def __init__(self):
self.opts = SaltEventsdLoader().getopts()
self._pre_startup(self.opts)
if type(self.opts) is not dict:
log.info("Received invalid configdata, startup cancelled")
sys.exit(1)
self.config = self.opts['general']
super(SaltEventsDaemon, self).__init__(self.config['pidfile'])
# the map of events are stored here, loaded in _init_events()
self.event_map = None
self._init_events(self.opts['events'])
self.backends = self._init_backends( self.config['backends'] )
log.info(self.backends)
# the socket to listen on for the events
self.sock_dir = self.config['sock_dir']
# two possible values for 'node': master and minion
# they do the same thing, just on different sockets
self.node = self.config['node']
# the id, usually 'master'
self.nodeid = self.config['id']
# the statefile where we write the daemon status
self.state_file = self.config['state_file']
# how many events to handle before updating the status
self.state_upd = self.config['state_upd']
# we dont know our pid (yet)
self.pid = None
# how many parallel workers to start max
self.max_workers = self.config['max_workers']
# the number of events to collect before starting a worker
self.event_limit = self.config['event_limit']
# a list to keep track of the currently running workers
# this is mainly for debugging to check wether all started
# workers are correctly joined over time so we dont leak memory
self.running_workers = []
# setup some counters used for the status
self.events_han = 0
self.events_rec = 0
self.threads_cre = 0
self.threads_join = 0
# the timer thats write data to the database every x seconds
# this is used to push data into the database even if
# self.event_limit is not reached regularly
self.ev_timer_ev = False
self.ev_timer_intrvl = self.config['dump_timer']
self.ev_timer = ResetTimer(self.ev_timer_intrvl,
self)
class SaltEventsDaemon(Daemon):
'''
The main daemon class where all the parsing, collecting
and dumping takes place
'''
def __init__(self, config, log_level=None, log_file=None, daemonize=False):
self.opts = SaltEventsdLoader(
config=config,
log_level=log_level,
log_file=log_file,
daemonize=daemonize,
).getopts()
self._pre_startup(self.opts)
if type(self.opts) is not dict:
log.info("Received invalid configdata, startup cancelled")
sys.exit(1)
self.config = self.opts['general']
super(SaltEventsDaemon, self).__init__(self.config['pidfile'])
# the map of events are stored here, loaded in _init_events()
self.event_map = None
self._init_events(self.opts['events'])
self.backends = self._init_backends(self.config['backends'])
# the socket to listen on for the events
self.sock_dir = self.config['sock_dir']
# two possible values for 'node': master and minion
# they do the same thing, just on different sockets
self.node = self.config['node']
# the id, usually 'master'
self.nodeid = self.config['id']
# the statefile where we write the daemon status
self.state_file = self.config['state_file']
# how many events to handle before updating the status
self.state_timer = self.config['state_timer']
# we dont know our pid (yet), its updated in run()
self.pid = None
# how many parallel workers to start max
self.max_workers = self.config['max_workers']
# the number of events to collect before starting a worker
self.event_limit = self.config['event_limit']
# a list to keep track of the currently running workers
# this is mainly for debugging to check wether all started
# workers are correctly joined over time so we dont leak memory
self.running_workers = []
# setup some counters used for the status
# the events we matched on
self.events_han = 0
self.events_rec = 0
# the threads we have created and joined
# if there numbers diverge, we have a memory leak
self.threads_cre = 0
self.threads_join = 0
# used to keep track of the delta between two stat_timer intervals
# to calculate events per second handled/received
self.stat_rec_count = 0
self.stat_hdl_count = 0
# the timer thats write data to the database every x seconds
# this is used to push data into the database even if
# self.event_limit is not reached regularly
self.ev_timer_ev = False
self.ev_timer_intrvl = self.config['dump_timer']
self.ev_timer = ResetTimer(self.ev_timer_intrvl, self, name='Event')
# the timer that writes statistical data to the status-file
self.state_timer_ev = False
self.state_timer_intrvl = self.config['state_timer']
self.state_timer = ResetTimer(self.state_timer_intrvl,
self,
name='Stat')
def timer_event(self, source=None):
'''
This is called whenever the timer started in __init__()
gets to the end of its counting loop
'''
if not source:
return
if source == 'Event':
self.ev_timer_ev = True
elif source == 'Stat':
self._write_state()
else:
pass
def stop(self, signal, frame):
'''
We override stop() to brake our main loop
and have a pretty log message
'''
log.info("Received signal {0}".format(signal))
# if we have running workers, run through all and join() the ones
# that have finished. if we still have running workers after that,
# wait 5 secs for the rest and then exit. Maybe we should improv
# this a litte bit more
if len(self.running_workers) > 0:
clean_workers = []
for count in range(0, 2):
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("Joined worker #{0}".format(
worker.getName()))
if len(clean_workers) > 0:
log.info("Waiting 5secs for remaining workers..")
time.sleep(5)
else:
break
log.info("salt-eventsd has shut down")
# leave the cleanup to the supers stop
try:
super(SaltEventsDaemon, self).stop(signal, frame)
except (IOError, OSError):
os._exit(0)
def start(self):
'''
We override start() just for our log message
'''
log.info("Starting salt-eventsd daemon")
# leave the startup to the supers daemon, thats where all
# the daemonizing and double-forking takes place
super(SaltEventsDaemon, self).start()
def run(self):
'''
This method is automatically called by start() from
our parent class
'''
log.info("Initializing event-listener")
self.pid = self._get_pid()
self._write_state()
self.listen()
def _pre_startup(self, opts):
'''
Does a startup-check if all needed parameters are
found in the configfile. this is really important
because we lose stdout in daemon mode and exceptions
might not be seen by the user
'''
required_general = [
'sock_dir',
'node',
'max_workers',
'id',
'event_limit',
'pidfile',
'state_file',
'state_timer',
'dump_timer',
]
for field in required_general:
if field not in opts['general']:
log.critical("Missing parameter " +
"'{0}' in configfile".format(field))
sys.exit(1)
def listen(self):
'''
The main event loop where we receive the events and
start the workers that dump our data into the database
'''
# log on to saltstacks event-bus
event = salt.utils.event.SaltEvent(
self.node,
self.sock_dir,
)
# we store our events in a list, we dont really care about an order
# or what kind of data is put in there. all that is configured with the
# templates configured in the configfile
event_queue = []
# start our timers
self.ev_timer.start()
self.state_timer.start()
# this is for logline chronology so the timer-message always comes
# _before_ the actual startup-message of the listening loop below :-)
time.sleep(1)
log.info("Entering main event loop")
log.info("Listening on: {0}".format(event.puburi))
# read everything we can get our hands on
while True:
# the zmq-socket does not like ^C very much, make the error
# a little more graceful. alright, alright, ignore the damn thing,
# we're exiting anyways...
try:
ret = event.get_event(full=True)
except zmq.ZMQError:
pass
except KeyboardInterrupt:
log.info('Received CTRL+C, shutting down')
self.stop(signal.SIGTERM, None)
except AssertionError:
# Incase the master restarts a reconnect needs to happen.
event = salt.utils.event.SaltEvent(
self.node,
self.sock_dir,
)
ret = event.get_event(full=True)
# if we have not received enough events in to reach event_limit
# and the timer has fired, dump the events collected so far
# to the workers
if (self.ev_timer_ev):
if (len(self.running_workers) < self.max_workers) and \
(len(event_queue) > 0):
self._init_worker(event_queue)
# reset our queue to prevent duplicate entries
del event_queue[:]
# we reset the timer.ev_timer_ev at the end of the loop
# so we can update the stats that are logged
if ret is None:
continue
# filter only the events we're interested in. all events have a tag
# we can filter them by. we match with a precompiled regex
if 'tag' in ret:
# filter out events with an empty tag. those are special
if ret['tag'] != '':
# run through our configured events and try to match the
# current events tag against the ones we're interested in
for key in self.event_map.keys():
if self.event_map[key]['tag'].match(ret['tag']):
log.debug("Matching on {0}:{1}".format(
key, ret['tag']))
prio = self.event_map[key].get('prio', 0)
# push prio1-events directly into a worker
if prio > 0:
log.debug(
'Prio1 event found, pushing immediately!')
self.events_han += 1
self._init_worker([ret])
else:
event_queue.append(ret)
self.events_han += 1
# once we reach the event_limit, start a worker that
# writes that data in to the database
if len(event_queue) >= self.event_limit:
# only start a worker if not too many workers are running
if len(self.running_workers) < self.max_workers:
self._init_worker(event_queue)
# reset the timer
self.ev_timer.reset()
# reset our queue to prevent duplicate entries
del event_queue[:]
else:
# FIXME: we need to handle this situation somehow if
# too many workers are running. just flush the events?
# there really is no sane way except queueing more and more
# until some sort of limit is reached and we care more about
# our saltmaster than about the collected events!
log.critical("Too many workers running, loosing data!!!")
# a list for the workers that are still running
clean_workers = []
# run through all the workers and join() the ones
# that have finished dumping their data and keep
# the running ones on our list
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("Joined worker #{0}".format(worker.getName()))
self.threads_join += 1
# get rid of the old reference and set a new one
# FIXME: is this really neccessary?
del self.running_workers
self.running_workers = clean_workers
self.events_rec += 1
if (self.ev_timer_ev):
self.ev_timer_ev = False
log.info("Listen loop ended...")
def _get_pid(self):
'''
Get our current pid from the pidfile and fall back
to os.getpid() if pidfile not present (in foreground mode)
'''
pid = None
try:
pidf = file(self.pidfile, 'r')
pid = int(pidf.read().strip())
pidf.close()
except IOError:
pid = os.getpid()
return pid
def _write_state(self):
'''
Writes a current status to the defined status-file
this includes the current pid, events received/handled
and threads created/joined
'''
ev_hdl_per_s = float((float(self.events_han - self.stat_hdl_count)) /
float(self.state_timer_intrvl))
ev_tot_per_s = float((float(self.events_rec - self.stat_rec_count)) /
float(self.state_timer_intrvl))
if self.config['stat_worker']:
stat_data = {
'events_rec': self.events_rec,
'events_hdl': self.events_han,
'events_hdl_sec': round(ev_hdl_per_s, 2),
'events_tot_sec': round(ev_tot_per_s, 2),
'threads_created': self.threads_cre,
'threads_joined': self.threads_join
}
self.threads_cre += 1
st_worker = SaltEventsdWorker(stat_data, self.threads_cre, None,
self.backends, **self.opts)
st_worker.start()
try:
self.running_workers.append(st_worker)
except AttributeError:
log.error('self is missing running_workers')
try:
log.info(self)
log.info(dir(self))
except Exception:
log.error('Failed to dump dir(self)')
try:
# write the info to the specified log
statf = open(self.state_file, 'w')
statf.writelines(
json.dumps({
'events_rec': self.events_rec,
'events_hdl': self.events_han,
'events_hdl_sec': round(ev_hdl_per_s, 2),
'events_tot_sec': round(ev_tot_per_s, 2),
'threads_created': self.threads_cre,
'threads_joined': self.threads_join
}))
# if we have the same pid as the pidfile, we are the running daemon
# and also print the current counters to the logfile with 'info'
if os.getpid() == self.pid:
log.info("Running with pid {0}".format(self.pid))
log.info("Events (han/recv): {0}/{1}".format(
self.events_han,
self.events_rec,
))
log.info("Threads (cre/joi):{0}/{1}".format(
self.threads_cre,
self.threads_join,
))
statf.write("\n")
statf.close()
sys.stdout.flush()
except IOError as ioerr:
log.critical("Failed to write state to {0}".format(
self.state_file))
log.exception(ioerr)
except OSError as oserr:
log.critical("Failed to write state to {0}".format(
self.state_file))
log.exception(oserr)
self.stat_rec_count = self.events_rec
self.stat_hdl_count = self.events_han
def _init_backends(self, backends):
'''
Loads the backends from the workers dir defined in the config
and fails back to the codebase dir incase they're not found there.
'''
backend_dirs = [
os.path.dirname(os.path.realpath(__file__)) + '/workers/'
]
if 'backend_dir' in self.config:
backend_dirs.insert(0, self.config['backend_dir'])
backend_mngr = BackendMngr(backend_dirs)
return backend_mngr.load_plugins()
def _init_events(self, events={}):
'''
Creates a dict of precompiled regexes for all defined events
from config for maximum performance.
'''
self.event_map = events
# we precompile all regexes
log.info("Initialising events...")
for key in events.keys():
# we compile the regex configured in the config
self.event_map[key]['tag'] = compile(events[key]['tag'])
log.info("Added event '{0}'".format(key))
# if subevents are configured, also update them with
# regex-matching object
if 'subs' in events[key]:
for sub_ev in events[key]['subs'].keys():
try:
self.event_map[key]['subs'][sub_ev]['fun'] = compile(
events[key]['subs'][sub_ev]['fun'])
except KeyError:
pass
try:
self.event_map[key]['subs'][sub_ev]['tag'] = compile(
events[key]['subs'][sub_ev]['tag'])
except KeyError:
pass
log.info("Added sub-event '{0}->{1}'".format(key, sub_ev))
def _init_worker(self, qdata):
'''
The method dumps the data into a worker thread which
handles pushing the data into different backends.
'''
self.threads_cre += 1
log.info("Starting worker #{0}".format(self.threads_cre))
# make sure we pass a copy of the list
worker = SaltEventsdWorker(list(qdata), self.threads_cre,
self.event_map, self.backends, **self.opts)
worker.start()
self.running_workers.append(worker)
class SaltEventsDaemon(salteventsd.daemon.Daemon):
'''
The main daemon class where all the parsing, collecting
and dumping takes place
'''
def __init__(self):
self.opts = SaltEventsdLoader().getopts()
self._pre_startup(self.opts)
if type(self.opts) is not dict:
log.info("Received invalid configdata, startup cancelled")
sys.exit(1)
self.config = self.opts['general']
super(SaltEventsDaemon, self).__init__(self.config['pidfile'])
# the map of events are stored here, loaded in _init_events()
self.event_map = None
self._init_events(self.opts['events'])
self.backends = self._init_backends( self.config['backends'] )
log.info(self.backends)
# the socket to listen on for the events
self.sock_dir = self.config['sock_dir']
# two possible values for 'node': master and minion
# they do the same thing, just on different sockets
self.node = self.config['node']
# the id, usually 'master'
self.nodeid = self.config['id']
# the statefile where we write the daemon status
self.state_file = self.config['state_file']
# how many events to handle before updating the status
self.state_upd = self.config['state_upd']
# we dont know our pid (yet)
self.pid = None
# how many parallel workers to start max
self.max_workers = self.config['max_workers']
# the number of events to collect before starting a worker
self.event_limit = self.config['event_limit']
# a list to keep track of the currently running workers
# this is mainly for debugging to check wether all started
# workers are correctly joined over time so we dont leak memory
self.running_workers = []
# setup some counters used for the status
self.events_han = 0
self.events_rec = 0
self.threads_cre = 0
self.threads_join = 0
# the timer thats write data to the database every x seconds
# this is used to push data into the database even if
# self.event_limit is not reached regularly
self.ev_timer_ev = False
self.ev_timer_intrvl = self.config['dump_timer']
self.ev_timer = ResetTimer(self.ev_timer_intrvl,
self)
def timer_event(self):
'''
this is called whenever the timer started in __init__()
gets to the end of its counting loop
'''
self.ev_timer_ev = True
def stop(self,
signal,
frame):
'''
we override stop() to brake our main loop
and have a pretty log message
'''
log.info("received signal {0}".format(signal))
# if we have running workers, run through all and join() the ones
# that have finished. if we still have running workers after that,
# wait 5 secs for the rest and then exit. Maybe we should improv
# this a litte bit more
if( len(self.running_workers) > 0 ):
clean_workers = []
for count in range(0, 2):
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("joined worker #{0}".format(worker.getName()))
if( len(clean_workers) > 0 ):
log.info("waiting 5secs for remaining workers..")
time.sleep(5)
else:
break
log.info("salt-eventsd has shut down")
# leave the cleanup to the supers stop
super(SaltEventsDaemon, self).stop(signal, frame)
def start(self):
'''
we override start() just for our log message
'''
log.info("starting salt-eventsd daemon")
# leave the startup to the supers daemon, thats where all
# the daemonizing and double-forking takes place
super(SaltEventsDaemon, self).start()
def run(self):
'''
the method automatically called by start() from
our parent class
'''
log.info("initializing event listener")
self.pid = self._get_pid()
self._write_state()
self.listen()
def _pre_startup(self,
opts):
'''
does a startup-check if all needed parameters are
found in the configfile. this is really important
because we lose stdout in daemon mode and exceptions
might not be seen by the user
'''
required_general = [ 'sock_dir',
'node',
'max_workers',
'id',
'event_limit',
'pidfile',
'state_file',
'state_upd',
'dump_timer' ]
for field in required_general:
if field not in opts['general']:
log.critical("Missing parameter " +
"'{0}' in configfile".format(field))
sys.exit(1)
def listen(self):
'''
the main event loop where we receive the events and
start the workers that dump our data into the database
'''
# log on to saltstacks event-bus
event = salt.utils.event.SaltEvent(self.node,
self.sock_dir)
# we store our events in a list, we dont really care about an order
# or what kind of data is put in there. all that is configured with the
# sql-template configured in the configfile
event_queue = []
# start our dump_timer
self.ev_timer.start()
# this is for logline chronology so the timer-message always comes
# _before_ the actual startup-message of the listening loop below :-)
time.sleep(1)
log.info("entering main event loop")
log.info("listening on: {0}".format(event.puburi))
# read everything we can get our hands on
while True:
# the zmq-socket does not like ^C very much, make the error
# a little more graceful. alright, alright, ignore the damn thing,
# we're exiting anyways...
try:
ret = event.get_event(full=True)
except zmq.ZMQError:
pass
if ret is None:
continue
# if the timer has expired, we may have not received enough
# events in the queue to reach event_limit, in that case we dump
# the data anyway to have it in the database
if(self.ev_timer_ev):
if (len(self.running_workers) < self.max_workers) and \
(len(event_queue) > 0):
self._init_worker(event_queue)
# reset our queue to prevent duplicate entries
del event_queue[:]
# we reset the timer.ev_timer_ev at the end of the loop
# so we can update the stats that are logged
# filter only the events we're interested in. all events have a tag
# we can filter them by. we match with a precompiled regex
if( 'tag' in ret ):
# filter out events with an empty tag. those are special
if( ret['tag'] != '' ):
# run through our configured events and try to match the
# current events tag against the ones we're interested in
for key in self.event_map.keys():
if( self.event_map[key]['tag'].match(ret['tag'])):
log.debug("matching on {0}:{1}".format(key,
ret['tag']))
prio = self.event_map[key].get('prio', 0)
# push prio1-events directly into a worker
if prio > 0:
log.debug('Prio1 event found, pushing immediately!')
self.events_han += 1
self._init_worker([ret])
else:
event_queue.append(ret)
self.events_han += 1
# once we reach the event_limit, start a worker that
# writes that data in to the database
if len(event_queue) >= self.event_limit:
# only start a worker if not too many workers are running
if len(self.running_workers) < self.max_workers:
self._init_worker(event_queue)
# reset the timer
self.ev_timer.reset()
# reset our queue to prevent duplicate entries
del event_queue[:]
else:
# FIXME: we need to handle this situation somehow if
# too many workers are running. just flush the events?
# there really is no sane way except queueing more and more
# until some sort of limit is reached and we care more about
# our saltmaster than about the collected events!
log.critical("too many workers running, loosing data!!!")
# a list for the workers that are still running
clean_workers = []
# run through all the workers and join() the ones
# that have finished dumping their data and keep
# the running ones on our list
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("joined worker #{0}".format(worker.getName()))
self.threads_join += 1
# get rid of the old reference and set a new one
# FIXME: is this really neccessary?
del self.running_workers
self.running_workers = clean_workers
self.events_rec += 1
# we update the stats every 'received div handled == 0'
# or if we recevied a timer event from our ResetTimer
if( (self.events_rec % self.state_upd) == 0 ):
self._write_state()
elif(self.ev_timer_ev):
self._write_state()
self.ev_timer_ev = False
log.info("listen loop ended...")
def _get_pid(self):
'''
get our current pid from the pidfile, basically the
same as os.getpid()
'''
try:
pidf = file(self.pidfile,'r')
pid = int(pidf.read().strip())
pidf.close()
return pid
except IOError:
return None
def _write_state(self):
'''
writes a current status to the defined status-file
this includes the current pid, events received/handled
and threads created/joined
'''
try:
# write the info to the specified log
statf = open(self.state_file, 'w')
statf.writelines(simplejson.dumps({'events_received':self.events_rec,
'events_handled':self.events_han,
'threads_created':self.threads_cre,
'threads_joined':self.threads_join}
))
# if we have the same pid as the pidfile, we are the running daemon
# and also print the current counters to the logfile with 'info'
if( os.getpid() == self.pid ):
log.info("running with pid {0}".format(self.pid))
log.info("events (han/recv): {0}/{1}".format(self.events_han,
self.events_rec))
log.info("threads (cre/joi):{0}/{1}".format(self.threads_cre,
self.threads_join))
statf.write("\n")
statf.close()
sys.stdout.flush()
except IOError as ioerr:
log.critical("Failed to write state to {0}".format(self.state_file))
log.exception(ioerr)
except OSError as oserr:
log.critical("Failed to write state to {0}".format(self.state_file))
log.exception(oserr)
def _init_backends(self, backends):
backend_mngr = BackendMngr( ['/usr/share/pyshared/salteventsd/',
self.config['backend_dir'] ] )
return backend_mngr.load_plugins()
def _init_events(self, events={}):
'''
this is used to tell the class about the events it should handle.
it has to be a dictionary with appropriate mappings in it. see the
config file for examples on how to compose the dict. each entry is
converted to a precompiled regex for maximum flexibility
'''
self.event_map = events
# we precompile all regexes
log.info("initialising events...")
for key in events.keys():
# we compile the regex configured in the config
self.event_map[key]['tag'] = compile( events[key]['tag'] )
log.info("Added event '{0}'".format(key))
# if subevents are configured, also update them with
# regex-macthing object
if( events[key].has_key('subs') ):
for sub_ev in events[key]['subs'].keys():
try:
self.event_map[key]['subs'][sub_ev]['fun'] = compile(events[key]['subs'][sub_ev]['fun'])
except KeyError:
pass
try:
self.event_map[key]['subs'][sub_ev]['tag'] = compile(events[key]['subs'][sub_ev]['tag'])
except KeyError:
pass
log.info("Added sub-event '{0}->{1}'".format(key,
sub_ev))
# the method dumps the data into a worker thread which
# handles pushing the data into different backends
def _init_worker(self, qdata):
'''
write a collection of events to the database. every invocation of
this methoed creates its own thread that writes into the database
'''
self.threads_cre += 1
log.info("starting worker #{0}".format(self.threads_cre))
# make sure we pass a copy of the list
worker = SaltEventsdWorker(list(qdata),
self.threads_cre,
self.event_map,
self.backends,
**self.opts)
worker.start()
self.running_workers.append(worker)
class SaltEventsDaemon(Daemon):
'''
The main daemon class where all the parsing, collecting
and dumping takes place
'''
def __init__(self, config, log_level=None, log_file=None, daemonize=False):
self.opts = SaltEventsdLoader(
config=config,
log_level=log_level,
log_file=log_file,
daemonize=daemonize,
).getopts()
self._pre_startup(self.opts)
if type(self.opts) is not dict:
log.info("Received invalid configdata, startup cancelled")
sys.exit(1)
self.config = self.opts['general']
super(SaltEventsDaemon, self).__init__(self.config['pidfile'])
# the map of events are stored here, loaded in _init_events()
self.event_map = None
self._init_events(self.opts['events'])
self.backends = self._init_backends(self.config['backends'])
# the socket to listen on for the events
self.sock_dir = self.config['sock_dir']
# two possible values for 'node': master and minion
# they do the same thing, just on different sockets
self.node = self.config['node']
# the id, usually 'master'
self.nodeid = self.config['id']
# the statefile where we write the daemon status
self.state_file = self.config['state_file']
# how many events to handle before updating the status
self.state_timer = self.config['state_timer']
# we dont know our pid (yet), its updated in run()
self.pid = None
# how many parallel workers to start max
self.max_workers = self.config['max_workers']
# the number of events to collect before starting a worker
self.event_limit = self.config['event_limit']
# a list to keep track of the currently running workers
# this is mainly for debugging to check wether all started
# workers are correctly joined over time so we dont leak memory
self.running_workers = []
# setup some counters used for the status
# the events we matched on
self.events_han = 0
self.events_rec = 0
# the threads we have created and joined
# if there numbers diverge, we have a memory leak
self.threads_cre = 0
self.threads_join = 0
# used to keep track of the delta between two stat_timer intervals
# to calculate events per second handled/received
self.stat_rec_count = 0
self.stat_hdl_count = 0
# the timer thats write data to the database every x seconds
# this is used to push data into the database even if
# self.event_limit is not reached regularly
self.ev_timer_ev = False
self.ev_timer_intrvl = self.config['dump_timer']
self.ev_timer = ResetTimer(
self.ev_timer_intrvl,
self,
name='Event'
)
# the timer that writes statistical data to the status-file
self.state_timer_ev = False
self.state_timer_intrvl = self.config['state_timer']
self.state_timer = ResetTimer(
self.state_timer_intrvl,
self,
name='Stat'
)
def timer_event(self, source=None):
'''
This is called whenever the timer started in __init__()
gets to the end of its counting loop
'''
if not source:
return
if source == 'Event':
self.ev_timer_ev = True
elif source == 'Stat':
self._write_state()
else:
pass
def stop(self, signal, frame):
'''
We override stop() to brake our main loop
and have a pretty log message
'''
log.info("Received signal {0}".format(signal))
# if we have running workers, run through all and join() the ones
# that have finished. if we still have running workers after that,
# wait 5 secs for the rest and then exit. Maybe we should improv
# this a litte bit more
if len(self.running_workers) > 0:
clean_workers = []
for count in range(0, 2):
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("Joined worker #{0}".format(worker.getName()))
if len(clean_workers) > 0:
log.info("Waiting 5secs for remaining workers..")
time.sleep(5)
else:
break
log.info("salt-eventsd has shut down")
# leave the cleanup to the supers stop
try:
super(SaltEventsDaemon, self).stop(signal, frame)
except (IOError, OSError):
os._exit(0)
def start(self):
'''
We override start() just for our log message
'''
log.info("Starting salt-eventsd daemon")
# leave the startup to the supers daemon, thats where all
# the daemonizing and double-forking takes place
super(SaltEventsDaemon, self).start()
def run(self):
'''
This method is automatically called by start() from
our parent class
'''
log.info("Initializing event-listener")
self.pid = self._get_pid()
self._write_state()
self.listen()
def _pre_startup(self, opts):
'''
Does a startup-check if all needed parameters are
found in the configfile. this is really important
because we lose stdout in daemon mode and exceptions
might not be seen by the user
'''
required_general = [
'sock_dir',
'node',
'max_workers',
'id',
'event_limit',
'pidfile',
'state_file',
'state_timer',
'dump_timer',
]
for field in required_general:
if field not in opts['general']:
log.critical("Missing parameter " +
"'{0}' in configfile".format(field))
sys.exit(1)
def listen(self):
'''
The main event loop where we receive the events and
start the workers that dump our data into the database
'''
# log on to saltstacks event-bus
event = salt.utils.event.SaltEvent(
self.node,
self.sock_dir,
)
# we store our events in a list, we dont really care about an order
# or what kind of data is put in there. all that is configured with the
# templates configured in the configfile
event_queue = []
# start our timers
self.ev_timer.start()
self.state_timer.start()
# this is for logline chronology so the timer-message always comes
# _before_ the actual startup-message of the listening loop below :-)
time.sleep(1)
log.info("Entering main event loop")
log.info("Listening on: {0}".format(event.puburi))
# read everything we can get our hands on
while True:
# the zmq-socket does not like ^C very much, make the error
# a little more graceful. alright, alright, ignore the damn thing,
# we're exiting anyways...
try:
ret = event.get_event(full=True)
except zmq.ZMQError:
pass
except KeyboardInterrupt:
log.info('Received CTRL+C, shutting down')
self.stop(signal.SIGTERM, None)
except AssertionError:
# Incase the master restarts a reconnect needs to happen.
event = salt.utils.event.SaltEvent(
self.node,
self.sock_dir,
)
ret = event.get_event(full=True)
# if we have not received enough events in to reach event_limit
# and the timer has fired, dump the events collected so far
# to the workers
if(self.ev_timer_ev):
if (len(self.running_workers) < self.max_workers) and \
(len(event_queue) > 0):
self._init_worker(event_queue)
# reset our queue to prevent duplicate entries
del event_queue[:]
# we reset the timer.ev_timer_ev at the end of the loop
# so we can update the stats that are logged
if ret is None:
continue
# filter only the events we're interested in. all events have a tag
# we can filter them by. we match with a precompiled regex
if 'tag' in ret:
# filter out events with an empty tag. those are special
if ret['tag'] != '':
# run through our configured events and try to match the
# current events tag against the ones we're interested in
for key in self.event_map.keys():
if self.event_map[key]['tag'].match(ret['tag']):
log.debug("Matching on {0}:{1}".format(key, ret['tag']))
prio = self.event_map[key].get('prio', 0)
# push prio1-events directly into a worker
if prio > 0:
log.debug('Prio1 event found, pushing immediately!')
self.events_han += 1
self._init_worker([ret])
else:
event_queue.append(ret)
self.events_han += 1
# once we reach the event_limit, start a worker that
# writes that data in to the database
if len(event_queue) >= self.event_limit:
# only start a worker if not too many workers are running
if len(self.running_workers) < self.max_workers:
self._init_worker(event_queue)
# reset the timer
self.ev_timer.reset()
# reset our queue to prevent duplicate entries
del event_queue[:]
else:
# FIXME: we need to handle this situation somehow if
# too many workers are running. just flush the events?
# there really is no sane way except queueing more and more
# until some sort of limit is reached and we care more about
# our saltmaster than about the collected events!
log.critical("Too many workers running, loosing data!!!")
# a list for the workers that are still running
clean_workers = []
# run through all the workers and join() the ones
# that have finished dumping their data and keep
# the running ones on our list
for worker in self.running_workers:
if worker.isAlive():
clean_workers.append(worker)
else:
worker.join()
log.debug("Joined worker #{0}".format(worker.getName()))
self.threads_join += 1
# get rid of the old reference and set a new one
# FIXME: is this really neccessary?
del self.running_workers
self.running_workers = clean_workers
self.events_rec += 1
if(self.ev_timer_ev):
self.ev_timer_ev = False
log.info("Listen loop ended...")
def _get_pid(self):
'''
Get our current pid from the pidfile and fall back
to os.getpid() if pidfile not present (in foreground mode)
'''
pid = None
try:
pidf = file(self.pidfile, 'r')
pid = int(pidf.read().strip())
pidf.close()
except IOError:
pid = os.getpid()
return pid
def _write_state(self):
'''
Writes a current status to the defined status-file
this includes the current pid, events received/handled
and threads created/joined
'''
ev_hdl_per_s = float((float(self.events_han - self.stat_hdl_count)) / float(self.state_timer_intrvl))
ev_tot_per_s = float((float(self.events_rec - self.stat_rec_count)) / float(self.state_timer_intrvl))
if self.config['stat_worker']:
stat_data = {
'events_rec': self.events_rec,
'events_hdl': self.events_han,
'events_hdl_sec': round(ev_hdl_per_s, 2),
'events_tot_sec': round(ev_tot_per_s, 2),
'threads_created': self.threads_cre,
'threads_joined': self.threads_join
}
self.threads_cre += 1
st_worker = SaltEventsdWorker(
stat_data,
self.threads_cre,
None,
self.backends,
**self.opts
)
st_worker.start()
try:
self.running_workers.append(st_worker)
except AttributeError:
log.error('self is missing running_workers')
try:
log.info(self)
log.info(dir(self))
except Exception:
log.error('Failed to dump dir(self)')
try:
# write the info to the specified log
statf = open(self.state_file, 'w')
statf.writelines(
json.dumps({
'events_rec': self.events_rec,
'events_hdl': self.events_han,
'events_hdl_sec': round(ev_hdl_per_s, 2),
'events_tot_sec': round(ev_tot_per_s, 2),
'threads_created': self.threads_cre,
'threads_joined': self.threads_join
})
)
# if we have the same pid as the pidfile, we are the running daemon
# and also print the current counters to the logfile with 'info'
if os.getpid() == self.pid:
log.info("Running with pid {0}".format(self.pid))
log.info("Events (han/recv): {0}/{1}".format(
self.events_han,
self.events_rec,
))
log.info("Threads (cre/joi):{0}/{1}".format(
self.threads_cre,
self.threads_join,
))
statf.write("\n")
statf.close()
sys.stdout.flush()
except IOError as ioerr:
log.critical("Failed to write state to {0}".format(self.state_file))
log.exception(ioerr)
except OSError as oserr:
log.critical("Failed to write state to {0}".format(self.state_file))
log.exception(oserr)
self.stat_rec_count = self.events_rec
self.stat_hdl_count = self.events_han
def _init_backends(self, backends):
'''
Loads the backends from the workers dir defined in the config
and fails back to the codebase dir incase they're not found there.
'''
backend_dirs = [
os.path.dirname(os.path.realpath(__file__)) + '/workers/'
]
if 'backend_dir' in self.config:
backend_dirs.insert(0, self.config['backend_dir'])
backend_mngr = BackendMngr(backend_dirs)
return backend_mngr.load_plugins()
def _init_events(self, events={}):
'''
Creates a dict of precompiled regexes for all defined events
from config for maximum performance.
'''
self.event_map = events
# we precompile all regexes
log.info("Initialising events...")
for key in events.keys():
# we compile the regex configured in the config
self.event_map[key]['tag'] = compile(events[key]['tag'])
log.info("Added event '{0}'".format(key))
# if subevents are configured, also update them with
# regex-matching object
if 'subs' in events[key]:
for sub_ev in events[key]['subs'].keys():
try:
self.event_map[key]['subs'][sub_ev]['fun'] = compile(events[key]['subs'][sub_ev]['fun'])
except KeyError:
pass
try:
self.event_map[key]['subs'][sub_ev]['tag'] = compile(events[key]['subs'][sub_ev]['tag'])
except KeyError:
pass
log.info("Added sub-event '{0}->{1}'".format(key, sub_ev))
def _init_worker(self, qdata):
'''
The method dumps the data into a worker thread which
handles pushing the data into different backends.
'''
self.threads_cre += 1
log.info("Starting worker #{0}".format(self.threads_cre))
# make sure we pass a copy of the list
worker = SaltEventsdWorker(
list(qdata),
self.threads_cre,
self.event_map,
self.backends,
**self.opts
)
worker.start()
self.running_workers.append(worker)