def make_beat_msg(pm, CFG, log=logging):
beat_msg = {}
beat_msg['eeagent_id'] = CFG.eeagent.name
# include node ID if it is present in config
node_id = CFG.eeagent.get('node_id')
if node_id is not None:
beat_msg['node_id'] = CFG.eeagent.node_id
# include timestamp in UTC, in iso8601/rfc3339
beat_msg['timestamp'] = datetime.now(utc).isoformat()
beat_processes = []
# we can have many process managers per eeagent, walk them all to get all the processes
pm.poll()
processes = pm.get_all()
for pname in processes:
p = processes[pname]
(name, round) = unmake_id(p.get_name())
try:
state = p.get_state()
except:
log.exception("Had a problem getting process state")
raise
beat_p = {'upid': name, 'round': round, 'state': state, 'msg': p.get_error_message()}
beat_processes.append(beat_p)
beat_msg['processes'] = beat_processes
return beat_msg
def lookup_id(self, process_name, ignore_round=False):
if ignore_round:
process_upid, process_round = unmake_id(process_name)
for name, proc in self._known_pws.iteritems():
upid, round = unmake_id(name)
if process_upid == upid:
return proc
else:
return None
else:
if process_name not in self._known_pws:
return None
return self._known_pws[process_name]
def make_beat_msg(pm, CFG, log=logging):
beat_msg = {}
beat_msg['eeagent_id'] = CFG.eeagent.name
# include node ID if it is present in config
node_id = CFG.eeagent.get('node_id')
if node_id is not None:
beat_msg['node_id'] = CFG.eeagent.node_id
# include timestamp in UTC, in iso8601/rfc3339
beat_msg['timestamp'] = datetime.now(utc).isoformat()
beat_processes = []
# we can have many process managers per eeagent, walk them all to get all the processes
pm.poll()
processes = pm.get_all()
for pname in processes:
p = processes[pname]
(name, round) = unmake_id(p.get_name())
try:
state = p.get_state()
except:
log.exception("Had a problem getting process state")
raise
beat_p = {
'upid': name,
'round': round,
'state': state,
'msg': p.get_error_message()
}
beat_processes.append(beat_p)
beat_msg['processes'] = beat_processes
return beat_msg
def lookup_id(self, process_name, ignore_round=False):
if ignore_round:
process_upid, process_round = unmake_id(process_name)
for name, proc in self._known_pws.iteritems():
upid, round = unmake_id(name)
if process_upid == upid:
return proc
else:
return None
else:
if process_name not in self._known_pws:
return None
return self._known_pws[process_name]
def launch_process(self, u_pid, round, run_type, parameters):
if run_type != self.CFG.eeagent.launch_type.name:
raise EEAgentParameterException("Unknown run type %s" % (run_type))
# check to see if download_code enabled
if (parameters.get('module_uri') and
(not self.CFG.eeagent.get('code_download') or
not self.CFG.eeagent.code_download.get('enabled', False))):
msg = "Code download not enabled in this eeagent"
raise EEAgentUnauthorizedException(msg)
# check if url in whitelist
if parameters.get('module_uri'):
uri = parameters.get('module_uri')
if not self._check_whitelist(uri):
msg = "%s not in code_download whitelist: '%s'" % (uri,
", ".join(self._get_whitelist()))
raise EEAgentUnauthorizedException(msg)
# Check to see whether this process already exists
process = self._find_proc(u_pid, round, ignore_round=True)
if process is not None and not process.restartable:
existing_upid, existing_round = unmake_id(process._name)
state = str(process.get_state())
if int(round) > int(existing_round):
msg = ("VERY BAD THING: A launch request for '%s' has been "
"recieved, but it is in state %s. The existing process has "
"round %s, but the PD is asking to start process with round "
"%s. Restarting the process anyway, but something is wrong."
% (u_pid, state, existing_round, round))
self._log.error(msg)
else:
msg = ("BAD THING: A launch request for '%s' has been "
"recieved, but it is in state %s. The existing process has "
"round %s, and the PD is asking to start process with round "
"%s. Maybe a message never arrived? Leaving process as-is."
% (u_pid, state, existing_round, round))
self._log.warning(msg)
return
try:
process.terminate()
except PIDanticStateException, pse:
self._log.warning("Attempt to terminate a process in the state %s" % (str(process.get_state())))
try:
process.clean_up()
except Exception, ex:
self._log.warning("Failed to cleanup: %s" % (str(ex)))
def beat_it(dashi, CFG, pm, log=logging):
try:
beat_msg = {}
beat_msg['eeagent_id'] = ""
beat_msg['timestamp'] = str(datetime.datetime.now())
beat_processes = []
# we can have many process managers per eeagent, walk them all to get all the processes
pm.poll()
processes = pm.get_all()
for pname in processes:
p = processes[pname]
(name, round) = unmake_id(p.get_name())
beat_p = {'upid': name, 'round': round, 'state': p.get_state(), 'msg': p.get_error_message()}
beat_processes.append(beat_p)
beat_msg['processes'] = beat_processes
log.log(logging.DEBUG, "Sending the heartbeat : %s" % (json.dumps(beat_msg)))
dashi.fire(CFG.pd.name, "heartbeat", message=beat_msg)
except Exception, ex:
log.log(logging.ERROR, "Error Sending the heartbeat : %s" % (str(ex)))
