python並發獲取snmp信息及性能測試

python & snmp

用python獲取snmp信息有多個現成的庫可以使用，其中比較常用的是netsnmp和pysnmp兩個庫。網上有較多的關於兩個庫的例子。
本文重點在於如何並發的獲取snmp的數據，即同時獲取多台機器的snmp信息。

netsnmp

先說netsnmp。python的netsnmp，其實是來自於net-snmp包。
python通過一個c文件調用net-snmp的接口獲取數據。

因此，在並發獲取多台機器的時候，不能夠使用協程獲取。因為使用協程，在get數據的時候，協程會一直等待net-snmp接口返回數據，而不會像socket使用時那樣在等待數據時把CPU切換給其他協程使用。從這點上來說，使用協程和串行獲取沒有區別。

那么如何解決並發獲取的問題呢？可以使用線程，多線程獲取(當然也可以使用多進程)。多個線程同時調用net-snmp的接口獲取數據，然后cpu在多個線程之間不停切換。當一個線程獲取一個結果后，可以繼續調用接口獲取下一個snmp數據。

這里我寫了一個樣例程序。首先把所有的host和oid做成任務放到隊列里，然后啟動多個線程，去執行獲取任務。程序樣例如下：

import threading
import time
import netsnmp
import Queue

start_time = time.time()
hosts = ["192.20.150.109", "192.20.150.110", "192.20.150.111", "192.20.150.112", "192.20.150.113", "192.20.150.114",
         "192.20.150.115", "192.20.150.116", "192.20.150.117", "192.20.150.118", "192.20.150.119", "192.20.150.120",
         "192.20.150.121", "192.20.80.148", "192.20.80.149", "192.20.96.59", "192.20.82.14", "192.20.82.15",
         "192.20.82.17", "192.20.82.19", "192.20.82.12", "192.20.80.139", "192.20.80.137", "192.20.80.136",
         "192.20.80.134", "192.20.80.133", "192.20.80.131", "192.20.80.130", "192.20.81.141", "192.20.81.140",
         "192.20.82.26", "192.20.82.28", "192.20.82.23", "192.20.82.21", "192.20.80.128", "192.20.80.127",
         "192.20.80.122", "192.20.81.159", "192.20.80.121", "192.20.80.124", "192.20.81.151", "192.20.80.118",
         "192.20.80.119", "192.20.80.113", "192.20.80.112", "192.20.80.116", "192.20.80.115", "192.20.78.62",
         "192.20.81.124", "192.20.81.125", "192.20.81.122", "192.20.81.121", "192.20.82.33", "192.20.82.31",
         "192.20.82.32", "192.20.82.30", "192.20.81.128", "192.20.82.39", "192.20.82.37", "192.20.82.35",
         "192.20.81.130", "192.20.80.200", "192.20.81.136", "192.20.81.137", "192.20.81.131", "192.20.81.133",
         "192.20.81.134", "192.20.82.43", "192.20.82.45", "192.20.82.41", "192.20.79.152", "192.20.79.155",
         "192.20.79.154", "192.25.76.235", "192.25.76.234", "192.25.76.233", "192.25.76.232", "192.25.76.231",
         "192.25.76.228", "192.25.20.96", "192.25.20.95", "192.25.20.94", "192.25.20.93", "192.24.163.14",
         "192.24.163.21", "192.24.163.29", "192.24.163.6", "192.18.136.22", "192.18.136.23", "192.24.193.2",
         "192.24.193.19", "192.24.193.18", "192.24.193.11", "192.20.157.132", "192.20.157.133", "192.24.212.232",
         "192.24.212.231", "192.24.212.230"]
oids = [".1.3.6.1.4.1.2021.11.9.0",".1.3.6.1.4.1.2021.11.10.0",".1.3.6.1.4.1.2021.11.11.0",".1.3.6.1.4.1.2021.10.1.3.1",
        ".1.3.6.1.4.1.2021.10.1.3.2",".1.3.6.1.4.1.2021.10.1.3.3",".1.3.6.1.4.1.2021.4.6.0",".1.3.6.1.4.1.2021.4.14.0",
        ".1.3.6.1.4.1.2021.4.15.0"]
myq = Queue.Queue()
rq = Queue.Queue()

#把host和oid組成任務
for host in hosts:
    for oid in oids:
        myq.put((host,oid))

def poll_one_host():
    while True:
        try:
            #死循環從隊列中獲取任務，直到隊列任務為空
            host, oid = myq.get(block=False)
            session = netsnmp.Session(Version=2, DestHost=host, Community="cluster",Timeout=3000000,Retries=0)
            var_list = netsnmp.VarList()
            var_list.append(netsnmp.Varbind(oid))
            ret = session.get(var_list)
            rq.put((host, oid, ret, (time.time() - start_time)))
        except Queue.Empty:
            break

thread_arr = []

#開啟多線程
num_thread = 50
for i in range(num_thread):
    t = threading.Thread(target=poll_one_host, kwargs={})
    t.setDaemon(True)
    t.start()
    thread_arr.append(t)

#等待任務執行完畢
for i in range(num_thread):
    thread_arr[i].join()

while True:
    try:
        info = rq.get(block=False)
        print info
    except Queue.Empty:
        print time.time() - start_time
        break

netsnmp除了支持get操作之外，還支持walk操作，即遍歷某個oid。
但是walk使用的時候需要謹慎，以免導致高延時等問題，具體可以參見之前的一篇snmpwalk高延時問題分析的博客。

pysnmp

pysnmp是用python實現的一套snmp協議的庫。其自身提供了對於異步的支持。

import time
import Queue
from pysnmp.hlapi.asyncore import *
t = time.time()
myq = Queue.Queue()

#回調函數。在有數據返回時觸發
def cbFun(snmpEngine, sendRequestHandle, errorIndication, errorStatus, errorIndex, varBinds, cbCtx):
     myq.put((time.time()-t, varBinds))
hosts = ["192.20.150.109", "192.20.150.110", "192.20.150.111", "192.20.150.112", "192.20.150.113", "192.20.150.114",
         "192.20.150.115", "192.20.150.116", "192.20.150.117", "192.20.150.118", "192.20.150.119", "192.20.150.120",
         "192.20.150.121", "192.20.80.148", "192.20.80.149", "192.20.96.59", "192.20.82.14", "192.20.82.15",
         "192.20.82.17", "192.20.82.19", "192.20.82.12", "192.20.80.139", "192.20.80.137", "192.20.80.136",
         "192.20.80.134", "192.20.80.133", "192.20.80.131", "192.20.80.130", "192.20.81.141", "192.20.81.140",
         "192.20.82.26", "192.20.82.28", "192.20.82.23", "192.20.82.21", "192.20.80.128", "192.20.80.127",
         "192.20.80.122", "192.20.81.159", "192.20.80.121", "192.20.80.124", "192.20.81.151", "192.20.80.118",
         "192.20.80.119", "192.20.80.113", "192.20.80.112", "192.20.80.116", "192.20.80.115", "192.20.78.62",
         "192.20.81.124", "192.20.81.125", "192.20.81.122", "192.20.81.121", "192.20.82.33", "192.20.82.31",
         "192.20.82.32", "192.20.82.30", "192.20.81.128", "192.20.82.39", "192.20.82.37", "192.20.82.35",
         "192.20.81.130", "192.20.80.200", "192.20.81.136", "192.20.81.137", "192.20.81.131", "192.20.81.133",
         "192.20.81.134", "192.20.82.43", "192.20.82.45", "192.20.82.41", "192.20.79.152", "192.20.79.155",
         "192.20.79.154", "192.25.76.235", "192.25.76.234", "192.25.76.233", "192.25.76.232", "192.25.76.231",
         "192.25.76.228", "192.25.20.96", "192.25.20.95", "192.25.20.94", "192.25.20.93", "192.24.163.14",
         "192.24.163.21", "192.24.163.29", "192.24.163.6", "192.18.136.22", "192.18.136.23", "192.24.193.2",
         "192.24.193.19", "192.24.193.18", "192.24.193.11", "192.20.157.132", "192.20.157.133", "192.24.212.232",
         "192.24.212.231", "192.24.212.230"]

oids = [".1.3.6.1.4.1.2021.11.9.0",".1.3.6.1.4.1.2021.11.10.0",".1.3.6.1.4.1.2021.11.11.0",".1.3.6.1.4.1.2021.10.1.3.1",
        ".1.3.6.1.4.1.2021.10.1.3.2",".1.3.6.1.4.1.2021.10.1.3.3",".1.3.6.1.4.1.2021.4.6.0",".1.3.6.1.4.1.2021.4.14.0",
        ".1.3.6.1.4.1.2021.4.15.0"]
        
snmpEngine = SnmpEngine()

#添加任務
for oid in oids:
    for h in hosts:
        getCmd(snmpEngine,
            CommunityData('cluster'),
            UdpTransportTarget((h, 161), timeout=3, retries=0,),
            ContextData(),
            ObjectType(ObjectIdentity(oid)),
            cbFun=cbFun)
time1 = time.time() - t

#執行異步獲取snmp
snmpEngine.transportDispatcher.runDispatcher()

#打印結果
while True:
    try:
        info = myq.get(block=False)
        print info
    except Queue.Empty:
        print time1
        print time.time() - t
        break

pysnmp本身只支持最基礎的get和getnext命令，因此如果想使用walk，需要自己進行實現。

性能測試

在同一個環境下，對兩者進行了性能測試。兩者對198個host，10個oid進行采集。

測試組	耗時(sec)
netsnmp(20線程)	6.252
netsnmp(50線程)	3.269
netsnmp(200線程)	3.265
pysnmp	4.812

可以看到netsnmp的采集速度跟線程數有關。當線程數增大到一定程度，采集時間不再縮短。因為開辟線程同樣會消耗時間。而已有的線程已經足夠處理。

pysnmp性能較之略差一下。詳細分析pysnmp在添加任務(執行getCmd時)消耗了約1.2s，之后的采集約消耗3.3秒。

在增加了oid數，在進行實驗。host仍然是198個，oid是42個。

測試組	耗時(sec)
netsnmp(20線程)	30.935
netsnmp(50線程)	12.914
netsnmp(200線程)	4.044
pysnmp	11.043

可以看到差距被進一步拉大。在線程足夠多的情況下，netsnmp的效率要明顯強於pysnmp。

因為二者都支持可以並行采集多個host，從易用性來說，netsnmp更為簡單一些，且netsnmp支持walk功能。本文更加推薦netsnmp。

安裝netsnmp需要安裝net-snmp。如果centos，則使用yum會較為方便。