)
448 self.node.install_packages(depends, self.app_home, self.run_home)
451 build = self.get("build")
454 self.info("Building sources ")
456 # replace application specific paths in the command
457 return self.replace_paths(build)
460 install = self.get("install")
463 self.info("Installing sources ")
465 # replace application specific paths in the command
466 return self.replace_paths(install)
469 # Wait until node is associated and deployed
471 if not node or node.state < ResourceState.READY:
472 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
473 self.ec.schedule(reschedule_delay, self.deploy)
476 command = self.get("command") or ""
477 self.info("Deploying command '%s' " % command)
484 super(LinuxApplication, self).deploy()
487 command = self.get("command")
489 self.info("Starting command '%s'" % command)
492 # If no command was given (i.e. Application was used for dependency
493 # installation), then the application is directly marked as FINISHED
494 self._state = ResourceState.FINISHED
497 if self.in_foreground:
498 self._run_in_foreground()
500 self._run_in_background()
502 super(LinuxApplication, self).start()
504 def _run_in_foreground(self):
505 command = self.get("command")
506 sudo = self.get("sudo") or False
507 x11 = self.get("forwardX11")
509 # For a command being executed in foreground, if there is stdin,
510 # it is expected to be text string not a file or pipe
511 stdin = self.get("stdin") or None
513 # Command will be launched in foreground and attached to the
514 # terminal using the node 'execute' in non blocking mode.
516 # We save the reference to the process in self._proc
517 # to be able to kill the process from the stop method.
518 # We also set blocking = False, since we don't want the
519 # thread to block until the execution finishes.
520 (out, err), self._proc = self.execute_command(self, command,
527 if self._proc.poll():
529 self.error(msg, out, err)
530 raise RuntimeError, msg
532 def _run_in_background(self):
533 command = self.get("command")
534 env = self.get("env")
535 sudo = self.get("sudo") or False
539 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
542 # Command will be run as a daemon in baground and detached from any
544 # The command to run was previously uploaded to a bash script
545 # during deployment, now we launch the remote script using 'run'
546 # method from the node.
547 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
548 (out, err), proc = self.node.run(cmd, self.run_home,
554 # check if execution errors occurred
555 msg = " Failed to start command '%s' " % command
559 self.error(msg, out, err)
560 raise RuntimeError, msg
562 # Wait for pid file to be generated
563 pid, ppid = self.node.wait_pid(self.run_home)
564 if pid: self._pid = int(pid)
565 if ppid: self._ppid = int(ppid)
567 # If the process is not running, check for error information
568 # on the remote machine
569 if not self.pid or not self.ppid:
570 (out, err), proc = self.node.check_errors(self.run_home,
573 # Out is what was written in the stderr file
576 msg = " Failed to start command '%s' " % command
577 self.error(msg, out, err)
578 raise RuntimeError, msg
581 """ Stops application execution
583 command = self.get('command') or ''
585 if self.state == ResourceState.STARTED:
589 self.info("Stopping command '%s'" % command)
591 # If the command is running in foreground (it was launched using
592 # the node 'execute' method), then we use the handler to the Popen
593 # process to kill it. Else we send a kill signal using the pid and ppid
594 # retrieved after running the command with the node 'run' method
599 # Only try to kill the process if the pid and ppid
601 if self.pid and self.ppid:
602 (out, err), proc = self.node.kill(self.pid, self.ppid, sudo =
606 # check if execution errors occurred
607 msg = " Failed to STOP command '%s' " % self.get("command")
608 self.error(msg, out, err)
613 super(LinuxApplication, self).stop()
616 self.info("Releasing resource")
618 tear_down = self.get("tearDown")
620 self.node.execute(tear_down)
624 if self.state == ResourceState.STOPPED:
625 super(LinuxApplication, self).release()
629 """ Returns the state of the application
631 if self._state == ResourceState.STARTED:
632 if self.in_foreground:
633 # Check if the process we used to execute the command
634 # is still running ...
635 retcode = self._proc.poll()
637 # retcode == None -> running
638 # retcode > 0 -> error
639 # retcode == 0 -> finished
642 msg = " Failed to execute command '%s'" % self.get("command")
643 err = self._proc.stderr.read()
644 self.error(msg, out, err)
647 self._state = ResourceState.FINISHED
650 # We need to query the status of the command we launched in
651 # background. In oredr to avoid overwhelming the remote host and
652 # the local processor with too many ssh queries, the state is only
653 # requested every 'state_check_delay' seconds.
654 state_check_delay = 0.5
655 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
656 # check if execution errors occurred
657 (out, err), proc = self.node.check_errors(self.run_home)
660 msg = " Failed to execute command '%s'" % self.get("command")
661 self.error(msg, out, err)
664 elif self.pid and self.ppid:
665 # No execution errors occurred. Make sure the background
666 # process with the recorded pid is still running.
667 status = self.node.status(self.pid, self.ppid)
669 if status == ProcStatus.FINISHED:
670 self._state = ResourceState.FINISHED
672 self._last_state_check = tnow()
676 def execute_command(self, command,
685 environ = self.node.format_environment(env, inline = True)
686 command = environ + command
687 command = self.replace_paths(command)
689 return self.node.execute(command,
692 forward_x11 = forward_x11,
695 def replace_paths(self, command):
697 Replace all special path tags with shell-escaped actual paths.
700 .replace("${USR}", self.node.usr_dir)
701 .replace("${LIB}", self.node.lib_dir)
702 .replace("${BIN}", self.node.bin_dir)
703 .replace("${SRC}", self.node.src_dir)
704 .replace("${SHARE}", self.node.share_dir)
705 .replace("${EXP}", self.node.exp_dir)
706 .replace("${EXP_HOME}", self.node.exp_home)
707 .replace("${APP_HOME}", self.app_home)
708 .replace("${RUN_HOME}", self.run_home)
709 .replace("${NODE_HOME}", self.node.node_home)
710 .replace("${HOME}", self.node.home_dir)
713 def valid_connection(self, guid):