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 # create run dir for application
275 self.node.mkdir(self.run_home)
277 # List of all the provision methods to invoke
284 self.upload_binaries,
286 self.upload_libraries,
291 # install dependencies
292 self.install_dependencies,
300 # Since provisioning takes a long time, before
301 # each step we check that the EC is still
304 self.debug("Interrupting provisioning. EC says 'ABORT")
311 # upload deploy script
312 deploy_command = ";".join(command)
313 self.execute_deploy_command(deploy_command)
315 # upload start script
316 self.upload_start_command()
318 self.info("Provisioning finished")
320 super(LinuxApplication, self).do_provision()
322 def upload_start_command(self, overwrite = False):
323 # Upload command to remote bash script
324 # - only if command can be executed in background and detached
325 command = self.get("command")
327 if command and not self.in_foreground:
328 self.info("Uploading command '%s'" % command)
330 # replace application specific paths in the command
331 command = self.replace_paths(command)
333 # replace application specific paths in the environment
334 env = self.get("env")
335 env = env and self.replace_paths(env)
337 shfile = os.path.join(self.app_home, "start.sh")
339 self.node.upload_command(command,
342 overwrite = overwrite)
344 def execute_deploy_command(self, command):
346 # Upload the command to a bash script and run it
347 # in background ( but wait until the command has
348 # finished to continue )
349 shfile = os.path.join(self.app_home, "deploy.sh")
350 self.node.run_and_wait(command, self.run_home,
353 pidfile = "deploy_pidfile",
354 ecodefile = "deploy_exitcode",
355 stdout = "deploy_stdout",
356 stderr = "deploy_stderr")
358 def upload_sources(self):
359 sources = self.get("sources")
364 self.info("Uploading sources ")
366 sources = sources.split(' ')
368 # Separate sources that should be downloaded from
369 # the web, from sources that should be uploaded from
372 for source in list(sources):
373 if source.startswith("http") or source.startswith("https"):
374 # remove the hhtp source from the sources list
375 sources.remove(source)
377 command.append( " ( "
378 # Check if the source already exists
379 " ls ${SRC}/%(basename)s "
381 # If source doesn't exist, download it and check
382 # that it it downloaded ok
383 " wget -c --directory-prefix=${SRC} %(source)s && "
384 " ls ${SRC}/%(basename)s "
386 "basename": os.path.basename(source),
390 command = " && ".join(command)
392 # replace application specific paths in the command
393 command = self.replace_paths(command)
396 sources = ' '.join(sources)
397 self.node.upload(sources, self.node.src_dir, overwrite = False)
401 def upload_files(self):
402 files = self.get("files")
405 self.info("Uploading files %s " % files)
406 self.node.upload(files, self.node.share_dir, overwrite = False)
408 def upload_libraries(self):
409 libs = self.get("libs")
412 self.info("Uploading libraries %s " % libaries)
413 self.node.upload(libs, self.node.lib_dir, overwrite = False)
415 def upload_binaries(self):
416 bins = self.get("bins")
419 self.info("Uploading binaries %s " % binaries)
420 self.node.upload(bins, self.node.bin_dir, overwrite = False)
422 def upload_code(self):
423 code = self.get("code")
426 self.info("Uploading code")
428 dst = os.path.join(self.app_home, "code")
429 self.node.upload(code, dst, overwrite = False, text = True)
431 def upload_stdin(self):
432 stdin = self.get("stdin")
434 # create dir for sources
435 self.info("Uploading stdin")
437 # upload stdin file to ${SHARE_DIR} directory
438 basename = os.path.basename(stdin)
439 dst = os.path.join(self.node.share_dir, basename)
440 self.node.upload(stdin, dst, overwrite = False, text = True)
442 # create "stdin" symlink on ${APP_HOME} directory
443 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
444 "app_home": self.app_home,
449 def install_dependencies(self):
450 depends = self.get("depends")
452 self.info("Installing dependencies %s" % depends)
453 return self.node.install_packages_command(depends)
456 build = self.get("build")
459 self.info("Building sources ")
461 # replace application specific paths in the command
462 return self.replace_paths(build)
465 install = self.get("install")
468 self.info("Installing sources ")
470 # replace application specific paths in the command
471 return self.replace_paths(install)
474 # Wait until node is associated and deployed
476 if not node or node.state < ResourceState.READY:
477 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
478 self.ec.schedule(reschedule_delay, self.deploy)
480 command = self.get("command") or ""
481 self.info("Deploying command '%s' " % command)
485 super(LinuxApplication, self).do_deploy()
488 command = self.get("command")
490 self.info("Starting command '%s'" % command)
493 # If no command was given (i.e. Application was used for dependency
494 # installation), then the application is directly marked as STOPPED
495 super(LinuxApplication, self).set_stopped()
497 if self.in_foreground:
498 self._run_in_foreground()
500 self._run_in_background()
502 super(LinuxApplication, self).do_start()
504 def _run_in_foreground(self):
505 command = self.get("command")
506 sudo = self.get("sudo") or False
507 x11 = self.get("forwardX11")
508 env = self.get("env")
510 # For a command being executed in foreground, if there is stdin,
511 # it is expected to be text string not a file or pipe
512 stdin = self.get("stdin") or None
514 # Command will be launched in foreground and attached to the
515 # terminal using the node 'execute' in non blocking mode.
517 # We save the reference to the process in self._proc
518 # to be able to kill the process from the stop method.
519 # We also set blocking = False, since we don't want the
520 # thread to block until the execution finishes.
521 (out, err), self._proc = self.execute_command(command,
528 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
558 self.error(msg, out, err)
559 raise RuntimeError, msg
561 # Wait for pid file to be generated
562 pid, ppid = self.node.wait_pid(self.run_home)
563 if pid: self._pid = int(pid)
564 if ppid: self._ppid = int(ppid)
566 # If the process is not running, check for error information
567 # on the remote machine
568 if not self.pid or not self.ppid:
569 (out, err), proc = self.node.check_errors(self.run_home,
572 # 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:
585 self.info("Stopping command '%s' " % command)
587 # If the command is running in foreground (it was launched using
588 # the node 'execute' method), then we use the handler to the Popen
589 # process to kill it. Else we send a kill signal using the pid and ppid
590 # retrieved after running the command with the node 'run' method
594 # Only try to kill the process if the pid and ppid
596 if self.pid and self.ppid:
597 (out, err), proc = self.node.kill(self.pid, self.ppid,
598 sudo = self._sudo_kill)
600 # TODO: check if execution errors occurred
601 if proc.poll() or err:
602 msg = " Failed to STOP command '%s' " % self.get("command")
603 self.error(msg, out, err)
605 super(LinuxApplication, self).do_stop()
607 def do_release(self):
608 self.info("Releasing resource")
610 tear_down = self.get("tearDown")
612 self.node.execute(tear_down)
616 super(LinuxApplication, self).do_release()
620 """ Returns the state of the application
622 if self._state == ResourceState.STARTED:
623 if self.in_foreground:
624 # Check if the process we used to execute the command
625 # is still running ...
626 retcode = self._proc.poll()
628 # retcode == None -> running
629 # retcode > 0 -> error
630 # retcode == 0 -> finished
633 msg = " Failed to execute command '%s'" % self.get("command")
634 err = self._proc.stderr.read()
635 self.error(msg, out, err)
641 # We need to query the status of the command we launched in
642 # background. In order to avoid overwhelming the remote host and
643 # the local processor with too many ssh queries, the state is only
644 # requested every 'state_check_delay' seconds.
645 state_check_delay = 0.5
646 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
647 if self.pid and self.ppid:
648 # Make sure the process is still running in background
649 status = self.node.status(self.pid, self.ppid)
651 if status == ProcStatus.FINISHED:
652 # If the program finished, check if execution
654 (out, err), proc = self.node.check_errors(
658 msg = "Failed to execute command '%s'" % \
660 self.error(msg, out, err)
665 self._last_state_check = tnow()
669 def execute_command(self, command,
678 environ = self.node.format_environment(env, inline = True)
679 command = environ + command
680 command = self.replace_paths(command)
682 return self.node.execute(command,
685 forward_x11 = forward_x11,
688 def replace_paths(self, command):
690 Replace all special path tags with shell-escaped actual paths.
693 .replace("${USR}", self.node.usr_dir)
694 .replace("${LIB}", self.node.lib_dir)
695 .replace("${BIN}", self.node.bin_dir)
696 .replace("${SRC}", self.node.src_dir)
697 .replace("${SHARE}", self.node.share_dir)
698 .replace("${EXP}", self.node.exp_dir)
699 .replace("${EXP_HOME}", self.node.exp_home)
700 .replace("${APP_HOME}", self.app_home)
701 .replace("${RUN_HOME}", self.run_home)
702 .replace("${NODE_HOME}", self.node.node_home)
703 .replace("${HOME}", self.node.home_dir)
706 def valid_connection(self, guid):