bug fix redesign

data-collector
Steve L. Nyemba 6 years ago
parent 8ecca833b3
commit ccc52fdaea

@ -7,146 +7,125 @@ h="""
python data-collector.py --path config.json
The configuration file is structured as JSON object as follows :
{
id: node identifier
key: customer's identification key
apps:"app_1,app_2,...",
folders:"path_1,path_2, ..."
id : identifier
key: customer's identification key,
api: http://localhost/monitor/1/client
folders:[]
}
NOTE: You can download a sample configuration file from https://the-phi.com/smart-top
"""
from utils.params import PARAMS as SYS_ARGS, Logger
import os
import requests
import pickle
import json
from threading import Thread, RLock
import monitor
import utils.agents.actor as actor
from utils.agents.manager import Manager
SYS_ARGS['host']='localhost'
ENDPOINT="http://:host/monitor".replace(":host",SYS_ARGS['host'])
class Collector(Thread) :
# from threading import Thread, RLock
from monitor import Apps, Folders
import time
from datetime import datetime
class Collector :
def __init__(self) :
Thread.__init__(self)
"""
This function initializes the data collector with critical information
The process will include validating the user's account and plan
@param key customer's key
@param id node identifier
The configuration file is passed to the class for basic initialization of variables
"""
#
# Let's open the file with the key (nothing else should be in the file
#f = open(SYS_ARGS['key'])
#SYS_ARGS['key'] = f.read()
#f.close()
headers = {"key":SYS_ARGS["key"],"id":SYS_ARGS["id"]} #,"scope":json.dumps(scope)}
self.plan = None
self.store= None
self.httpclient = requests.Session()
#headers['content-type'] = 'application/json'
try:
self.key = SYS_ARGS['key']
# Logger.log(subject='Collector',object='api',action='request',value=ENDPOINT)
# url = "/".join([ENDPOINT,"init/collector"])
Logger.log(subject='Collector',object='api',action='request',value=SYS_ARGS['api'])
url = "/".join([SYS_ARGS['api'],"init/collector"])
r = requests.post(url,headers=headers)
def init(self):
# if 'folders' not in SYS_ARGS :
# #
# # If nothing is set it will monitor the temporary directory
# self.locations = [os.environ[name] for name in ['TEMP','TMP','TMPDIR'] if name in os.environ]
# else:
# self.locations = SYS_ARGS['folders']
r = json.loads(r.text)
#
# -- let's get the list of features we are interested .
url = SYS_ARGS['api']+'/1/client/login'
key = SYS_ARGS['key']
id = SYS_ARGS['id'] if 'id' in SYS_ARGS else os.environ['HOSTNAME']
headers = {"key":key,"id":id}
if r :
#
# Persisting plan and data-store ...
self.plan = r['plan']
self.store = r['store']
info = {"store":self.store,"plan":self.plan}
#-- what features are allowed
r = self.httpclient.post(url,headers=headers)
if info['plan'] is not None and info['store'] is not None:
info['plan'] = self.plan['name']
info['store'] = self.store['args']['dbname']
_action = 'init'
self.initialize()
if r.status_code == 200 :
r = r.json()
self.features = r['features']
self.config = r['config'] #-- contains apps and folders
Logger.log(action="login",value=r)
else:
info['plan'] = self.plan is not None
info['store']= self.store is not None
_action = 'init.error'
Logger.log(subject='collector',object='api',action=_action,value=info)
except Exception as e:
print(e)
Logger.log(subject='collector',object='api',action='init.error',value=str(e))
self.monitor = None
def initialize(self):
self.features = None
self.config = None
Logger.log(action='login',value='error')
def callback(self,channel,method,header,stream):
pass
def listen(self):
factory = DataSourceFactory()
# self.qlistener = factory.instance(type="QueueListener",args=_args)
# self.qlistener.callback = self.callback
# self.qlistener.init(SYS_ARGS['id'])
def post(self,**args) :
"""
This function creates a monitoring object with the associated parameters from the plan
plan.metadata = {"agents":...,"folder_size":...,"delay":...,"limit":...,"actors":...}
This function will post data to the endpoint
"""
_agents = [monitor.DetailProcess(),monitor.FileWatch()]
_actors = [actor.Apps(),actor.Folders(),actor.Mailer()]
# Initialiing the agents with the parameter values we know of
r = []
for agent in _agents :
if agent.getName() in SYS_ARGS :
values = SYS_ARGS[agent.getName()]
# print (["init ",agent.getName(),values])
Logger.log(subject='Collector',object=agent.getName(),action='init.agent',value=values )
agent.init(values)
r.append(agent)
_agents = r
config = {"store":self.store,"plan":self.plan}
#@TODO: add SYS_ARGS content so apps that are configured reboot/delete folders that are marked
# This is an important security measure!
#
self.manager = Manager()
self.manager.init(node=SYS_ARGS['id'],agents=_agents,actors=_actors,config=config,key=self.key,host=SYS_ARGS['host'])
url = SYS_ARGS['api']+'/1/client/log'
key = SYS_ARGS['key']
id = SYS_ARGS['id'] if 'id' in SYS_ARGS else os.environ['HOSTNAME']
headers = {"key":key,"id":id,"context":args['context'],"content-type":"application/json"}
body = args['data'].to_json(orient='records')
if args.shape[0] > 0 :
r = self.httpclient.post(url,headers=headers,data=body)
Logger.log(action="post."+args['context'],value=r.status_code)
else:
Logger.log(action="data.error",value="no data")
def run(self):
"""
This funtion runs the authorized features and
This function will execute the basic functions to monitor folders and apps running on the system
given the configuration specified on the server .
"""
#self.monitor.start()
while True :
try:
self.init()
if self.config and self.features :
ELAPSED_TIME = 60* int(self.features['schedule'].replace("min","").strip())
if 'apps' in self.config :
self.post( data=(Apps()).get(filter=self.config['apps']),context="apps")
if 'folders' in self.config and self.config['folders'] :
folder = Folders()
f = folder.get(path=self.config['folders'])
self.post(data = f ,context="folders")
Logger.log(action='sleeping',value=ELAPSED_TIME)
#
# In case no configuration is provided, the system will simply fall asleep and wait
# @TODO: Evaluate whether to wake up the system or not (security concerns)!
#
time.sleep(ELAPSED_TIME)
# Logger.log(subject='Collector',object='monitor',action='rpc',value=(self.manager is None) )
thread = Thread(target=self.manager.run)
thread.start()
# print self.manager
# print self.monitor.config['store']
# for agent in self.pool :
# try:
# agent = pickle.loads(agent)
# agent.start()
except Exception,e:
Logger.log(action='error',value=e.message)
print e
break
# #p = pickle.loads(self.pool[key])
# #p.init(SYS_ARGS[key].split(','))
# #p.start()
# except Exception,e:
# print e
pass
if __name__ == '__main__' :
#
#
if 'path' in SYS_ARGS :
path = SYS_ARGS['path']
f = open(path)
p = json.loads(f.read())
SYS_ARGS = json.loads(f.read())
f.close()
else:
for id in ['apps','folders']:
if id in SYS_ARGS :
SYS_ARGS[id] = SYS_ARGS[id].split(',')
p = dict(SYS_ARGS)
Logger.init('data-collector')
SYS_ARGS = dict(SYS_ARGS,** p)
if 'apps' in SYS_ARGS :
SYS_ARGS['apps'] += [__file__]
thread = Collector()
thread.start()
collector = Collector()
collector.run()
else:
print (h)

@ -1,275 +1,113 @@
"""
This program is designed to inspect an application environment
This program should only be run on unix friendly systems
Steve L. Nyemba <steve@the-phi.com>
The Phi Technology - Smart Top
We enable the engines to be able to run a several configurations
Similarly to what a visitor design-pattern would do
This program is the core for evaluating folders and applications. Each class is specialized to generate a report in a pandas data-frame
The classes will focus on Apps, Folders and Protocols
- SmartTop.get(**args)
@TODO:
Protocols (will be used in anomaly detection)
"""
from __future__ import division
import os
import subprocess
from sets import Set
import re
import datetime
import urllib2 as http, base64
from threading import Thread, RLock
import time
import numpy as np
from utils.ml import ML
import sys
from scipy import stats
class Analysis:
def __init__(self):
self.logs = []
pass
def post(self,object):
self.logs.append(object)
def init(self):
d = datetime.datetime.now()
self.now = {"month":d.month,"year":d.year, "day":d.day,"hour":d.hour,"minute":d.minute}
def getNow(self):
d = datetime.datetime.now()
return {"month":d.month,"year":d.year, "day":d.day,"hour":d.hour,"minute":d.minute,"second":d.second}
def getName(self):
return self.__class__.__name__
def reboot(self,row,conf) :
return False
def cleanup(self,text):
return re.sub('[^a-zA-Z0-9\s:]',' ',str(text)).strip()
"""
This class is designed to analyze environment variables. Environment variables can either be folders, files or simple values
The class returns a quantifiable assessment of the environment variables (expected 100%)
"""
class Env(Analysis):
def __init__(self):
Analysis.__init__(self)
def init(self,values):
#Analysis.init(self)
self.values = values
"""
This function evaluate the validity of an environment variable by returning a 1 or 0 (computable)
The function will use propositional logic (https://en.wikipedia.org/wiki/Propositional_calculus)
"""
def evaluate(self,id):
if id in os.environ :
#
# We can inspect to make sure the environment variable is not a path or filename.
# Using propositional logic we proceed as follows:
# - (p) We determine if the value is an folder or file name (using regex)
# - (q) In case of a file or folder we check for existance
# The final result is a conjuction of p and q
#
value = os.environ[id]
expressions = [os.sep,'(\\.\w+)$']
p = sum([ re.search(xchar,value) is not None for xchar in expressions])
q = os.path.exists(value)
return int(p and q)
else:
return 0
def composite (self):
#Analysis.init(self)
r = [ self.evaluate(id) for id in self.values] ;
N = len(r)
n = sum(r)
value = 100 * round(n/N,2)
missing = [self.values[i] for i in range(0,N) if r[i] == 0]
return dict(self.getNow(),**{"value":value,"missing":missing})
"""
This class is designed to handle analaysis of the a python virtual environment i.e deltas between requirments file and a virtualenv
@TODO: update the virtual environment
"""
class Sandbox(Analysis):
def __init__(self):
Analysis.__init__(self)
def init(self,conf):
#Analysis.init(self)
if os.path.exists(conf['sandbox']) :
self.sandbox_path = conf['sandbox']
else:
self.sandbox_path = None
if os.path.exists(conf['requirements']) :
self.requirements_path = conf['requirements']
else:
self.requirements_path = None
def get_requirements (self):
f = open(self.requirements_path)
return [ name.replace('-',' ').replace('_',' ') for name in f.read().split('\n') if name != '']
"""
This function will return the modules installed in the sandbox (virtual environment)
"""
def get_sandbox_requirements(self):
cmd = ['freeze']
xchar = ''.join([os.sep]*2)
pip_vm = ''.join([self.sandbox_path,os.sep,'bin',os.sep,'pip']).replace(xchar,os.sep)
cmd = [pip_vm]+cmd
r = subprocess.check_output(cmd).split('\n')
return [row.replace('-',' ').replace('_',' ') for row in r if row.strip() != '']
def evaluate(self):
pass
def reboot(self,rows,limit=None) :
return sum([ len(item['missing']) for item in rows ]) > 0
"""
This function returns the ratio of existing modules relative to the ones expected
"""
def composite(self):
Analysis.init(self)
if self.sandbox_path and self.requirements_path :
required_modules= self.get_requirements()
sandbox_modules = self.get_sandbox_requirements()
N = len(required_modules)
n = len(Set(required_modules) - Set(sandbox_modules))
value = round(1 - (n/N),2)*100
missing = list(Set(required_modules) - Set(sandbox_modules))
return dict(self.getNow(),**{"value":value,"missing":missing})
else:
import pandas as pd
import datetime
class SmartTop:
def get(self,**args):
return None
"""
This class performs the analysis of a list of processes and determines
The class provides a quantifiable measure of how many processes it found over all
"""
class ProcessCounter(Analysis):
class Apps(SmartTop) :
def __init__(self):
Analysis.__init__(self)
def init(self,names):
#Analysis.init(self)
self.names = names
def evaluate(self,name):
cmd = "".join(['ps -eo comm |grep ',name,' |wc -l'])
handler = subprocess.Popen(cmd,shell=True,stdout=subprocess.PIPE)
return int(handler.communicate()[0].replace("\n","") )
def composite(self):
#Analysis.init(self)
r = {}
for name in self.names :
r[name] = self.evaluate(name)
#N = len(r)
#n = sum(r)
#return n/N
return dict(self.getNow(),**r)
"""
This class returns an application's both memory and cpu usage
This class will process a system command and parse the outpout accordingly given a parser
@param parse is a parser pointer
"""
class DetailProcess(Analysis):
def __init__(self):
Analysis.__init__(self)
self.cmd = "ps -eo pid,user,pmem,pcpu,stat,etime,args|awk 'OFS=\";\" {$1=$1; if($5 > 9) print }'"
self.xchar = ';'
def get_app(self,stream):
index = 1 if os.path.exists(" ".join(stream[:1])) else len(stream)-1
def init (self,names):
#Analysis.init(self)
self.names = names;
def getName(self):
return "apps"
def split(self,name,stream):
cmd = " ".join(stream[:index]) if index > 0 else " ".join(stream)
pattern = "(\d+.{0,1}\d*)\x20*(\d+.{0,1}\d*)\x20*(\d+.{0,1}\d*)".replace(":name",name).strip()
g = re.match(pattern,stream.strip())
if g :
return list(g.groups())+['1']+[name]
if ' ' in cmd.split('/')[len(cmd.split('/'))-1] :
p = cmd.split('/')[len(cmd.split('/'))-1].split(' ')
name = p[0]
args = " ".join(p[1:])
else:
return ''
def reboot(self,rows,conf=None) :
return np.sum([int(item['label']=='crash') for item in rows]) > 0
def parse(self,row,fields):
"""
The last field should be the command in its integrity
@pre len(fields) > len(row)
"""
r = {}
now = self.getNow()
r['date'] = now
row = [term for term in row.split() if term.strip() != '']
for name in fields :
index = fields.index(name)
r[name] = row[index] if row else 0
if name not in ['user','cmd','status','pid'] :
r[name] = float(r[name])
r[name] = row[index: ] if row else []
#
# Let's set the status give the data extracted
#
if r['status'] == 0 :
r['status'] = 'crash'
elif 'Z' in r['status'] :
r['status'] = 'zombie'
elif r['memory_usage'] > 0 and r['cpu_usage'] > 0:
r['status'] = 'running'
name = cmd.split('/')[len(cmd.split('/'))-1]
args = " ".join(stream[index:]) if index > 0 else ""
return [name,cmd,args]
def to_pandas(self,m):
"""
This function will convert the output of ps to a data-frame
@param m raw matrix i.e list of values like a csv
"""
m = [item for item in m if len(item) != len (m[0])]
m = "\n".join(m[1:])
df = pd.read_csv(pd.compat.StringIO(m),sep=self.xchar)
df.columns =['pid','user','mem','cpu','status','started','name','cmd','args']
return df
def empty(self,name):
return pd.DataFrame([{"pid":None,"user":None,"mem":0,"cpu":0,"status":"X","started":None,"name":name,"cmd":None,"args":None}])
def parse(self,rows):
m = []
TIME_INDEX = 5
ARGS_INDEX = 6
for item in rows :
if rows.index(item) != 0 :
parts = item.split(self.xchar)
row = parts[:TIME_INDEX]
row.append(' '.join(parts[TIME_INDEX:ARGS_INDEX]))
row += self.get_app(parts[ARGS_INDEX:])
else:
r['status'] = 'idle'
row = item.split(self.xchar)
row = (self.xchar.join(row)).strip()
if len(row.replace(";","")) > 0 :
m.append(row)
return m
def get(self,**args):
"""
This function returns a the output of a command to the calling code that is piped into the class
The output will be stored in a data frame with columns
@
"""
try:
handler = subprocess.Popen(self.cmd,shell=True,stdout=subprocess.PIPE)
stream = handler.communicate()[0]
rows = stream.split('\n')
df = self.to_pandas(self.parse(rows))
r = pd.DataFrame()
if 'filter' in args :
pattern = "|".join(args['filter'])
r = df[df.name.str.contains(pattern)]
for name in args['filter'] :
filter = "name == '"+name+"'"
if r.query(filter).size == 0 :
r = r.append(self.empty(name))
return r
except Exception,e:
print (e)
return None
def evaluate(self,name=None) :
if name is None :
name = ".*"
fields = ["user","pid","memory_usage","cpu_usage","memory_available","status","cmd"]
cmd = "ps -eo user,pid,pmem,pcpu,vsize,stat,command|grep -Ei \":app\"".replace(":app",name)
handler = subprocess.Popen(cmd,shell=True,stdout=subprocess.PIPE)
logs = handler.communicate()[0].split('\n')
logs = [row for row in logs if (row.strip() != '') and ('grep -Ei' in row )== False ]
if len(logs) == 0:
return [dict(self.parse('',fields),**{'label':name}) ]
else :
return [dict(self.parse(row,fields),**{'label':name}) for row in logs if row.strip() != '' and 'grep' not in row and '-Ei' not in row]
def status(self,row):
x = row['memory_usage']
y = row['cpu_usage']
z = row['memory_available']
if z :
if y :
return "running"
return "idle"
else:
return "crash"
#def format(self,row):
# r= {"memory_usage":row[0],"cpu_usage":row[1],"memory_available":row[2]/1000,"proc_count":row[3],"label":self.cleanup(row[4])}
# status = self.status(r)
# r['status'] = status
# return r
def composite(self):
ma = []
for name in self.names:
row = self.evaluate(name)
ma += row
return ma
class Folders(SmartTop):
"""
This class evaluates a list of folders and provides detailed informaiton about age/size of each file
Additionally the the details are summarized in terms of global size, and oldest file.
This class will assess a folder and produce a report in a data-frame that can be later on used for summary statistics
"""
class FileWatch(Analysis):
def __init__(self):
pass
def init(self,folders):
self.folders = folders;
self.cache = []
def getName(self):
return "folders"
def evaluate(self,dir_path,r=[]):
"""
This function will recursively scan a folder and retrieve file sizes and age of the files.
The data will be returned as an array of {size,age,label} items
"""
def _get(self,dir_path,r=[]):
for child in os.listdir(dir_path):
path = os.path.join(dir_path, child)
if os.path.isdir(path):
self.evaluate(path,r)
self._get(path,r)
else:
size = os.path.getsize(path)
@ -277,62 +115,25 @@ class FileWatch(Analysis):
file_date = datetime.datetime.fromtimestamp(file_date)
now = datetime.datetime.now()
age = (now - file_date ).days
r.append({"label":path,"size":size,"age":age,"date":self.getNow()})
return r
def reboot(self,rows,limit) :
return np.sum([ 1 for item in rows if rows['size'] > limit ]) > 0
def composite(self):
out = []
for folder in self.folders :
r = self.evaluate(folder,[])
file_count = len(r)
age_mode = [item[0] for item in stats.mode([item['age'] for item in r])]
size_mode= [item[0] for item in stats.mode([item['size'] for item in r])]
age = {"mode":age_mode,"median":np.median([item['age'] for item in r] ),"mean":np.mean([item['age'] for item in r] ),"var":np.mean([item['age'] for item in r])}
size = {"mode":size_mode,"median":np.median([item['size'] for item in r] ), "mean":np.mean([item['size'] for item in r] ),"var":np.mean([item['size'] for item in r])}
out.append({"label":folder,"date":self.getNow(),"stats":{"age":age,"size":size,"file_count":file_count},"logs":r})
return out
# class Monitor (Thread):
# def __init__(self,pConfig,pWriter,id='processes') :
# Thread.__init__(self)
# self.config = pConfig[id]
# self.writer = pWriter;
# self.logs = []
# self.handler = self.config['class']
# self.mconfig = self.config['config']
name = os.path.basename(path)
r.append({"name":name,"path":path,"size":size,"age":age})
return r
# def stop(self):
# self.keep_running = False
# def run(self):
# r = {}
# self.keep_running = True
# lock = RLock()
# while self.keep_running:
# lock.acquire()
# for label in self.mconfig:
# self.handler.init(self.mconfig[label])
# r = self.handler.composite()
# self.writer.write(label=label,row = r)
# time.sleep(2)
# lock.release()
# self.prune()
# TIME_LAPSE = 60*2
# time.sleep(TIME_LAPSE)
# print "Stopped ..."
# def prune(self) :
# MAX_ENTRIES = 100
# if len(self.logs) > MAX_ENTRIES :
# BEG = len(self.logs) - MAX_SIZE -1
# self.logs = self.logs[BEG:]
def get(self,**args):
# path = args['path']
if isinstance(args['path'],list) == False:
paths = [args['path']]
else:
paths = args['path']
_out = pd.DataFrame()
for path in paths :
if os.path.exists(path) :
#
# If the folder does NOT exists it should not be treated.
#
r = pd.DataFrame(self._get(path))
r = pd.DataFrame([{"name":path,"files":r.shape[0],"age_in_days":r.age.mean(),"size_in_kb":r.size.sum()}])
_out = _out.append(r)
return _out

Loading…
Cancel
Save