def discovery_redis(*args):
"""
find redis instance
@return: [(ip, prot, passwd)]
"""
redises = []
redis_conf_path_root = '/data'
redis_conf_file_name = 'redis.conf'
for redis_process in [x
for x in psutil.process_iter()
if len(x.cmdline()) > 0 and (
os.path.basename(x.exe()) == BINNAME or os.path.basename(x.exe()) == CODISBINNAME)]:
try:
redis_ip, redis_port = sorted([laddr.laddr
for laddr in redis_process.connections()
if laddr.status == 'LISTEN'])[0]
except:
continue
redis_passwd = ''
config_files = []
if os.path.isdir(redis_process.cwd()):
cwd_try = os.path.join(redis_process.cwd(), redis_conf_file_name)
if os.path.exists(cwd_try):
config_files.append(cwd_try)
else:
for root_dir, dirs, files in os.walk(os.path.dirname(redis_process.cwd())):
if 'redis.conf' in files:
config_files.append(os.path.join(root_dir, redis_conf_file_name))
elif len(redis_process.cmdline()) > 1 and os.path.isfile(redis_process.cmdline()[1]):
config_files.append(redis_process.cmdline()[1])
else:
for root_dir, dirs, files in os.walk(redis_conf_path_root):
if 'redis.conf' in files:
config_files.append(os.path.join(root_dir, redis_conf_file_name))
for config_file in config_files:
try:
with open(config_file, 'r') as f:
passwd = None
port = None
for line in f.readlines():
if re.search('^requirepass', line):
passwd = line.split()[1]
if re.search('^port', line):
port = line.split()[1]
if passwd and port and str(redis_port) == port:
redis_passwd = Monitor.encode_password(passwd)
break
except:
pass
if redis_ip == '0.0.0.0' or redis_ip == '::' or redis_ip == '127.0.0.1' or redis_ip == '':
redis_ip = Monitor.get_local_ip()
if [redis_ip, redis_port, redis_passwd] not in redises:
redises.append([redis_ip, redis_port, redis_passwd])
return redises
def discovery_redis():
"""
find redis instance
@return: [(ip, prot, passwd)]
"""
redises = []
redis_conf_path_root = "/data"
redis_conf_file_name = "redis.conf"
for redis_process in [
x for x in psutil.process_iter() if len(x.cmdline()) > 0 and os.path.basename(x.exe()) == BINNAME
]:
try:
redis_ip, redis_port = sorted(
[laddr.laddr for laddr in redis_process.get_connections() if laddr.status == "LISTEN"]
)[0]
except:
continue
redis_passwd = ""
config_files = []
if os.path.isdir(redis_process.getcwd()):
cwd_try = os.path.join(redis_process.getcwd(), redis_conf_file_name)
if os.path.exists(cwd_try):
config_files.append(cwd_try)
else:
for root_dir, dirs, files in os.walk(os.path.dirname(redis_process.getcwd())):
if "redis.conf" in files:
config_files.append(os.path.join(root_dir, redis_conf_file_name))
elif len(redis_process.cmdline()) > 1 and os.path.isfile(redis_process.cmdline()[1]):
config_files.append(redis_process.cmdline()[1])
else:
for root_dir, dirs, files in os.walk(redis_conf_path_root):
if "redis.conf" in files:
config_files.append(os.path.join(root_dir, redis_conf_file_name))
for config_file in config_files:
try:
with open(config_file, "r") as f:
passwd = None
port = None
for line in f.readlines():
if re.search("^requirepass", line):
passwd = line.split()[1]
if re.search("^port", line):
port = line.split()[1]
if passwd and port and str(redis_port) == port:
redis_passwd = Monitor.encode_password(passwd)
break
except:
pass
if redis_ip == "0.0.0.0":
redis_ip = Monitor.get_local_ip()
redises.append([redis_ip, redis_port, redis_passwd])
return redises
def start(args):
"""
get service monitor data, or discovery service instance
@param args:
@return: string when get service data, json when discovery
"""
fs = args.fieldseparator
arg_list = []
if args.extend:
arg_list = args.extend.split(fs)
if args.discovery:
assert not args.macros is None, 'must have macros'
elif args.list:
print Monitor.get_service_list(fs, *arg_list)
else:
assert not args.instance is None, 'must have instance'
# assert not args.item is None, 'must have item'
if args.service:
monitor = Monitor(args.service, cache_path=args.cache if args.cache else None)
if args.discovery:
print monitor.discovery(args.macros.split(fs), *arg_list)
else:
if args.item:
print monitor.load_data(args.instance, args.item, *arg_list)
if args.list:
print "Monitor Items (in %s)" % args.instance
for it in sorted(monitor.load_keys(args.instance, *arg_list)):
print it
def discovery_tomcat(*args):
'''
discovery tomcat instance's host and port
'''
result = []
port = ''
cmdline = []
host = Monitor.get_local_ip()
for i in psutil.process_iter():
if i.name() == BINNAME:
pid = int(i.pid)
cmdline = i.cmdline()
for opt in cmdline:
if opt.find('-Dcom.sun.management.jmxremote.port') == 0:
monitor_port = opt.strip().split('=')[1]
if opt.find('-Dcatalina.base') == 0:
config_file = opt.strip().split(
'=')[1] + '/conf/server.xml'
tree = ET.ElementTree(file=config_file)
root = tree.getroot()
for root_child in root:
if root_child.tag == 'Service':
for service_child in root_child:
if service_child.tag == 'Connector':
listen_port = service_child.attrib['port']
if monitor_port.isdigit() and listen_port.isdigit():
result.append([host, listen_port, monitor_port])
return result
def discovery_kingshard(*args):
'''
discovery kingshard instance's host and port
'''
result = []
port = ''
address = ''
cmdline = []
host = Monitor.get_local_ip()
for i in psutil.process_iter():
if i.name() == BINNAME:
pid = int(i.pid)
cmdline = i.cmdline()
for opt in cmdline:
if opt.find('-config=') == 0:
config_file = opt.strip().split('=')[1]
if os.path.exists(config_file):
f = open(config_file, 'r')
cfg_content = f.read()
f.close()
for line in cfg_content.split("\n"):
if line.find('addr : ') == 0:
address = line.split(':')[1].strip()
port = line.split(':')[2].strip()
if address == '0.0.0.0':
address = host
result.append([address, port])
return result
def discovery_tomcat(*args):
'''
discovery tomcat instance's host and port
'''
result = []
port = ''
cmdline = []
host = Monitor.get_local_ip()
for i in psutil.process_iter():
if i.name() == BINNAME:
pid = int(i.pid)
cmdline = i.cmdline()
for opt in cmdline:
if opt.find('-Dcom.sun.management.jmxremote.port') == 0:
monitor_port = opt.strip().split('=')[1]
if opt.find('-Dcatalina.base') == 0:
config_file = opt.strip().split('=')[1] + '/conf/server.xml'
tree = ET.ElementTree(file=config_file)
root = tree.getroot()
for root_child in root:
if root_child.tag == 'Service':
for service_child in root_child:
if service_child.tag == 'Connector':
listen_port = service_child.attrib['port']
if monitor_port.isdigit() and listen_port.isdigit():
result.append([host, listen_port, monitor_port])
return result
def discovery_etcd(*args):
'''
discovery etcd instance's host and port
'''
result = []
port = ''
address = ''
cmdline = []
host = Monitor.get_local_ip()
for i in psutil.process_iter():
if i.name() == BINNAME:
pid = int(i.pid)
cmdline = i.cmdline()
for opt in cmdline:
if opt.find('--name=') == 0:
name = opt.strip().split('=')[1].strip()
config_file = CFG_DIR + 'etcd-' + name + '.conf'
if os.path.exists(config_file):
f = open(config_file, 'r')
cfg_content = f.read()
f.close()
for line in cfg_content.split("\n"):
if line.find('ETCD_LISTEN_CLIENT_URLS=') == 0:
address_port = line.split('//')[1].strip('"')
address = address_port.split(':')[0]
port = address_port.split(':')[1]
if address == '0.0.0.0':
address = host
result.append([address, port, name])
return result
def discovery_nginx(status_path=False, *args):
'''
discovery nginx instance's host and port
'''
nginx_pid = ''
instance_list = []
for i in psutil.process_iter():
if i.name() == BINNAME and i.username() == 'root':
nginx_pid = i.pid
nginx_proc_file = '/proc/{0}/cmdline'.format(nginx_pid)
if os.path.exists(nginx_proc_file):
f = open(nginx_proc_file, 'r')
nginx_cmd = f.read()
f.close()
nginx_cfg_file = nginx_cmd.strip().split()[-1]
f = open(nginx_cfg_file, 'r')
cfg_content = f.read()
f.close()
# print cfg_content
p_status_on = re.compile('stub_status\s+on;', re.I)
m_status_on = p_status_on.search(cfg_content)
if m_status_on:
p_path = re.compile('location.*', re.I)
m_path = p_path.search(cfg_content)
if m_path:
path = m_path.group(0).strip().split()[1]
p_port = re.compile('listen\s+.*', re.I)
m_port = p_port.search(cfg_content)
if m_port:
l_port = m_port.group(0).strip().strip(';').split()[-1].split(
':')
if (len(l_port) == 2):
port = l_port[1]
else:
port = l_port[0]
ip = Monitor.get_local_ip()
result = []
if not status_path:
result.append([str(ip), str(port)])
return result
else:
return path
def discovery_nginx(status_path=False, *args):
'''
discovery nginx instance's host and port
'''
nginx_pid = ''
instance_list = []
for i in psutil.process_iter():
if i.name() == BINNAME and i.username() == 'root':
nginx_pid = i.pid
nginx_proc_file = '/proc/{0}/cmdline'.format(nginx_pid)
if os.path.exists(nginx_proc_file):
f = open(nginx_proc_file, 'r')
nginx_cmd = f.read()
f.close()
nginx_cfg_file = nginx_cmd.strip().split()[-1]
f = open(nginx_cfg_file, 'r')
cfg_content = f.read()
f.close()
# print cfg_content
p_status_on = re.compile('stub_status\s+on;', re.I)
m_status_on = p_status_on.search(cfg_content)
if m_status_on:
p_path = re.compile('location.*', re.I)
m_path = p_path.search(cfg_content)
if m_path:
path = m_path.group(0).strip().split()[1]
p_port = re.compile('listen\s+.*', re.I)
m_port = p_port.search(cfg_content)
if m_port:
l_port = m_port.group(0).strip().strip(';').split()[-1].split(':')
if (len(l_port) == 2):
port = l_port[1]
else:
port = l_port[0]
ip = Monitor.get_local_ip()
result = []
if not status_path:
result.append([str(ip), str(port)])
return result
else:
return path
def discovery_codisHa(*args):
'''
discovery codis-ha instance's host and port
'''
result = []
port = ''
address = ''
cmdline = []
host = Monitor.get_local_ip()
for i in psutil.process_iter():
if i.name() == BINNAME:
pid = int(i.pid)
cmdline = i.cmdline()
for opt in cmdline:
if opt.find('--dashboard=') == 0:
address = opt.strip().split('=')[1].split(':')[0].strip()
port = opt.strip().split('=')[1].split(':')[1].strip()
result.append([address, port])
return result
def get_redis_data(instance_name, *args):
"""
get monitor data from redis
@param instance_name: ip:port:passwd
@return: dict
"""
ip, port, passwd = instance_name.split("/")
passwd = Monitor.decode_password(passwd)
r = redis.StrictRedis(host=ip, port=port, password=passwd)
d = r.info()
check_items = {
"redis_version": str,
"redis_mode": str, # standalone,
"uptime_in_seconds": int,
"process_id": int,
# Clients
"connected_clients": int, # 当前客户端连接数
"blocked_clients": int, # 正在等待阻塞命令（BLPOP、BRPOP、BRPOPLPUSH）的客户端的数量
"connected_slaves": int, # 当前从连接的数量
"rejected_connections": int,
"total_connections_received": int, # 运行以来连接过的客户端的总数量
"client_longest_output_list": int, # 当前连接的客户端当中，最长的输出列表
"client_biggest_input_buf": int, # 当前连接的客户端当中，最大输入缓存
# Memory
"used_memory": int, # 由redis分配器分配的内存总量，以字节（byte）为单位
"used_memory_rss": int, # 从操作系统的角度，返回rRedis已分配的内存总量（俗称常驻集大小),这个值和top、ps等命令的输出一致。
"used_memory_peak": int, # redis的内存消耗峰值（以字节为单位）
"used_memory_lua": int, # 引擎所使用的内存大小（以字节为单位）
"mem_fragmentation_ratio": float, # 内存碎片比率:userd_memory_rss和used_memory之间的比率
# Persistence
"rdb_changes_since_last_save": int, # 上次保存数据库之后，执行命令的次数
"rdb_bgsave_in_progress": int, # 后台进行中的save操作的数量
"rdb_last_save_time": int, # 最后一次成功保存的时间点，以 UNIX 时间戳格式显示
"rdb_last_bgsave_status": str,
"rdb_last_bgsave_time_sec": int,
"rdb_current_bgsave_time_sec": int,
# Stats
"total_commands_processed": int, # 运行以来执行过的命令的总数量
"instantaneous_ops_per_sec": int, # 每秒瞬间ops数
"expired_keys": int, # 运行以来过期的 key 的数量
"evicted_keys": int, # 运行以来删除过的key的数量
"keyspace_hits": int, # 命中 key 的次数
"keyspace_misses": int, # 不命中 key 的次数
"pubsub_channels": int, # 当前使用中的频道数量
"pubsub_patterns": int, # 当前使用的模式的数量
#'latest_fork_usec':int,
# Replication
"role": str, # 当前实例的角色master还是slave
"master_host": str,
"master_port": int,
"master_link_status": str, # up or down
"master_last_io_seconds_ago": int,
"master_sync_in_progress": int,
"slave_lists": str,
#'slave0:ip=192.168.200.25,port=62710,state=online,offset=823669419,lag=1 #offset 当前从的数据偏移量位置
# CPU
"used_cpu_sys": float,
"used_cpu_user": float,
"used_cpu_sys_children": float,
"used_cpu_user_children": float,
}
redis_stats = {k: d[k] if d.has_key(k) else v() for k, v in check_items.iteritems()}
redis_stats["role"] = 1 if redis_stats.get("role", "master") == "master" else 2
redis_stats["keyspace_hits_rate"] = "{0:.2f}".format(
float(redis_stats.get("keyspace_hits", 0))
/ float(redis_stats.get("keyspace_hits", 1) + redis_stats.get("keyspace_misses", 0) + 0.01)
* 100
)
if redis_stats["connected_slaves"] > 0:
slave_lists = set()
for i in range(redis_stats["connected_slaves"]):
if d.has_key("slave%s" % i):
slave_lists.add(str(d["slave%s" % i]))
redis_stats["slave_lists"] = ",".join(list(slave_lists))
return redis_stats
def get_redis_data(instance_name, *args):
"""
get monitor data from redis
@param instance_name: ip:port:passwd
@return: dict
"""
ip, port, passwd = instance_name.split('/')
passwd = Monitor.decode_password(passwd)
r = redis.StrictRedis(host=ip, port=port, password=passwd)
d = r.info()
c = r.config_get('maxmemory')
d['max_memory'] = c['maxmemory']
check_items = {
'redis_version': str,
'redis_mode': str, # standalone,
'uptime_in_seconds': int,
'process_id': int,
# Clients
'connected_clients': int, # 当前客户端连接数
'blocked_clients': int, # 正在等待阻塞命令（BLPOP、BRPOP、BRPOPLPUSH）的客户端的数量
'connected_slaves': int, # 当前从连接的数量
'rejected_connections': int,
'total_connections_received': int, # 运行以来连接过的客户端的总数量
'client_longest_output_list': int, # 当前连接的客户端当中，最长的输出列表
'client_biggest_input_buf': int, # 当前连接的客户端当中，最大输入缓存
# Memory
'max_memory': int, # 配置文件限制的最大内存 以 byte 为单位，0代表不限制
'used_memory': int, # 由redis分配器分配的内存总量，以字节（byte）为单位
'used_memory_rss': int, # 从操作系统的角度，返回rRedis已分配的内存总量（俗称常驻集大小),这个值和top、ps等命令的输出一致。
'used_memory_peak': int, # redis的内存消耗峰值（以字节为单位）
'used_memory_lua': int, # 引擎所使用的内存大小（以字节为单位）
'mem_fragmentation_ratio': float, # 内存碎片比率:userd_memory_rss和used_memory之间的比率
# Persistence
'rdb_changes_since_last_save': int, # 上次保存数据库之后，执行命令的次数
'rdb_bgsave_in_progress': int, # 后台进行中的save操作的数量
'rdb_last_save_time': int, # 最后一次成功保存的时间点，以 UNIX 时间戳格式显示
'rdb_last_bgsave_status': str,
'rdb_last_bgsave_time_sec': int,
'rdb_current_bgsave_time_sec': int,
# Stats
'total_commands_processed': int, # 运行以来执行过的命令的总数量
'instantaneous_ops_per_sec': int, # 每秒瞬间ops数
'expired_keys': int, # 运行以来过期的 key 的数量
'evicted_keys': int, # 运行以来删除过的key的数量
'keyspace_hits': int, # 命中 key 的次数
'keyspace_misses': int, # 不命中 key 的次数
'pubsub_channels': int, # 当前使用中的频道数量
'pubsub_patterns': int, # 当前使用的模式的数量
# 'latest_fork_usec':int,
# Replication
'role': str, # 当前实例的角色master还是slave
'master_host': str,
'master_port': int,
'master_link_status': str, # up or down
'master_last_io_seconds_ago': int,
'master_sync_in_progress': int,
'slave_lists': str,
# 'slave0:ip=192.168.200.25,port=62710,state=online,offset=823669419,lag=1 #offset 当前从的数据偏移量位置
# CPU
'used_cpu_sys': float,
'used_cpu_user': float,
'used_cpu_sys_children': float,
'used_cpu_user_children': float,
}
redis_stats = {k: d[k] if d.has_key(k) else v() for k, v in check_items.iteritems()}
redis_stats['role'] = 1 if redis_stats.get('role', 'master') == 'master' else 2
redis_stats['keyspace_hits_rate'] = '{0:.2f}'.format(float(redis_stats.get('keyspace_hits', 0)) / float(
redis_stats.get('keyspace_hits', 1) + redis_stats.get('keyspace_misses', 0) + 0.01) * 100)
if not redis_stats.get('max_memory') == '0':
redis_stats['memory_used_percent'] = '{0:.2f}'.format(
float((redis_stats.get('used_memory')) / float(redis_stats.get('max_memory')) * 100))
if redis_stats['connected_slaves'] > 0:
slave_lists = set()
for i in range(redis_stats['connected_slaves']):
if d.has_key("slave%s" % i):
slave_lists.add(str(d["slave%s" % i]))
redis_stats['slave_lists'] = ','.join(list(slave_lists))
return redis_stats
def discovery_ipmi(arg='', *args):
'''
discovery ipmi instance
'''
result = []
host = Monitor.get_local_ip()
if not arg:
result.append([host])
elif arg == 'FAN':
(status, output) = commands.getstatusoutput(
"/usr/bin/ipmitool sdr type fan | grep -v 'Fully Redundant' | cut -d '|' -f 1 "
)
if status == 0:
for fan in output.split('\n'):
fan = fan.lower()
fan = fan.strip()
fan = fan.replace(' ', '_')
result.append([host, fan])
# cpu memory system io
elif arg == 'CMSI':
(status, output) = commands.getstatusoutput(
"/usr/bin/ipmitool sdr elist full | grep -E '(CPU|IO|MEM|SYS)'| cut -d '|' -f 1"
)
if status == 0:
for cmsi in output.split('\n'):
cmsi = cmsi.lower()
cmsi = cmsi.strip()
cmsi = cmsi.replace(' ', '_')
result.append([host, cmsi])
# current voltage temp
elif arg == 'VCT':
(status, output) = commands.getstatusoutput(
"/usr/bin/ipmitool sdr elist full | grep -E '(Inlet Temp|Exhaust Temp|Current|Voltage)'| cut -d '|' -f 1"
)
if status == 0:
for vct in output.split('\n'):
vct = vct.lower()
vct = vct.strip()
vct = vct.replace(' ', '_')
result.append([host, vct])
elif arg == 'PW':
(status, output) = commands.getstatusoutput(
"/usr/bin/ipmitool chassis status | grep -E '(System Power|Power Overload)' | cut -d ':' -f 1"
)
if status == 0:
for pw in output.split('\n'):
pw = pw.lower()
pw = pw.strip()
pw = pw.replace(' ', '_')
result.append([host, pw])
elif arg == 'SEL':
(status, output) = commands.getstatusoutput(
"/usr/bin/ipmitool sel info | grep -v Information | cut -d ':' -f 1"
)
if status == 0:
for sel in output.split('\n'):
sel = sel.lower()
sel = sel.strip()
sel = sel.replace(' ', '_')
result.append([host, sel])
return result
