# Author: Alina Quereilhac <alina.quereilhac@inria.fr>
from nepi.execution.attribute import Attribute, Flags
-from nepi.execution.resource import ResourceManager, clsinit, ResourceState
+from nepi.execution.resource import ResourceManager, clsinit, ResourceState, \
+ reschedule_delay
from nepi.resources.linux import rpmfuncs, debfuncs
-from nepi.util import sshfuncs, execfuncs
+from nepi.util import sshfuncs, execfuncs
+from nepi.util.sshfuncs import ProcStatus
import collections
import os
# TODO: Unify delays!!
# TODO: Validate outcome of uploads!!
-reschedule_delay = "0.5s"
+class ExitCode:
+ """
+ Error codes that the rexitcode function can return if unable to
+ check the exit code of a spawned process
+ """
+ FILENOTFOUND = -1
+ CORRUPTFILE = -2
+ ERROR = -3
+ OK = 0
+
+class OSType:
+ """
+ Supported flavors of Linux OS
+ """
+ FEDORA_8 = "f8"
+ FEDORA_12 = "f12"
+ FEDORA_14 = "f14"
+ FEDORA = "fedora"
+ UBUNTU = "ubuntu"
+ DEBIAN = "debian"
@clsinit
class LinuxNode(ResourceManager):
+ """
+ .. class:: Class Args :
+
+ :param ec: The Experiment controller
+ :type ec: ExperimentController
+ :param guid: guid of the RM
+ :type guid: int
+
+ .. note::
+
+ There are different ways in which commands can be executed using the
+ LinuxNode interface (i.e. 'execute' - blocking and non blocking, 'run',
+ 'run_and_wait').
+
+ Brief explanation:
+
+ * 'execute' (blocking mode) :
+
+ HOW IT WORKS: 'execute', forks a process and run the
+ command, synchronously, attached to the terminal, in
+ foreground.
+ The execute method will block until the command returns
+ the result on 'out', 'err' (so until it finishes executing).
+
+ USAGE: short-lived commands that must be executed attached
+ to a terminal and in foreground, for which it IS necessary
+ to block until the command has finished (e.g. if you want
+ to run 'ls' or 'cat').
+
+ * 'execute' (NON blocking mode - blocking = False) :
+
+ HOW IT WORKS: Same as before, except that execute method
+ will return immediately (even if command still running).
+
+ USAGE: long-lived commands that must be executed attached
+ to a terminal and in foreground, but for which it is not
+ necessary to block until the command has finished. (e.g.
+ start an application using X11 forwarding)
+
+ * 'run' :
+
+ HOW IT WORKS: Connects to the host ( using SSH if remote)
+ and launches the command in background, detached from any
+ terminal (daemonized), and returns. The command continues to
+ run remotely, but since it is detached from the terminal,
+ its pipes (stdin, stdout, stderr) can't be redirected to the
+ console (as normal non detached processes would), and so they
+ are explicitly redirected to files. The pidfile is created as
+ part of the process of launching the command. The pidfile
+ holds the pid and ppid of the process forked in background,
+ so later on it is possible to check whether the command is still
+ running.
+
+ USAGE: long-lived commands that can run detached in background,
+ for which it is NOT necessary to block (wait) until the command
+ has finished. (e.g. start an application that is not using X11
+ forwarding. It can run detached and remotely in background)
+
+ * 'run_and_wait' :
+
+ HOW IT WORKS: Similar to 'run' except that it 'blocks' until
+ the command has finished execution. It also checks whether
+ errors occurred during runtime by reading the exitcode file,
+ which contains the exit code of the command that was run
+ (checking stderr only is not always reliable since many
+ commands throw debugging info to stderr and the only way to
+ automatically know whether an error really happened is to
+ check the process exit code).
+
+ Another difference with respect to 'run', is that instead
+ of directly executing the command as a bash command line,
+ it uploads the command to a bash script and runs the script.
+ This allows to use the bash script to debug errors, since
+ it remains at the remote host and can be run manually to
+ reproduce the error.
+
+ USAGE: medium-lived commands that can run detached in
+ background, for which it IS necessary to block (wait) until
+ the command has finished. (e.g. Package installation,
+ source compilation, file download, etc)
+
+ """
_rtype = "LinuxNode"
@classmethod
self.error(msg, out, err)
raise RuntimeError, "%s - %s - %s" %( msg, out, err )
- if out.find("Fedora release 12") == 0:
- self._os = "f12"
+ if out.find("Fedora release 8") == 0:
+ self._os = OSType.FEDORA_8
+ elif out.find("Fedora release 12") == 0:
+ self._os = OSType.FEDORA_12
elif out.find("Fedora release 14") == 0:
- self._os = "f14"
+ self._os = OSType.FEDORA_14
elif out.find("Debian") == 0:
- self._os = "debian"
+ self._os = OSType.DEBIAN
elif out.find("Ubuntu") ==0:
- self._os = "ubuntu"
+ self._os = OSType.UBUNTU
else:
msg = "Unsupported OS"
self.error(msg, out)
return self._os
+ @property
+ def use_deb(self):
+ return self.os in [OSType.DEBIAN, OSType.UBUNTU]
+
+ @property
+ def use_rpm(self):
+ return self.os in [OSType.FEDORA_12, OSType.FEDORA_14, OSType.FEDORA_8,
+ OSType.FEDORA]
+
@property
def localhost(self):
return self.get("hostname") in ['localhost', '127.0.0.7', '::1']
def deploy(self):
if self.state == ResourceState.NEW:
try:
- self.discover()
- self.provision()
+ self.discover()
+ self.provision()
except:
self._state = ResourceState.FAILED
raise
if not self.localhost:
# Build destination as <user>@<server>:<path>
dst = "%s@%s:%s" % (self.get("username"), self.get("hostname"), dst)
-
result = self.copy(src, dst)
# clean up temp file
src = "%s@%s:%s" % (self.get("username"), self.get("hostname"), src)
return self.copy(src, dst)
- def install_packages(self, packages, home = None):
- home = home or self.node_home
-
- cmd = ""
- if self.os in ["f12", "f14"]:
- cmd = rpmfuncs.install_packages_command(self.os, packages)
- elif self.os in ["debian", "ubuntu"]:
- cmd = debfuncs.install_packages_command(self.os, packages)
+ def install_packages(self, packages, home):
+ command = ""
+ if self.use_rpm:
+ command = rpmfuncs.install_packages_command(self.os, packages)
+ elif self.use_deb:
+ command = debfuncs.install_packages_command(self.os, packages)
else:
msg = "Error installing packages ( OS not known ) "
self.error(msg, self.os)
raise RuntimeError, msg
out = err = ""
- (out, err), proc = self.run_and_wait(cmd, home,
- pidfile = "instpkg_pid",
- stdout = "instpkg_out",
- stderr = "instpkg_err",
+ (out, err), proc = self.run_and_wait(command, home,
+ shfile = "instpkg.sh",
+ pidfile = "instpkg_pidfile",
+ ecodefile = "instpkg_exitcode",
+ stdout = "instpkg_stdout",
+ stderr = "instpkg_stderr",
raise_on_error = True)
return (out, err), proc
- def remove_packages(self, packages, home = None):
- home = home or self.node_home
-
- cmd = ""
- if self.os in ["f12", "f14"]:
- cmd = rpmfuncs.remove_packages_command(self.os, packages)
- elif self.os in ["debian", "ubuntu"]:
- cmd = debfuncs.remove_packages_command(self.os, packages)
+ def remove_packages(self, packages, home):
+ command = ""
+ if self.use_rpm:
+ command = rpmfuncs.remove_packages_command(self.os, packages)
+ elif self.use_deb:
+ command = debfuncs.remove_packages_command(self.os, packages)
else:
msg = "Error removing packages ( OS not known ) "
self.error(msg)
raise RuntimeError, msg
out = err = ""
- (out, err), proc = self.run_and_wait(cmd, home,
- pidfile = "rmpkg_pid",
- stdout = "rmpkg_out",
- stderr = "rmpkg_err",
+ (out, err), proc = self.run_and_wait(command, home,
+ shfile = "rmpkg.sh",
+ pidfile = "rmpkg_pidfile",
+ ecodefile = "rmpkg_exitcode",
+ stdout = "rmpkg_stdout",
+ stderr = "rmpkg_stderr",
raise_on_error = True)
return (out, err), proc
def rmdir(self, path):
return self.execute("rm -rf %s" % path, with_lock = True)
-
- def run_and_wait(self, command,
- home = ".",
- pidfile = "pid",
+
+ def run_and_wait(self, command, home,
+ shfile = "cmd.sh",
+ env = None,
+ pidfile = "pidfile",
+ ecodefile = "exitcode",
stdin = None,
- stdout = 'stdout',
- stderr = 'stderr',
+ stdout = "stdout",
+ stderr = "stderr",
sudo = False,
tty = False,
raise_on_error = False):
- """ runs a command in background on the remote host, but waits
- until the command finishes execution.
- This is more robust than doing a simple synchronized 'execute',
- since in the remote host the command can continue to run detached
- even if network disconnections occur
"""
+ Uploads the 'command' to a bash script in the host.
+ Then runs the script detached in background in the host, and
+ busy-waites until the script finishes executing.
+ """
+ self.upload_command(command, home,
+ shfile = shfile,
+ ecodefile = ecodefile,
+ env = env)
+
+ command = "bash ./%s" % shfile
# run command in background in remote host
(out, err), proc = self.run(command, home,
pidfile = pidfile,
tty = tty)
# check no errors occurred
- if proc.poll() and err:
+ if proc.poll():
msg = " Failed to run command '%s' " % command
self.error(msg, out, err)
if raise_on_error:
# wait until command finishes to execute
self.wait_run(pid, ppid)
-
- # check if execution errors occurred
- (out, err), proc = self.check_output(home, stderr)
-
- if err or out:
+
+ (out, err), proc = self.check_errors(home,
+ ecodefile = ecodefile,
+ stdout = stdout,
+ stderr= stderr)
+
+ # Out is what was written in the stderr file
+ if err:
msg = " Failed to run command '%s' " % command
self.error(msg, out, err)
raise RuntimeError, msg
return (out, err), proc
+
+ def exitcode(self, home, ecodefile = "exitcode"):
+ """
+ Get the exit code of an application.
+ Returns an integer value with the exit code
+ """
+ (out, err), proc = self.check_output(home, ecodefile)
+
+ # Succeeded to open file, return exit code in the file
+ if proc.wait() == 0:
+ try:
+ return int(out.strip())
+ except:
+ # Error in the content of the file!
+ return ExitCode.CORRUPTFILE
+
+ # No such file or directory
+ if proc.returncode == 1:
+ return ExitCode.FILENOTFOUND
+
+ # Other error from 'cat'
+ return ExitCode.ERROR
+
+ def upload_command(self, command, home,
+ shfile = "cmd.sh",
+ ecodefile = "exitcode",
+ env = None):
+ """ Saves the command as a bash script file in the remote host, and
+ forces to save the exit code of the command execution to the ecodefile
+ """
+
+ if not (command.strip().endswith(";") or command.strip().endswith("&")):
+ command += ";"
+
+ # The exit code of the command will be stored in ecodefile
+ command = " { %(command)s } ; echo $? > %(ecodefile)s ;" % {
+ 'command': command,
+ 'ecodefile': ecodefile,
+ }
+
+ # Export environment
+ environ = self.format_environment(env)
+
+ # Add environ to command
+ command = environ + command
+
+ dst = os.path.join(home, shfile)
+ return self.upload(command, dst, text = True)
+
+ def format_environment(self, env, inline = False):
+ """Format environmental variables for command to be executed either
+ as an inline command
+ (i.e. export PYTHONPATH=src/..; export LALAL= ..;python script.py) or
+ as a bash script (i.e. export PYTHONPATH=src/.. \n export LALA=.. \n)
+ """
+ if not env: return ""
+
+ # Remove extra white spaces
+ env = re.sub(r'\s+', ' ', env.strip())
+
+ sep = ";" if inline else "\n"
+ return sep.join(map(lambda e: " export %s" % e, env.split(" "))) + sep
+
+ def check_errors(self, home,
+ ecodefile = "exitcode",
+ stdout = "stdout",
+ stderr = "stderr"):
+ """
+ Checks whether errors occurred while running a command.
+ It first checks the exit code for the command, and only if the
+ exit code is an error one it returns the error output.
+
+ """
+ proc = None
+ err = ""
+ # retrive standard output from the file
+ (out, oerr), oproc = self.check_output(home, stdout)
+
+ # get exit code saved in the 'exitcode' file
+ ecode = self.exitcode(home, ecodefile)
+
+ if ecode in [ ExitCode.CORRUPTFILE, ExitCode.ERROR ]:
+ err = "Error retrieving exit code status from file %s/%s" % (home, ecodefile)
+ elif ecode > 0 or ecode == ExitCode.FILENOTFOUND:
+ # The process returned an error code or didn't exist.
+ # Check standard error.
+ (err, eerr), proc = self.check_output(home, stderr)
+
+ # If the stderr file was not found, assume nothing bad happened,
+ # and just ignore the error.
+ # (cat returns 1 for error "No such file or directory")
+ if ecode == ExitCode.FILENOTFOUND and proc.poll() == 1:
+ err = ""
+
+ return (out, err), proc
- def wait_pid(self, home = ".", pidfile = "pid", raise_on_error = False):
+ def wait_pid(self, home, pidfile = "pidfile", raise_on_error = False):
""" Waits until the pid file for the command is generated,
and returns the pid and ppid of the process """
pid = ppid = None
delay = 1.0
- for i in xrange(5):
- pidtuple = self.checkpid(home = home, pidfile = pidfile)
+
+ for i in xrange(4):
+ pidtuple = self.getpid(home = home, pidfile = pidfile)
if pidtuple:
pid, ppid = pidtuple
break
else:
time.sleep(delay)
- delay = min(30,delay*1.2)
+ delay = delay * 1.5
else:
msg = " Failed to get pid for pidfile %s/%s " % (
home, pidfile )
def wait_run(self, pid, ppid, trial = 0):
""" wait for a remote process to finish execution """
- delay = 1.0
- first = True
- bustspin = 0
+ start_delay = 1.0
while True:
status = self.status(pid, ppid)
- if status is sshfuncs.FINISHED:
+ if status is ProcStatus.FINISHED:
break
- elif status is not sshfuncs.RUNNING:
- bustspin += 1
- time.sleep(delay*(5.5+random.random()))
- if bustspin > 12:
+ elif status is not ProcStatus.RUNNING:
+ delay = delay * 1.5
+ time.sleep(delay)
+ # If it takes more than 20 seconds to start, then
+ # asume something went wrong
+ if delay > 20:
break
else:
- if first:
- first = False
-
- time.sleep(delay*(0.5+random.random()))
- delay = min(30,delay*1.2)
- bustspin = 0
+ # The app is running, just wait...
+ time.sleep(0.5)
def check_output(self, home, filename):
- """ checks file content """
+ """ Retrives content of file """
(out, err), proc = self.execute("cat %s" %
os.path.join(home, filename), retry = 1, with_lock = True)
return (out, err), proc
def copy(self, src, dst):
if self.localhost:
- (out, err), proc = execfuncs.lcopy(source, dest,
+ (out, err), proc = execfuncs.lcopy(source, dest,
recursive = True,
strict_host_checking = False)
else:
connect_timeout = 30,
strict_host_checking = False,
persistent = True,
+ blocking = True,
with_lock = False
):
""" Notice that this invocation will block until the
err_on_timeout = err_on_timeout,
connect_timeout = connect_timeout,
persistent = persistent,
+ blocking = blocking,
strict_host_checking = strict_host_checking
)
else:
retry = retry,
err_on_timeout = err_on_timeout,
connect_timeout = connect_timeout,
- persistent = persistent
+ persistent = persistent,
+ blocking = blocking,
+ strict_host_checking = strict_host_checking
)
return (out, err), proc
- def run(self, command,
- home = None,
+ def run(self, command, home,
create_home = False,
- pidfile = "pid",
+ pidfile = 'pidfile',
stdin = None,
stdout = 'stdout',
stderr = 'stderr',
sudo = False,
tty = False):
-
+
self.debug("Running command '%s'" % command)
if self.localhost:
sudo = sudo,
user = user)
else:
- # Start process in a "daemonized" way, using nohup and heavy
- # stdin/out redirection to avoid connection issues
with self._lock:
- (out,err), proc = sshfuncs.rspawn(
+ (out, err), proc = sshfuncs.rspawn(
command,
pidfile = pidfile,
home = home,
return (out, err), proc
- def checkpid(self, home = ".", pidfile = "pid"):
+ def getpid(self, home, pidfile = "pidfile"):
if self.localhost:
- pidtuple = execfuncs.lcheckpid(os.path.join(home, pidfile))
+ pidtuple = execfuncs.lgetpid(os.path.join(home, pidfile))
else:
with self._lock:
- pidtuple = sshfuncs.rcheckpid(
+ pidtuple = sshfuncs.rgetpid(
os.path.join(home, pidfile),
host = self.get("hostname"),
user = self.get("username"),
)
return pidtuple
-
+
def status(self, pid, ppid):
if self.localhost:
status = execfuncs.lstatus(pid, ppid)
proc = None
status = self.status(pid, ppid)
- if status == sshfuncs.RUNNING:
+ if status == sshfuncs.ProcStatus.RUNNING:
if self.localhost:
(out, err), proc = execfuncs.lkill(pid, ppid, sudo)
else:
identity = self.get("identity"),
server_key = self.get("serverKey")
)
- return (out, err), proc
- def check_bad_host(self, out, err):
- badre = re.compile(r'(?:'
- r'|Error: disk I/O error'
- r')',
- re.I)
- return badre.search(out) or badre.search(err)
+ return (out, err), proc
git://git.onelab.eu / nepi.git / blobdiff