def _get_current_processes(self):
# Add all current processes to make sure nothing misses the radar
process_state_db = {}
# list of all processes on the node is made here
all_processes = self.process_info_manager.get_all_processes()
self.msg_log("DBG: get_current_processes: '%s'" % all_processes,
SandeshLevel.SYS_DEBUG)
for proc_info in all_processes:
proc_name = self._get_process_name(proc_info)
proc_pid = int(proc_info['pid'])
stat = ProcessStat(proc_name,
host_ip=self.hostip,
hostname=self.hostname)
stat.process_state = proc_info['statename']
if 'start' in proc_info:
stat.start_time = str(proc_info['start'])
stat.start_count += 1
stat.pid = proc_pid
if stat.group not in self.group_names:
self.group_names.append(stat.group)
if not stat.group in process_state_db:
process_state_db[stat.group] = {}
process_state_db[stat.group][proc_name] = stat
return process_state_db
def _send_process_state(self, process_info):
pname = self._get_process_name(process_info)
# update process stats
if pname in list(key for group in self.process_state_db
for key in self.process_state_db[group]):
for group in self.process_state_db:
if pname in self.process_state_db[group]:
proc_stat = self.process_state_db[group][pname]
else:
proc_stat = ProcessStat(pname,
host_ip=self.hostip,
hostname=self.hostname)
pstate = process_info['state']
proc_stat.process_state = pstate
send_uve = False
if (pstate == 'PROCESS_STATE_RUNNING'):
proc_stat.start_count += 1
proc_stat.start_time = str(int(time.time() * 1000000))
send_uve = True
proc_stat.pid = int(process_info['pid'])
if (pstate == 'PROCESS_STATE_STOPPED'):
proc_stat.stop_count += 1
send_uve = True
proc_stat.stop_time = str(int(time.time() * 1000000))
proc_stat.last_exit_unexpected = False
proc_stat.last_cpu = None
proc_stat.last_time = 0
if (pstate == 'PROCESS_STATE_EXITED'):
proc_stat.exit_count += 1
send_uve = True
proc_stat.exit_time = str(int(time.time() * 1000000))
proc_stat.last_cpu = None
proc_stat.last_time = 0
if not process_info['expected']:
self.msg_log(
'%s with pid: %s exited abnormally' %
(pname, process_info['pid']), SandeshLevel.SYS_ERR)
proc_stat.last_exit_unexpected = True
send_init_uve = False
# update process state database
if not proc_stat.group in self.process_state_db:
self.process_state_db[proc_stat.group] = {}
send_init_uve = True
self.process_state_db[proc_stat.group][pname] = proc_stat
if send_uve:
if (send_init_uve):
self._send_init_info(proc_stat.group)
self.send_process_state_db([proc_stat.group])
def _send_process_state(self, process_info):
pname = self._get_process_name(process_info)
# update process stats
if pname in list(key for group in self.process_state_db for key in self.process_state_db[group]):
for group in self.process_state_db:
if pname in self.process_state_db[group]:
proc_stat = self.process_state_db[group][pname]
else:
proc_stat = ProcessStat(pname, host_ip=self.hostip)
pstate = process_info['state']
proc_stat.process_state = pstate
send_uve = False
if (pstate == 'PROCESS_STATE_RUNNING'):
proc_stat.start_count += 1
proc_stat.start_time = str(int(time.time() * 1000000))
send_uve = True
proc_stat.pid = int(process_info['pid'])
if (pstate == 'PROCESS_STATE_STOPPED'):
proc_stat.stop_count += 1
send_uve = True
proc_stat.stop_time = str(int(time.time() * 1000000))
proc_stat.last_exit_unexpected = False
proc_stat.last_cpu = None
proc_stat.last_time = 0
if (pstate == 'PROCESS_STATE_EXITED'):
proc_stat.exit_count += 1
send_uve = True
proc_stat.exit_time = str(int(time.time() * 1000000))
proc_stat.last_cpu = None
proc_stat.last_time = 0
if not process_info['expected']:
self.msg_log('%s with pid: %s exited abnormally' %
(pname, process_info['pid']), SandeshLevel.SYS_ERR)
proc_stat.last_exit_unexpected = True
send_init_uve = False
# update process state database
if not proc_stat.group in self.process_state_db:
self.process_state_db[proc_stat.group] = {}
send_init_uve = True
self.process_state_db[proc_stat.group][pname] = proc_stat
if send_uve:
if (send_init_uve):
self._send_init_info(proc_stat.group)
self.send_process_state_db([proc_stat.group])
def _get_current_processes(self):
# Add all current processes to make sure nothing misses the radar
process_state_db = {}
# list of all processes on the node is made here
all_processes = self.process_info_manager.get_all_processes()
self.msg_log("DBG: get_current_processes: '%s'" % all_processes,
SandeshLevel.SYS_DEBUG)
for proc_info in all_processes:
proc_name = self._get_process_name(proc_info)
proc_pid = int(proc_info['pid'])
stat = ProcessStat(proc_name, host_ip=self.hostip)
stat.process_state = proc_info['statename']
if 'start' in proc_info:
stat.start_time = str(proc_info['start'])
stat.start_count += 1
stat.pid = proc_pid
if stat.group not in self.group_names:
self.group_names.append(stat.group)
if not stat.group in process_state_db:
process_state_db[stat.group] = {}
process_state_db[stat.group][proc_name] = stat
return process_state_db
def send_process_state(self, process_info):
pname = self.get_process_name(process_info)
# update process stats
if pname in list(key for group in self.process_state_db
for key in self.process_state_db[group]):
for group in self.process_state_db:
if pname in self.process_state_db[group]:
proc_stat = self.process_state_db[group][pname]
else:
proc_stat = ProcessStat(pname)
pstate = process_info['state']
proc_stat.process_state = pstate
send_uve = False
if (pstate == 'PROCESS_STATE_RUNNING'):
proc_stat.start_count += 1
proc_stat.start_time = str(int(time.time() * 1000000))
send_uve = True
proc_stat.pid = int(process_info['pid'])
if (pstate == 'PROCESS_STATE_STOPPED'):
proc_stat.stop_count += 1
send_uve = True
proc_stat.stop_time = str(int(time.time() * 1000000))
proc_stat.last_exit_unexpected = False
proc_stat.last_cpu = None
proc_stat.last_time = 0
if (pstate == 'PROCESS_STATE_EXITED'):
proc_stat.exit_count += 1
send_uve = True
proc_stat.exit_time = str(int(time.time() * 1000000))
proc_stat.last_cpu = None
proc_stat.last_time = 0
if not process_info['expected']:
self.msg_log(
'%s with pid: %s exited abnormally' %
(pname, process_info['pid']), SandeshLevel.SYS_ERR)
proc_stat.last_exit_unexpected = True
# check for core file for this exit
find_command_option = ("find " + self.get_corefile_path() +
" -name core.[A-Za-z]*." +
process_info['pid'] + "*")
self.msg_log(
'find command option for cores: %s' %
(find_command_option), SandeshLevel.SYS_DEBUG)
(corename, _) = Popen(find_command_option.split(),
stdout=PIPE,
close_fds=True).communicate()
if ((corename is not None) and (len(corename.rstrip()) >= 1)):
self.msg_log('core file: %s' % (corename),
SandeshLevel.SYS_ERR)
# before adding to the core file list make
# sure that we do not have too many cores
self.msg_log(
'core_file_list: %s, max_cores: %d' %
(str(proc_stat.core_file_list), self.max_cores),
SandeshLevel.SYS_DEBUG)
if (len(proc_stat.core_file_list) >= self.max_cores):
# get rid of old cores
start = self.max_old_cores
end = len(proc_stat.core_file_list) - \
self.max_new_cores + 1
core_files_to_be_deleted = \
proc_stat.core_file_list[start:end]
self.remove_corefiles(core_files_to_be_deleted)
# now delete the cores from the list as well
del proc_stat.core_file_list[start:end]
# now add the new core to the core file list
proc_stat.core_file_list.append(corename.rstrip())
self.msg_log(
'# of cores for %s: %d' %
(pname, len(proc_stat.core_file_list)),
SandeshLevel.SYS_DEBUG)
send_init_uve = False
# update process state database
if not proc_stat.group in self.process_state_db:
self.process_state_db[proc_stat.group] = {}
send_init_uve = True
self.process_state_db[proc_stat.group][pname] = proc_stat
if send_uve:
if (send_init_uve):
self.send_init_info(proc_stat.group)
self.send_process_state_db([proc_stat.group])
def _send_process_state(self, process_info):
pname = self._get_process_name(process_info)
# update process stats
if pname in list(key for group in self.process_state_db for key in self.process_state_db[group]):
for group in self.process_state_db:
if pname in self.process_state_db[group]:
proc_stat = self.process_state_db[group][pname]
else:
proc_stat = ProcessStat(pname, host_ip=self.hostip)
pstate = process_info['state']
proc_stat.process_state = pstate
send_uve = False
if (pstate == 'PROCESS_STATE_RUNNING'):
proc_stat.start_count += 1
proc_stat.start_time = str(int(time.time() * 1000000))
send_uve = True
proc_stat.pid = int(process_info['pid'])
if (pstate == 'PROCESS_STATE_STOPPED'):
proc_stat.stop_count += 1
send_uve = True
proc_stat.stop_time = str(int(time.time() * 1000000))
proc_stat.last_exit_unexpected = False
proc_stat.last_cpu = None
proc_stat.last_time = 0
if (pstate == 'PROCESS_STATE_EXITED'):
proc_stat.exit_count += 1
send_uve = True
proc_stat.exit_time = str(int(time.time() * 1000000))
proc_stat.last_cpu = None
proc_stat.last_time = 0
if not process_info['expected']:
self.msg_log('%s with pid: %s exited abnormally' %
(pname, process_info['pid']), SandeshLevel.SYS_ERR)
proc_stat.last_exit_unexpected = True
# check for core file for this exit
corename = self.system_data.find_corefile(
"core.[A-Za-z]*." + process_info['pid'] + "*")
if ((corename is not None) and (len(corename.rstrip()) >= 1)):
self.msg_log('core file: %s' % (corename),
SandeshLevel.SYS_ERR)
# before adding to the core file list make
# sure that we do not have too many cores
self.msg_log('core_file_list: %s, max_cores: %d' %
(str(proc_stat.core_file_list), self.max_cores),
SandeshLevel.SYS_DEBUG)
if (len(proc_stat.core_file_list) >= self.max_cores):
# get rid of old cores
start = self.max_old_cores
end = len(proc_stat.core_file_list) - \
self.max_new_cores + 1
core_files_to_be_deleted = \
proc_stat.core_file_list[start:end]
self.system_data.remove_corefiles(core_files_to_be_deleted)
# now delete the cores from the list as well
del proc_stat.core_file_list[start:end]
# now add the new core to the core file list
proc_stat.core_file_list.append(corename.rstrip())
self.msg_log('# of cores for %s: %d' % (pname,
len(proc_stat.core_file_list)), SandeshLevel.SYS_DEBUG)
send_init_uve = False
# update process state database
if not proc_stat.group in self.process_state_db:
self.process_state_db[proc_stat.group] = {}
send_init_uve = True
self.process_state_db[proc_stat.group][pname] = proc_stat
if send_uve:
if (send_init_uve):
self._send_init_info(proc_stat.group)
self.send_process_state_db([proc_stat.group])