def exited(self, status):
# Log last output
for line in self.readlines():
pass
# Display exit code
if status is not None:
log_func = self.warning
info = []
if WCOREDUMP(status):
info.append("core.%s dumped!" % self._pid)
log_func = self.error
if WIFSIGNALED(status):
signal = WTERMSIG(status)
signal = SIGNAME.get(signal, signal)
info.append("signal %s" % signal)
if WIFEXITED(status):
info.append("exitcode=%s" % WEXITSTATUS(status))
if info:
log_func("Exit (%s)" % ", ".join(info))
else:
log_func("Exit")
else:
self.error("Process exited (ECHILD error)")
# Delete process
self.process = None
self._pid = None
def _debug_event_iteration(self):
# (pid, status) = waitpid(self.pid, 0)
(pid, status) = waitpid(0, 0)
self.pid = pid
signum = 0
logger.debug('status:' + hex(status) + ' pid:' + str(pid))
# Process exited?
if WIFEXITED(status):
self._event_handle_process_exit(status, pid)
# Process killed by a signal?
elif WIFSIGNALED(status):
self._event_handle_process_kill(status, pid)
# Invalid process status?
elif not WIFSTOPPED(status):
self._event_handle_process_unknown_status(status, pid)
# Ptrace event?
elif self.WPTRACEEVENT(status):
self._event_handle_process_ptrace_event(status, pid)
else:
signum = self._event_handle_process_signal(status, pid)
# continue
if signum is None:
signum = 0
if self.single_step_flag | (self._restore_breakpoint is not None):
self._ptrace(PTRACE_SINGLESTEP, pid, 0, signum)
elif self.trace_sys_call_flag:
self._ptrace(PTRACE_SYSCALL, pid, 0, signum)
else:
self._ptrace(PTRACE_CONT, pid, 0, signum)
def wait_and_reap_zombies(pid):
_, status = waitpid(pid, 0)
try:
while True:
waitpid(-1, WNOHANG)
except ChildProcessError:
pass
if WIFSIGNALED(status):
return -WTERMSIG(status)
return WEXITSTATUS(status)
def execute(self, *args):
process = Popen4([self.facility.path] + self.facility.args +
list(args))
timers = [
Timer(time, self.kill, [process.pid, signal])
for time, signal in self.facility.wait.iteritems()
]
for timer in timers:
timer.start()
status = process.wait()
for timer in timers:
# No penalty, btw, for cancelling a dead timer
if timer.isAlive():
timer.cancel()
process.tochild.close()
if __debug__:
while True:
line = process.fromchild.readline()
if line:
syslog(line)
else:
break
process.fromchild.close()
command = basename(self.facility.path)
if WIFEXITED(status):
exit_status = WEXITSTATUS(status)
if exit_status != EX_OK:
raise ExecutionError(
EX_SOFTWARE, '`%(command)s\' exited with \
error-code %(exit_status)d' % {
'command': command,
'exit_status': exit_status
})
elif WIFSIGNALED(status):
raise ExecutionError(
EX_SOFTWARE, '`%(command)s\' terminated \
with signal %(signal)d' % {
'command': command,
'signal': WTERMSIG(status)
})
elif WIFSTOPPED(status):
raise ExecutionError(
EX_SOFTWARE, '`%(command)s\' stopped with \
signal %(signal)d' % {
'command': command,
'signal': WSTOPSIG(status)
})
else:
# Failsafe: timers should have killed the process by this point, or
# it should have ended naturally.
kill(process.pid, SIGKILL)
raise ExecutionError(
EX_SOFTWARE, 'Failed timer on `%(command)s\'; \
terminating the process extraordinarily.' % {'command': command})
def formatProcessStatus(status, title="Process"):
if WIFSTOPPED(status):
signum = WSTOPSIG(status)
return "%s stopped by signal %s" % (title, signalName(signum))
if WIFSIGNALED(status):
signum = WTERMSIG(status)
return "%s killed by signal %s" % (title, signalName(signum))
if not WIFEXITED(status):
raise ValueError("Invalid status: %r" % status)
exitcode = WEXITSTATUS(status)
if exitcode:
return "%s exited with code %s" % (title, exitcode)
else:
return "%s exited normally" % title
def _waitpid(pid):
"""Convenience wrapper around the original waitpid invocation."""
# 0 and -1 trigger a different behavior in waitpid. We disallow those
# values.
assert pid > 0
while True:
pid_, status = waitpid_(pid, 0)
assert pid_ == pid
if WIFEXITED(status):
return WEXITSTATUS(status)
elif WIFSIGNALED(status):
# Signals are usually represented as the negated signal number.
return -WTERMSIG(status)
elif WIFSTOPPED(status) or WIFCONTINUED(status):
# In our current usage scenarios we can simply ignore SIGSTOP and
# SIGCONT by restarting the wait.
continue
else:
assert False
return 1
def formatProcessStatus(status, title="Process"):
"""
Format a process status (integer) as a string.
"""
if WIFSTOPPED(status):
signum = WSTOPSIG(status)
text = "%s stopped by signal %s" % (title, signalName(signum))
elif WIFSIGNALED(status):
signum = WTERMSIG(status)
text = "%s killed by signal %s" % (title, signalName(signum))
else:
if not WIFEXITED(status):
raise ValueError("Invalid status: %r" % status)
exitcode = WEXITSTATUS(status)
if exitcode:
text = "%s exited with code %s" % (title, exitcode)
else:
text = "%s exited normally" % title
if WCOREDUMP(status):
text += " (core dumped)"
return text
def processStatus(self, status):
# Process exited?
if WIFEXITED(status):
code = WEXITSTATUS(status)
event = self.processExited(code)
# Process killed by a signal?
elif WIFSIGNALED(status):
signum = WTERMSIG(status)
event = self.processKilled(signum)
# Invalid process status?
elif not WIFSTOPPED(status):
raise ProcessError(self, "Unknown process status: %r" % status)
# Ptrace event?
elif HAS_PTRACE_EVENTS and WPTRACEEVENT(status):
event = WPTRACEEVENT(status)
event = self.ptraceEvent(event)
else:
signum = WSTOPSIG(status)
event = self.processSignal(signum)
return event
done += 1
elif not msg.has_key('cmd'):
mpd_raise('mpdrun: from man, invalid msg=:%s:' % (msg))
elif msg['cmd'] == 'execution_problem':
# print 'rank %d (%s) in job %s failed to find executable %s' % \
# ( msg['rank'], msg['src'], msg['jobid'], msg['exec'] )
host = msg['src'].split('_')[0]
reason = unquote(msg['reason'])
print 'problem with execution of %s on %s: %s ' % \
(msg['exec'],host,reason)
# keep going until all man's finish
elif msg['cmd'] == 'job_aborted_early':
print 'rank %d in job %s caused collective abort of all ranks' % \
( msg['rank'], msg['jobid'] )
status = msg['exit_status']
if WIFSIGNALED(status):
if status > myExitStatus:
myExitStatus = status
killed_status = status & 0x007f # AND off core flag
print ' exit status of rank %d: killed by signal %d ' % \
(msg['rank'],killed_status)
else:
exit_status = WEXITSTATUS(status)
if exit_status > myExitStatus:
myExitStatus = exit_status
print ' exit status of rank %d: return code %d ' % \
(msg['rank'],exit_status)
elif msg['cmd'] == 'job_aborted':
print 'job aborted; reason = %s' % (msg['reason'])
elif msg['cmd'] == 'client_exit_status':
if outXmlDoc:
