增加根据每块计算板的不同, 灵活判断应该重启哪些应用

6 files changed, 219 insertions, 37 deletions

diff --git a/py_cmd/common_modules_deploy.json b/py_cmd/common_modules_deploy.json
new file mode 100644
index 0000000..786644f
--- /dev/null
+++ b/py_cmd/common_modules_deploy.json
@@ -0,0 +1,12 @@
+#此文件定义四块计算板分别运行什么模块

+#先根据tsg_chassis_ip.json, 根据当前设备的ip地址, 知道当前设备的sled名称

+#然后根据设备类型名称, 找到当前运行的所有模块(应用)名称

+#然后根据模块名称, 到common_modules_operator.json查找每个模块的操作方法

+{

+    "modules_dispatch":

+     

+       "mcn0": ["kni"]

+       "mcn1": ["tfe"]

+        "mcn2": ["tfe"]

+        "mcn3": ["tfe"]

+}
\ No newline at end of file
diff --git a/py_cmd/common_modules_operator.json b/py_cmd/common_modules_operator.json
new file mode 100644
index 0000000..1bd6a16
--- /dev/null
+++ b/py_cmd/common_modules_operator.json
@@ -0,0 +1,26 @@
+#此文件定义, 每个模块的启动、停止、检查运行状态的方法

+{

+   "modules":[

+   {sapp:

+   start: 

+       "cwd"

+       "exe"

+   stop:

+       "killall -9 r3 sapp"

+   check healthy:

+        "ps -ef | grep sapp "

+   }

+   

+    {telegraf:

+   start: 

+       "cwd" : ""

+       "exe" : "service influxdb restart"

+   stop:

+       "service influxdb stop"

+   check healthy:

+        "service influxdb status | grep Active "

+   }  

+   

+   ]

+

+}
\ No newline at end of file
diff --git a/py_tools/get_traffic_by_marsio.py b/py_tools/get_traffic_by_marsio.py
index 8bf1282..7dbc53f 100644
--- a/py_tools/get_traffic_by_marsio.py
+++ b/py_tools/get_traffic_by_marsio.py
@@ -117,7 +117,7 @@ def get_and_send_marsio_traffic(logger, json_fp, telegraf_client, devsym, arg_fl
         if value == "":
             logger.critical("can't get device:%s traffic!" %(devsym))
             sys.exit(1)
-        __metric_dict_speed[item] = value
+        __metric_dict_speed[item] = int(value)
         
     telegraf_client.metric('interface', __metric_dict_speed, tags = metric_tag)
     #logger.info("telegraf_client send metric")
diff --git a/py_tools/get_traffic_by_proc.py b/py_tools/get_traffic_by_proc.py
new file mode 100644
index 0000000..fec12bb
--- /dev/null
+++ b/py_tools/get_traffic_by_proc.py
@@ -0,0 +1,115 @@
+#coding=utf-8

+import sys

+import re

+import syslog

+import subprocess

+import time

+import telegraf

+import logging

+import logging.handlers

+from common_telegraf import *

+from common_logger import *

+

+Receive_bytes_index = 1

+Receive_packets_index = 2

+Receive_err_index = 3

+Receive_drop_index = 4

+

+Transmit_bytes_index = 9

+Transmit_packets_index = 10

+Transmit_err_index = 11

+Transmit_drop_index = 12

+

+#return exitcode value + output message:

+#    0: succ

+#    1: error

+def system_cmd_run(cmd_str):

+    dangerous_cmd = {"rm", "mv", "poweroff", "shutdown"}

+    

+    for cmd in dangerous_cmd:

+        pattern = "\s*%s" %(cmd)

+        match_str = re.match(pattern, cmd_str)

+        if not match_str is None:

+            print("can't run this cmd:%s" %(cmd_str))

+            sys.exit(1)

+    

+    try:

+        exitcode, output = subprocess.getstatusoutput(cmd_str)

+    except Exception as e:  

+        #if exitcode != 0:

+        #    output = ""

+        print(e)

+        return 1, e

+

+    return exitcode, output

+    

+def get_traffic_from_proc(logger, device_name):

+    cmd_str = "cat /proc/net/dev | grep %s:" %(device_name)

+    res_code, res_buf = system_cmd_run(cmd_str)

+    if res_code != 0:

+        return ""

+    if len(res_buf) <= 0:

+        return ""

+    

+    logger.debug(res_buf)    

+    

+    sections_array = re.split(r'[:\t\s\r\n]\s*', res_buf.strip())

+

+    packets_recv = sections_array[Receive_packets_index]

+    bytes_recv = sections_array[Receive_bytes_index]

+    errin = sections_array[Receive_err_index]

+    dropin = sections_array[Receive_drop_index]

+    

+    packets_sent = sections_array[Transmit_packets_index]

+    bytes_sent = sections_array[Transmit_bytes_index]

+    errout = sections_array[Transmit_err_index]

+    dropout = sections_array[Transmit_drop_index]     

+

+    return packets_recv,bytes_recv,errin,dropin,packets_sent,bytes_sent,errout,dropout

+    

+def send_metrics_by_proc(logger, telegraf_client, dev_name, arg_flow_type, arg_node_name):  

+    metric_tag = {'device': dev_name, 'flow_type':arg_flow_type, 'node':arg_node_name}

+    

+    old_packets_recv,old_bytes_recv,old_errin,old_dropin,old_packets_sent,old_bytes_sent,old_errout,old_dropout = get_traffic_from_proc(logger,dev_name)

+    if len(old_packets_recv) <= 0:

+        logger.error("can't get %s status" %(dev_name))

+        return 1 

+    time.sleep(1)

+    new_packets_recv,new_bytes_recv,new_errin,new_dropin,new_packets_sent,new_bytes_sent,new_errout,new_dropout = get_traffic_from_proc(logger,dev_name)

+    if len(new_packets_recv) <= 0:

+        logger.error("can't get %s status" %(dev_name))

+        return 1 

+        

+    metrict_val = {}

+    metrict_val['PhyRXBits'] = int(8 *(int(new_bytes_recv) - int(old_bytes_recv)))

+    metrict_val['PhyRXError'] = int(new_errin) - int(old_errin)

+    metrict_val['PhyRXFrame'] = int(new_packets_recv) - int(old_packets_recv)

+    metrict_val['PhyRXMissed'] = 0

+    metrict_val['PhyRXNoBUF'] = 0

+

+    metrict_val['PhyTXBits'] = int(8 *(int(new_bytes_recv) - int(old_bytes_recv)))

+    metrict_val['PhyTXError'] = int(new_errin) - int(old_errin)

+    metrict_val['PhyTXFrame'] = int(new_packets_recv) - int(old_packets_recv)

+    metrict_val['PhyRXMissed'] = 0

+    metrict_val['PhyRXNoBUF'] = 0

+    

+    metrict_val['UsrRXDrops'] = int(new_dropin) - int(old_dropin)

+    metrict_val['UsrTXDrops'] = int(new_dropout) - int(old_dropout)

+    

+    telegraf_client.metric('interface', metrict_val, tags = metric_tag)

+    

+    logger.debug(metrict_val)

+    

+if __name__ == '__main__':

+    global logger 

+    global time_interval

+    if len(sys.argv) <= 1:

+        print("Usage: %s timeinterval" %(sys.argv[0]))    

+        time_interval = 5

+    else:        

+        time_interval = int(sys.argv[1])    

+    tele_tags = {}

+    telegraf_client = telegraf_init('127.0.0.1', 8126, tele_tags)

+    logger = logger_init(10) 

+    

+    send_metrics_by_proc(logger, telegraf_client, 'ens33', 'inline', 'mcn1')
\ No newline at end of file
diff --git a/py_tools/get_traffic_by_psutil.py b/py_tools/get_traffic_by_psutil.py
index 9fb77c2..263414e 100644
--- a/py_tools/get_traffic_by_psutil.py
+++ b/py_tools/get_traffic_by_psutil.py
@@ -4,7 +4,11 @@ import sys
 import psutil
 import time
 import telegraf
-        
+import logging
+import logging.handlers
+from common_telegraf import *
+from common_logger import *
+
 def get_stats_for_device(dev_name):
     
     devlist = psutil.net_io_counters(pernic=True).keys()
@@ -38,23 +42,23 @@ def send_metrics_by_psutil(logger, telegraf_client, dev_name, arg_flow_type, arg
     metric_tag = {'device': dev_name, 'flow_type':arg_flow_type, 'node':arg_node_name}
     
     old_packets_recv,old_bytes_recv,old_errin,old_dropin,old_packets_sent,old_bytes_sent,old_errout,old_dropout = get_stats_for_device(dev_name)
-    time.sleep(3)
+    time.sleep(1)
     new_packets_recv,new_bytes_recv,new_errin,new_dropin,new_packets_sent,new_bytes_sent,new_errout,new_dropout = get_stats_for_device(dev_name)
 
     metrict_val = {}
     metrict_val['PhyRXBits'] = int(8 *(new_bytes_recv - old_bytes_recv))
     metrict_val['PhyRXError'] = int(new_errin-old_errin)
     metrict_val['PhyRXFrame'] = int(new_packets_recv - old_packets_recv)
-    metrict_val['PhyRXMissed'] = 0
-    metrict_val['PhyRXNoBUF'] = 0
+    metrict_val['PhyRXMissed'] = int('0')
+    metrict_val['PhyRXNoBUF'] = int('0')
 
     metrict_val['PhyTXBits'] = int(8 *(new_bytes_recv - old_bytes_recv))
     metrict_val['PhyTXError'] = int(new_errin-old_errin)
     metrict_val['PhyTXFrame'] = int(new_packets_recv - old_packets_recv)
-    metrict_val['PhyRXMissed'] = 0
-    metrict_val['PhyRXNoBUF'] = 0
+    metrict_val['PhyRXMissed'] = int('0')
+    metrict_val['PhyRXNoBUF'] = int('0')
     
-    metrict_val['UsrRXDrops'] = int(new_dropin - new_dropin)
+    metrict_val['UsrRXDrops'] = int(new_dropin - old_dropin)
     metrict_val['UsrTXDrops'] = int(new_dropout - old_dropout)
     
     telegraf_client.metric('interface', metrict_val, tags = metric_tag)
@@ -71,8 +75,8 @@ if __name__ == '__main__':
     else:        
         time_interval = int(sys.argv[1])    
 
-    telegraf_client = telegraf_init('127.0.0.1', 8126)
+    tele_tags = {}
+    telegraf_client = telegraf_init('127.0.0.1', 8126, tele_tags)
     logger = logger_init(10) 
-    telegraf_client = telegraf_init()
     
-    send_metrics_by_psutil(logger, telegraf_client, 'ens33', 'inline', 'mcn_1')
+    send_metrics_by_psutil(logger, telegraf_client, 'ens33', 'inline', 'mcn1')
diff --git a/py_tools/tsg_monit_interface.py b/py_tools/tsg_monit_interface.py
index 64a3729..62ccfbe 100644
--- a/py_tools/tsg_monit_interface.py
+++ b/py_tools/tsg_monit_interface.py
@@ -5,8 +5,10 @@ import psutil
 import time
 import json
 import telegraf
+import logging
 from get_traffic_by_psutil import *
 from get_traffic_by_marsio import *
+from get_traffic_by_proc import *
 from common_telegraf import *
 from common_logger import *
 from common_args import *
@@ -16,37 +18,60 @@ from common_get_tags import *
 INTERFACE_JSON_PATH = '/opt/tsg/etc/tsg_chassis_interface.json'
 
 
-def get_local_node():
+#return value:
+# 0, error msg, json_dict : succ
+# 1, error msg, "" : error
+def tsg_json_parse(file_name):
     try:
-        with open(INTERFACE_JSON_PATH) as json_fp:
-            json_dict = json.load(json_fp)
-            return json_dict['local_chassis_node']
-    except IOError:
-        return ""
+        with open(file_name) as json_fp:
+            try:
+                json_dict = json.load(json_fp)
+                return 0, "", json_dict
+            except Exception as e:
+                return 1, e, ""
+    except Exception as e:
+        return 1, e , ""
+
 
+def get_local_node():
+    ret, msg, json_dict = tsg_json_parse(INTERFACE_JSON_PATH)
+    
+    if ret == 0:
+        return json_dict['local_chassis_node']
+    else:
+        logger.critical("%s!" %(msg))
+        sys.exit(1)  
 
 def get_interface_list():
-    try:
-        with open(INTERFACE_JSON_PATH) as json_fp:
-            json_dict = json.load(json_fp)
-            interface_list =  json_dict['interface_list']
-            #print(interface_list)
-            for dev in interface_list:
-                if 'marsio' == dev['dev_type']:
-                    logger.debug("get traffic stat by %s for: %s" %(dev['dev_type'], dev['dev_name']))
-                    # call function in get_traffic_by_marsio.py
-                    send_metrics_by_marsio(logger, telegraf_client, dev['dev_name'], dev['flow_type'], node_name)
-                elif 'pcap' == dev['dev_type']: 
-                    logger.debug("get traffic stat by %s for: %s, please wait for 1s..." %(dev['dev_type'], dev['dev_name']))
-                    # call function in get_traffic_by_psutil.py
-                    send_metrics_by_psutil(logger, telegraf_client, dev['dev_name'], dev['flow_type'], node_name)
-                else:
-                    logger.critical("not support driver type %s for: %s" %(dev['dev_type'], dev['dev_name']))
-                    sys.exit(1)                
-    except IOError:
-        logger.critical("can't open file %s!" %(INTERFACE_JSON_PATH))
+    ret, msg, json_dict = tsg_json_parse(INTERFACE_JSON_PATH)
+    
+    if ret == 0:
+        interface_list =  json_dict['interface_list']
+    else:
+        logger.critical("%s!" %(msg))
         sys.exit(1)  
-        return ""
+        
+    if len(interface_list) <= 0:
+        logger.critical("can't get interface_list from json file %s!" %(INTERFACE_JSON_PATH))
+        sys.exit(1)   
+    
+    #print(interface_list)
+    for dev in interface_list: 
+        if 'marsio' == dev['dev_type']:
+            logger.debug("get traffic stat by %s for: %s" %(dev['dev_type'], dev['dev_name']))
+            # call function in get_traffic_by_marsio.py
+            send_metrics_by_marsio(logger, telegraf_client, dev['dev_name'], dev['flow_type'], node_name)
+        elif 'pcap' == dev['dev_type']: 
+            logger.debug("get traffic stat by %s for: %s, please wait for 1s..." %(dev['dev_type'], dev['dev_name']))
+            # call function in get_traffic_by_psutil.py
+            #send_metrics_by_psutil(logger, telegraf_client, dev['dev_name'], dev['flow_type'], node_name)
+            
+            #Ҳ��������ԭʼ�İ취, ����cat ����/proc/net/dev
+            send_metrics_by_proc(logger, telegraf_client, dev['dev_name'], dev['flow_type'], node_name)
+        else:
+            logger.critical("not support driver type %s for: %s" %(dev['dev_type'], dev['dev_name']))
+            sys.exit(1)                
+
 
     
 if __name__ == '__main__':