source: trunk/server/common/oursrc/hacron/hacron @ 1466

#!/usr/bin/env python
import glob
import logging.handlers
import fcntl
import optparse
import os
import socket
import shutil
import subprocess
import sys
import time
from os import path

OCF_SUCCESS=0
OCF_ERR_GENERIC=1
OCF_ERR_ARGS=2
OCF_ERR_UNIMPLEMENTED=3
OCF_ERR_PERM=4
OCF_ERR_INSTALLED=5
OCF_ERR_CONFIGURED=6
OCF_NOT_RUNNING=7

logger = logging.getLogger('cron')

HA_LOGD = os.environ.get('HA_LOGD') == 'yes'

class HacronError(Exception):
    def __init__(self, errno, msg='Something went wrong'):
        self.errno = errno
        self.msg = msg
    
class HaLogHandler(logging.Handler):
    """
    A handler class which writes to ha_logger.
    """
    def __init__(self, ha_tag):
        """
        Initialize the handler.  ha_tag is the name of this resource.
        """
        logging.Handler.__init__(self)
        self.ha_tag = ha_tag

    def emit(self, record):
        """
        Emit a record.
        """
        print 'Passed', record
        try:
            levelname = record.levelname
            msg = self.format(record)
            subprocess.call(['/usr/sbin/ha_logger', '-t', self.ha_tag, msg])
        except (KeyboardInterrupt, SystemExit):
            raise
        except:
            self.handleError(record)

class lock(object):
    def __init__(self, filename):
        self.filename = filename
        if not _touch(filename):
            raise

    def __enter__(self):
        f = open(self.filename)
        fcntl.flock(f, fcntl.LOCK_EX)
            
    def __exit__(self, type, value, traceback):
        f = open(self.filename)
        fcntl.flock(f, fcntl.LOCK_UN)
        
def _touch(path):
    """Effectively touches a file.  Returns true if successful, false
    otherwise"""
    try:
        open(path, 'a').close()
    except IOError:
        return False
    else:
        return True

def _remove(dest):
    try:
        if path.isdir(dest):
            os.rmdir(dest)
        else:
            os.remove(dest)
    except OSError, e:
        logging.error('Could not remove %s: %s' % (dest, e))
        return False
    else:
        return True

def _mkdir(dir):
    try:
        os.mkdir(dir)
    except OSError, e:
        logging.error('Could not mkdir %s: %s' % (dir, e))
        return False
    else:
        return True
    
def _strip(name):
    """Strip off the file extension, and leading /'s, if they exist"""
    return path.splitext(path.basename(name))[0]

def _suffix(name, suffix):
    return '%s.%s' % (name, suffix)

def _crondir(server):
    return path.join(CRONSPOOL_DIR, _suffix(server, 'cronspool'))

def _serverfile(server):
    return path.join(SERVER_DIR, server)

def _servers():
    """Get a list of the servers."""
    return [_strip(f) for f in glob.glob(path.join(SERVER_DIR, '*'))]

def _is_master(server):
    crondir = path.join(CRONSPOOL_DIR, _suffix(server, 'cronspool'))
    return path.islink(crondir)

def _restart_crond(args, options):
    # TODO: insert correct cmd here.  Also, should we capture and log
    # stdout?
    if options.development:
        cmd = ['echo', 'called crond reset']
    else:
        cmd = ['service', 'crond', 'reload']
    subprocess.check_call(cmd)

def start_cron(args, options):
    if not _touch(_serverfile(HOSTNAME)):
        return OCF_ERR_CONFIGURED
    elif _is_master(HOSTNAME):
        logger.error('%s is already the master!' % HOSTNAME)
        return OCF_SUCCESS

    logger.info('Starting %s' % HOSTNAME)
    for server in _servers():
        crondir = _crondir(server)
        if server == HOSTNAME:
            # Get rid of current crondir, and leave if that fails.
            if not _remove(crondir):
                logger.error("Could not remove dummy cronspool dir %s" % crondir)
                return OCF_ERR_GENERIC
            os.symlink('../cronspool', crondir)
            logger.info('Created master symlink %s' % crondir)
        else:
            if path.islink(crondir):
                _remove(crondir)
                logger.info('Removed old master symlink: %s' % crondir)
            if not path.exists(crondir):
                _mkdir(crondir)
                logger.info('Created slave dummy directory %s' % crondir)
    try:
        _restart_crond()
    except OSError, e:
        logger.error('Cron restart exited with return code %d' % e.errno)
        return OCF_ERR_GENERIC
    else:
        logger.info('Restarted crond')
    return OCF_SUCCESS

def stop_cron(args, options):
    """Stop cron."""
    if not _is_master(HOSTNAME):
        logger.error('I am not the master!')
        return OCF_NOT_RUNNING
    else:
        crondir = _crondir(HOSTNAME)
        logger.info('Removing symlink %s' % crondir)
        _remove(crondir)
        _mkdir(crondir)
        # TODO: should we do something else here?
        try:
            _restart_crond()
        except OSError, e:
            logger.error('Cron restart exited with return code %d' % e.errno)
            return OCF_ERR_GENERIC
        else:
            logger.info('Restarted crond')
        return OCF_SUCCESS

def monitor_cron(args, options):
    """Check whether cron is running.  For now just makes sure that the
    current machine is the master, although this should likely be fixed."""
    if _is_master(HOSTNAME):
        return OCF_SUCCESS
    else:
        return OCF_NOT_RUNNING

def validate_all_cron(args, options):
    if not _touch(_serverfile(HOSTNAME)):
        logger.error('Could not touch %s' % _serverfile(HOSTNAME))
        return OCF_GENERIC_ERR
    elif not path.exists(CRONSPOOL_DIR):
        return OCF_GENERIC_ERR
    else:
        return OCF_SUCCESS

def setup(args, options):
    for d in [CRONSPOOL_DIR, SERVER_DIR]:
        if not path.exists(d):
            os.makedirs(d)
            logger.info('Created %s' % d)
        else:
            logger.info('Already exists: %s' % d)

def remove_servers(servers, options):
    """Remove servers from the list of available ones."""
    for server in servers:
        os.unlink(_serverfile(server))

def meta_data_cron(args, options):
    print """<?xml version="1.0"?>
<!DOCTYPE resource-agent SYSTEM "ra-api-1.dtd">
<resource-agent name="hacron" version="0.1">
<version>1.0</version>

<longdesc lang="en">
This is the high-availability cron manager.  It uses an extremely overpowered
clustering solution to make it so that people can have their crontabs.  Yay.
</longdesc>
<shortdesc lang="en">HA Cron</shortdesc>

<parameters>
<parameter name="cron_root" required="1">
<longdesc lang="en">
Base directory for storage of crontabs and server information.
</longdesc>
<shortdesc lang="en">Cron base directory</shortdesc>
<content type="string" />
</parameter>
</parameters>

<actions>
<action name="start"        timeout="90" />
<action name="stop"         timeout="100" />
<action name="monitor"      timeout="20" interval="10" depth="0" start-delay="0" />
<action name="reload"       timeout="90" />
<action name="meta-data"    timeout="5" />
<action name="validate-all"   timeout="30" />
</actions>
</resource-agent>
"""
    return OCF_SUCCESS

def usage(parser):
    parser.print_help()
    return 1

def _set_globals(args, options):
    global HOSTNAME, CRONROOT, CRONSPOOL_DIR, SERVER_DIR, \
        HA_RSCTMP, OCF_RESOURCE_INSTANCE
    if options.development:
        logging.basicConfig(level=logging.DEBUG)
    else:
        if HA_LOGD:
            handler = HaLogHandler('hacron')
        else:
            handler = logging.handlers.SysLogHandler('/dev/log')
        formatter = logging.Formatter("%(module)s: %(levelname)s %(message)s")
        handler.setLevel(logging.INFO)
        handler.setFormatter(formatter)
        logger.addHandler(handler)
    HOSTNAME = options.server or os.environ.get('HA_CURHOST') or socket.gethostname()
    CRONROOT = options.cronroot or os.environ.get('OCF_RESKEY_cron_root')
    if not CRONROOT:
        raise HacronError(OCF_ERR_CONFIGURED, 'No cron_root specified.')
    CRONSPOOL_DIR = path.join(CRONROOT, 'server-cronspools')
    SERVER_DIR = path.join(CRONROOT, 'servers')
    HA_RSCTMP = os.environ.get('HA_RSCTMP', '/tmp')
    OCF_RESOURCE_INSTANCE = os.environ.get('OCF_RESOURCE_INSTANCE', 'default')
    return OCF_SUCCESS

def main():
    usage_str = """usage: %%prog [-s server] [-c cronroot] [-d] cmd

Script for starting and stopping cron in a multiserver environment.
One server is designated the master.

== HA available commands: ==
start: Make this server into the master and reload crond.
reload: Same as start.
stop: Demote this server to a spare and reload crond.
monitor: Indicate whether this server is successfully the master.
validate-all: Make sure that things look right and this server is
  ready to be promoted to master.
meta-data: Print out the XML meta data for this service

== User-only commands: ==
setup: Create the folders, etc. necessary for running hacron.
remove-servers server1 server2 ...: Take a list of servers out of the
  list of available ones.
    """
    parser = optparse.OptionParser(usage=usage_str)
    parser.add_option("-s", "--server",
                      action="store", dest="server",
                      default=None,
                      help="choose which server to run script as")
    parser.add_option("-c", "--cronroot",
                      action="store", dest="cronroot",
                      default=None,
                      help="pick root of cron dir")
    parser.add_option("-d", "--development",
                      action="store_true", dest="development",
                      default=False,
                      help="run in development mode")
    (options, args) = parser.parse_args()
    if len(args) < 1:
        return usage(parser)
    command = args[0]
    args = args[1:]

    if command == 'meta-data':
        return meta_data_cron(args, options)

    try:
        _set_globals(args, options)
    except HacronError, e:
        logger.error(e.msg)
        return e.errno

    with lock('%s/hacron-%s.lock' % (HA_RSCTMP, OCF_RESOURCE_INSTANCE)):
        if command == 'start':
            return start_cron(args, options)
        elif command == 'reload':
            return start_cron(args, options)
        elif command == 'stop':
            return stop_cron(args, options)
        elif command == 'monitor':
            return monitor_cron(args, options)
        elif command == 'validate-all':
            return validate_all_cron(args, options)
        elif command == 'setup':
            return setup(args, options)
        elif command == 'remove-servers':
            return remove_servers(args, options)
        else:
            usage(parser)
            return OCF_ERR_UNIMPLEMENTED

if __name__ == '__main__':
    try:
        ret = main()
    except Exception, e:
        logger.error('exception from main: %s' % e)
        ret = OCF_ERR_GENERIC
        raise
    sys.exit(ret)