server/autoserv

#!/usr/bin/python -u
#
# Copyright 2007 Google Inc. Released under the GPL v2

"""
Run an control file through the server side engine
"""

__author__ = """\
mbligh@google.com (Martin J. Bligh)
"""

import sys, os, re, traceback, signal

import common
from autotest_lib.server import server_job, utils

usage = """\
usage: autoserv
	[-h, --help]               # This help message
	[-m machine,[machine,...]] # list of machines to pass to control file
	[-M machines_file]         # list of machines (from a file)
	[-c]                       # control file is a client side control
	[-r resultsdir]            # specify results directory (default '.')
	[-i]                       # reinstall machines before running the job
	[-I]                       # reinstall machines after running the job
	[-b]                       # reboot all specified machines after the job
	[-l label]                 # label for the job (arbitrary string)
	[-u user]                  # username for the job (email address)
	[-v]                       # verify the machines only
	[-R]                       # repair the machines
	[-n]                       # no teeing the status to stdout/stderr
	[-p]                       # write pidfile (.autoserv_execute)
	[-P jobname]               # parse the results of the job
	<control file>             # name of the control file to run
	[args ...]                 # args to pass through to the control file
"""


class PidFileManager(object):
	pid_file = None

	def open_pid_file(self, results_dir):
		pid_file_path = os.path.join(results_dir, '.autoserv_execute')
		assert not os.path.exists(pid_file_path)
		self.pid_file = open(pid_file_path, 'w')
		self.pid_file.write(str(os.getpid()) + '\n')
		self.pid_file.flush()


	def close_pid_file(self, exit_code, signal_code=0):
		if not self.pid_file:
			return
		real_exit_code = (exit_code << 8) | (signal_code & 0xFF)
		self.pid_file.write(str(real_exit_code) + '\n')
		self.pid_file.close()
		self.pid_file = None


def run_autoserv(pid_file_manager, results, parser):
	# send stdin to /dev/null
	dev_null = os.open(os.devnull, os.O_RDONLY)
	os.dup2(dev_null, sys.stdin.fileno())
	os.close(dev_null)

	# Create separate process group
	os.setpgrp()

	# Implement SIGTERM handler
	def handle_sigint(signum, frame):
		pid_file_manager.close_pid_file(1, signal.SIGTERM)
		os.killpg(os.getpgrp(), signal.SIGKILL)

	# Set signal handler
	signal.signal(signal.SIGTERM, handle_sigint)

	# Get a useful value for running 'USER'
	realuser = os.environ.get('USER')
	if not realuser:
		realuser = 'anonymous'

	machines = parser.parse_opts_param('-m', None, split = ',')
	machines_file = parser.parse_opts_param('-M', None)
	label    = parser.parse_opts_param('-l', '')
	user     = parser.parse_opts_param('-u', realuser)
	client   = parser.parse_opts('-c')
	reboot   = parser.parse_opts('-b')
	install_before = parser.parse_opts('-i')
	install_after  = parser.parse_opts('-I')
	verify   = parser.parse_opts('-v')
	repair   = parser.parse_opts('-R')
	no_tee   = parser.parse_opts('-n')
	help     = parser.parse_opts('-h') or parser.parse_opts('--help')
	parse_job = parser.parse_opts_param('-P', '')

	if help is True:
		print usage
		sys.exit(0)

	if len(parser.args) < 1 and not verify and not repair:
		print usage
		sys.exit(1)

	# We have a control file unless it's just a verify/repair job
	if len(parser.args) > 0:
		control = parser.args[0]
	else:
		control = None

	if machines_file:
		machines = []
		for m in open(machines_file, 'r').readlines():
			# remove comments, spaces
			m = re.sub('#.*', '', m).strip()
			if m:
				machines.append(m)
		print "Read list of machines from file: %s" % machines_file
		print ','.join(machines)

	if machines:
		for machine in machines:
			if not machine or re.search('\s', machine):
				print "Invalid machine %s" % str(machine)
				sys.exit(1)
		machines = list(set(machines))
		machines.sort()

	job = server_job.server_job(control, parser.args[1:], results, label,
				    user, machines, client, parse_job)
	debug_dir = os.path.join(results, 'debug')
	stdout = os.path.join(debug_dir, 'autoserv.stdout')
	stderr = os.path.join(debug_dir, 'autoserv.stderr')
	if no_tee:
		job.stdout.redirect(stdout)
		job.stderr.redirect(stderr)
	else:
		job.stdout.tee_redirect(stdout)
		job.stderr.tee_redirect(stderr)

	# run the job
	exit_code = 0
	try:
		if repair:
			job.repair()
		elif verify:
			job.verify()
		else:
			try:
				job.run(reboot, install_before, install_after)
			finally:
				job.cleanup_parser()
	except:
		job.aborted = True
		traceback.print_exc()

	if getattr(job, 'aborted', False):
		sys.exit(1)


def main():
	pid_file_manager = PidFileManager()

	args = sys.argv[1:]
	parser = utils.AutoservOptionParser(args)

	results  = parser.parse_opts_param('-r', '.')
	results  = os.path.abspath(results)
	write_pidfile = parser.parse_opts('-p')
	if write_pidfile:
		pid_file_manager.open_pid_file(results)

	exit_code = 0
	try:
		try:
			run_autoserv(pid_file_manager, results, parser)
		except SystemExit, e:
			exit_code = e.code
		except:
			# If we don't know what happened, we'll classify it as
			# an 'abort' and return 1.
			exit_code = 1
	finally:
		pid_file_manager.close_pid_file(exit_code)
	sys.exit(exit_code)


if __name__ == '__main__':
	main()