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. ",
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 "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.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 # create run dir for application
278 self.node.mkdir(self.run_home)
280 # List of all the provision methods to invoke
287 self.upload_binaries,
289 self.upload_libraries,
294 # install dependencies
295 self.install_dependencies,
303 # Since provisioning takes a long time, before
304 # each step we check that the EC is still
307 self.debug("Interrupting provisioning. EC says 'ABORT")
314 # upload deploy script
315 deploy_command = ";".join(command)
316 self.execute_deploy_command(deploy_command)
318 # upload start script
319 self.upload_start_command()
321 self.info("Provisioning finished")
323 super(LinuxApplication, self).do_provision()
325 def upload_start_command(self, overwrite = False):
326 # Upload command to remote bash script
327 # - only if command can be executed in background and detached
328 command = self.get("command")
330 if command and not self.in_foreground:
331 self.info("Uploading command '%s'" % command)
333 # replace application specific paths in the command
334 command = self.replace_paths(command)
336 # replace application specific paths in the environment
337 env = self.get("env")
338 env = env and self.replace_paths(env)
340 shfile = os.path.join(self.app_home, "start.sh")
342 self.node.upload_command(command,
345 overwrite = overwrite)
347 def execute_deploy_command(self, command):
349 # Upload the command to a bash script and run it
350 # in background ( but wait until the command has
351 # finished to continue )
352 shfile = os.path.join(self.app_home, "deploy.sh")
353 self.node.run_and_wait(command, self.run_home,
356 pidfile = "deploy_pidfile",
357 ecodefile = "deploy_exitcode",
358 stdout = "deploy_stdout",
359 stderr = "deploy_stderr")
361 def upload_sources(self, src_dir = None):
362 sources = self.get("sources")
367 src_dir = self.node.src_dir
370 self.info("Uploading sources ")
372 sources = sources.split(' ')
374 # Separate sources that should be downloaded from
375 # the web, from sources that should be uploaded from
378 for source in list(sources):
379 if source.startswith("http") or source.startswith("https"):
380 # remove the hhtp source from the sources list
381 sources.remove(source)
383 command.append( " ( "
384 # Check if the source already exists
385 " ls %(src_dir)s/%(basename)s "
387 # If source doesn't exist, download it and check
388 # that it it downloaded ok
389 " wget -c --directory-prefix=%(src_dir)s %(source)s && "
390 " ls %(src_dir)s/%(basename)s "
392 "basename": os.path.basename(source),
397 command = " && ".join(command)
399 # replace application specific paths in the command
400 command = self.replace_paths(command)
403 sources = ' '.join(sources)
404 self.node.upload(sources, src_dir, overwrite = False)
408 def upload_files(self):
409 files = self.get("files")
412 self.info("Uploading files %s " % files)
413 self.node.upload(files, self.node.share_dir, overwrite = False)
415 def upload_libraries(self):
416 libs = self.get("libs")
419 self.info("Uploading libraries %s " % libaries)
420 self.node.upload(libs, self.node.lib_dir, overwrite = False)
422 def upload_binaries(self):
423 bins = self.get("bins")
426 self.info("Uploading binaries %s " % binaries)
427 self.node.upload(bins, self.node.bin_dir, overwrite = False)
429 def upload_code(self):
430 code = self.get("code")
433 self.info("Uploading code")
435 dst = os.path.join(self.app_home, "code")
436 self.node.upload(code, dst, overwrite = False, text = True)
438 def upload_stdin(self):
439 stdin = self.get("stdin")
441 # create dir for sources
442 self.info("Uploading stdin")
444 # upload stdin file to ${SHARE_DIR} directory
445 basename = os.path.basename(stdin)
446 dst = os.path.join(self.node.share_dir, basename)
447 self.node.upload(stdin, dst, overwrite = False, text = True)
449 # create "stdin" symlink on ${APP_HOME} directory
450 command = "( cd %(app_home)s ; [ ! -f stdin ] && ln -s %(stdin)s stdin )" % ({
451 "app_home": self.app_home,
456 def install_dependencies(self):
457 depends = self.get("depends")
459 self.info("Installing dependencies %s" % depends)
460 return self.node.install_packages_command(depends)
463 build = self.get("build")
466 self.info("Building sources ")
468 # replace application specific paths in the command
469 return self.replace_paths(build)
472 install = self.get("install")
475 self.info("Installing sources ")
477 # replace application specific paths in the command
478 return self.replace_paths(install)
481 # Wait until node is associated and deployed
483 if not node or node.state < ResourceState.READY:
484 self.debug("---- RESCHEDULING DEPLOY ---- node state %s " % self.node.state )
485 self.ec.schedule(reschedule_delay, self.deploy)
487 command = self.get("command") or ""
488 self.info("Deploying command '%s' " % command)
492 super(LinuxApplication, self).do_deploy()
495 command = self.get("command")
497 self.info("Starting command '%s'" % command)
500 # If no command was given (i.e. Application was used for dependency
501 # installation), then the application is directly marked as STOPPED
502 super(LinuxApplication, self).set_stopped()
504 if self.in_foreground:
505 self._run_in_foreground()
507 self._run_in_background()
509 super(LinuxApplication, self).do_start()
511 def _run_in_foreground(self):
512 command = self.get("command")
513 sudo = self.get("sudo") or False
514 x11 = self.get("forwardX11")
515 env = self.get("env")
517 # Command will be launched in foreground and attached to the
518 # terminal using the node 'execute' in non blocking mode.
520 # We save the reference to the process in self._proc
521 # to be able to kill the process from the stop method.
522 # We also set blocking = False, since we don't want the
523 # thread to block until the execution finishes.
524 (out, err), self._proc = self.execute_command(command,
530 if self._proc.poll():
531 self.error(msg, out, err)
532 raise RuntimeError, msg
534 def _run_in_background(self):
535 command = self.get("command")
536 env = self.get("env")
537 sudo = self.get("sudo") or False
541 stdin = os.path.join(self.app_home, "stdin") if self.get("stdin") \
544 # Command will be run as a daemon in baground and detached from any
546 # The command to run was previously uploaded to a bash script
547 # during deployment, now we launch the remote script using 'run'
548 # method from the node.
549 cmd = "bash %s" % os.path.join(self.app_home, "start.sh")
550 (out, err), proc = self.node.run(cmd, self.run_home,
556 # check if execution errors occurred
557 msg = " Failed to start command '%s' " % command
560 self.error(msg, out, err)
561 raise RuntimeError, msg
563 # Wait for pid file to be generated
564 pid, ppid = self.node.wait_pid(self.run_home)
565 if pid: self._pid = int(pid)
566 if ppid: self._ppid = int(ppid)
568 # If the process is not running, check for error information
569 # on the remote machine
570 if not self.pid or not self.ppid:
571 (out, err), proc = self.node.check_errors(self.run_home,
574 # 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:
587 self.info("Stopping command '%s' " % command)
589 # If the command is running in foreground (it was launched using
590 # the node 'execute' method), then we use the handler to the Popen
591 # process to kill it. Else we send a kill signal using the pid and ppid
592 # 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,
600 sudo = self._sudo_kill)
602 # TODO: check if execution errors occurred
603 if (proc and proc.poll()) or err:
604 msg = " Failed to STOP command '%s' " % self.get("command")
605 self.error(msg, out, err)
607 super(LinuxApplication, self).do_stop()
609 def do_release(self):
610 self.info("Releasing resource")
612 tear_down = self.get("tearDown")
614 self.node.execute(tear_down)
618 super(LinuxApplication, self).do_release()
622 """ Returns the state of the application
624 if self._state == ResourceState.STARTED:
625 if self.in_foreground:
626 # Check if the process we used to execute the command
627 # is still running ...
628 retcode = self._proc.poll()
630 # retcode == None -> running
631 # retcode > 0 -> error
632 # retcode == 0 -> finished
635 msg = " Failed to execute command '%s'" % self.get("command")
636 err = self._proc.stderr.read()
637 self.error(msg, out, err)
643 # We need to query the status of the command we launched in
644 # background. In order to avoid overwhelming the remote host and
645 # the local processor with too many ssh queries, the state is only
646 # requested every 'state_check_delay' seconds.
647 state_check_delay = 0.5
648 if tdiffsec(tnow(), self._last_state_check) > state_check_delay:
649 if self.pid and self.ppid:
650 # Make sure the process is still running in background
651 status = self.node.status(self.pid, self.ppid)
653 if status == ProcStatus.FINISHED:
654 # If the program finished, check if execution
656 (out, err), proc = self.node.check_errors(
660 msg = "Failed to execute command '%s'" % \
662 self.error(msg, out, err)
667 self._last_state_check = tnow()
671 def execute_command(self, command,
679 environ = self.node.format_environment(env, inline = True)
680 command = environ + command
681 command = self.replace_paths(command)
683 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):