#!/usr/bin/python
-import plc
-import auth
-api = plc.PLC(auth.auth, auth.plc)
import sys
-import database
+from monitor import database
from nodecommon import *
-from policy import Diagnose
+from unified_model import Record
import glob
import os
-from reboot import pcu_name
+import traceback
import time
import re
+import string
-#fb = {}
-fb = database.dbLoad("findbad")
-fbpcu = {}
+from monitor.pcu import reboot
+from monitor.wrapper import plc
+api = plc.getAuthAPI()
+
+from monitor.database import FindbadNodeRecord, FindbadNodeRecordSync
+from monitor import util
+from monitor import config
+
+fb = None
+fbpcu = None
class NoKeyException(Exception): pass
def daysdown_print_nodeinfo(fbnode, hostname):
fbnode['hostname'] = hostname
- fbnode['daysdown'] = Diagnose.getStrDaysDown(fbnode)
- fbnode['intdaysdown'] = Diagnose.getDaysDown(fbnode)
+ fbnode['daysdown'] = Record.getStrDaysDown(fbnode)
+ fbnode['intdaysdown'] = Record.getDaysDown(fbnode)
print "%(intdaysdown)5s %(hostname)-44s | %(state)10.10s | %(daysdown)s" % fbnode
def fb_print_nodeinfo(fbnode, hostname, fields=None):
- fbnode['hostname'] = hostname
- fbnode['checked'] = diff_time(fbnode['checked'])
- if fbnode['bootcd']:
- fbnode['bootcd'] = fbnode['bootcd'].split()[-1]
+ #fbnode['hostname'] = hostname
+ #fbnode['checked'] = diff_time(fbnode['checked'])
+ if fbnode['bootcd_version']:
+ fbnode['bootcd_version'] = fbnode['bootcd_version'].split()[-1]
else:
- fbnode['bootcd'] = "unknown"
+ fbnode['bootcd_version'] = "unknown"
fbnode['pcu'] = color_pcu_state(fbnode)
if not fields:
- if 'ERROR' in fbnode['category']:
- fbnode['kernel'] = ""
+ if ( fbnode['observed_status'] is not None and \
+ 'DOWN' in fbnode['observed_status'] ) or \
+ fbnode['kernel_version'] is None:
+ fbnode['kernel_version'] = ""
+ else:
+ fbnode['kernel_version'] = fbnode['kernel_version'].split()[2]
+
+ if fbnode['plc_node_stats'] is not None:
+ fbnode['boot_state'] = fbnode['plc_node_stats']['boot_state']
else:
- fbnode['kernel'] = fbnode['kernel'].split()[2]
- fbnode['boot_state'] = fbnode['plcnode']['boot_state']
+ fbnode['boot_state'] = "unknown"
+
+ try:
+ if len(fbnode['nodegroups']) > 0:
+ fbnode['category'] = fbnode['nodegroups'][0]
+ except:
+ #print "ERROR!!!!!!!!!!!!!!!!!!!!!"
+ pass
- print "%(hostname)-39s | %(checked)11.11s | %(boot_state)5.5s| %(state)8.8s | %(ssh)5.5s | %(pcu)6.6s | %(bootcd)6.6s | %(category)8.8s | %(kernel)s" % fbnode
+ print "%(hostname)-45s | %(date_checked)11.11s | %(boot_state)5.5s| %(observed_status)8.8s | %(ssh_status)5.5s | %(pcu)6.6s | %(bootcd_version)6.6s | %(kernel_version)s" % fbnode
else:
format = ""
for f in fields:
return con_or_true
+def verifyDBrecord(constraints, record):
+ """
+ constraints is a list of key, value pairs.
+ # [ {... : ...}==AND , ... , ... , ] == OR
+ """
+ def has_key(obj, key):
+ try:
+ x = obj.__getattribute__(key)
+ return True
+ except:
+ return False
+
+ def get_val(obj, key):
+ try:
+ return obj.__getattribute__(key)
+ except:
+ return None
+
+ def get(obj, path):
+ indexes = path.split("/")
+ value = get_val(obj,indexes[0])
+ if value is not None and len(indexes) > 1:
+ for key in indexes[1:]:
+ if key in value:
+ value = value[key]
+ else:
+ raise NoKeyException(key)
+ return value
+
+ #print constraints, record
+
+ con_or_true = False
+ for con in constraints:
+ #print "con: %s" % con
+ if len(con.keys()) == 0:
+ con_and_true = False
+ else:
+ con_and_true = True
+
+ for key in con.keys():
+ #print "looking at key: %s" % key
+ if has_key(record, key):
+ value_re = re.compile(con[key])
+ if type([]) == type(get(record,key)):
+ local_or_true = False
+ for val in get(record,key):
+ local_or_true = local_or_true | (value_re.search(val) is not None)
+ con_and_true = con_and_true & local_or_true
+ else:
+ if get(record,key) is not None:
+ con_and_true = con_and_true & (value_re.search(get(record,key)) is not None)
+ else:
+ print "missing key %s" % key,
+ pass
+
+ con_or_true = con_or_true | con_and_true
+
+ return con_or_true
+
def verify(constraints, data):
"""
constraints is a list of key, value pairs.
#print "looking at key: %s" % key
if key in data:
value_re = re.compile(con[key])
- con_and_true = con_and_true & (value_re.search(data[key]) is not None)
+ if type([]) == type(data[key]):
+ local_or_true = False
+ for val in data[key]:
+ local_or_true = local_or_true | (value_re.search(val) is not None)
+ con_and_true = con_and_true & local_or_true
+ else:
+ if data[key] is not None:
+ con_and_true = con_and_true & (value_re.search(data[key]) is not None)
elif key not in data:
print "missing key %s" % key,
pass
- #print "missing key %s" % key
- #con_and_true = False
con_or_true = con_or_true | con_and_true
return False
def pcu_select(str_query, nodelist=None):
+ global fb
+ global fbpcu
pcunames = []
nodenames = []
if str_query is None: return (nodenames, pcunames)
+ if fb is None:
+ fb = database.dbLoad("findbad")
+ if fbpcu is None:
+ fbpcu = database.dbLoad("findbadpcus")
+
#print str_query
dict_query = query_to_dict(str_query)
#print dict_query
if verify(dict_query, pcuinfo):
nodenames.append(node)
str = "cmdhttps/locfg.pl -s %s -f iloxml/License.xml -u %s -p '%s' | grep MESSAGE" % \
- (pcu_name(pcuinfo), pcuinfo['username'], pcuinfo['password'])
- pcunames.append(str)
+ (reboot.pcu_name(pcuinfo), pcuinfo['username'], pcuinfo['password'])
+ #pcunames.append(str)
+ pcunames.append(pcuinfo['pcu_id'])
return (nodenames, pcunames)
def node_select(str_query, nodelist=None, fbdb=None):
+ global fb
+
hostnames = []
if str_query is None: return hostnames
#print str_query
dict_query = query_to_dict(str_query)
#print dict_query
- global fb
if fbdb is not None:
fb = fbdb
- for node in fb['nodes'].keys():
- if nodelist is not None:
- if node not in nodelist: continue
-
- fb_nodeinfo = fb['nodes'][node]['values']
-
- if fb_nodeinfo == []:
- #print node, "has lost values"
+ for node in nodelist:
+ #if nodelist is not None:
+ # if node not in nodelist: continue
+
+ try:
+ fb_noderec = None
+ fb_noderec = FindbadNodeRecord.query.filter(FindbadNodeRecord.hostname==node).order_by(FindbadNodeRecord.date_checked.desc()).first()
+ except:
+ print traceback.print_exc()
continue
- #sys.exit(1)
- fb_nodeinfo['pcu'] = color_pcu_state(fb_nodeinfo)
- fb_nodeinfo['hostname'] = node
- if 'plcnode' in fb_nodeinfo:
- fb_nodeinfo.update(fb_nodeinfo['plcnode'])
-
- if verify(dict_query, fb_nodeinfo):
- #print node #fb_nodeinfo
- hostnames.append(node)
- else:
- #print "NO MATCH", node
- pass
+
+ if fb_noderec:
+ fb_nodeinfo = fb_noderec.to_dict()
+
+ #fb_nodeinfo['pcu'] = color_pcu_state(fb_nodeinfo)
+ #if 'plcnode' in fb_nodeinfo:
+ # fb_nodeinfo.update(fb_nodeinfo['plcnode'])
+
+ #if verifyDBrecord(dict_query, fb_nodeinfo):
+ if verify(dict_query, fb_nodeinfo):
+ #print node #fb_nodeinfo
+ hostnames.append(node)
+ else:
+ #print "NO MATCH", node
+ pass
return hostnames
global fb
global fbpcu
- from config import config
- from optparse import OptionParser
- parser = OptionParser()
- parser.set_defaults(node=None, fromtime=None, select=None, list=None, pcuselect=None, nodelist=None, daysdown=None, fields=None)
+ from monitor import parser as parsermodule
+ parser = parsermodule.getParser()
+
+ parser.set_defaults(node=None, fromtime=None, select=None, list=None,
+ pcuselect=None, nodelist=None, daysdown=None, fields=None)
parser.add_option("", "--daysdown", dest="daysdown", action="store_true",
help="List the node state and days down...")
parser.add_option("", "--select", dest="select", metavar="key=value",
help="A list of nodes to bring out of debug mode.")
parser.add_option("", "--fromtime", dest="fromtime", metavar="YYYY-MM-DD",
help="Specify a starting date from which to begin the query.")
- config = config(parser)
- config.parse_args()
+
+ parser = parsermodule.getParser(['defaults'], parser)
+ config = parsermodule.parse_args(parser)
if config.fromtime:
path = "archive-pdb"
os.chdir("..")
fb = archive.load(file[:-4])
else:
- fb = database.dbLoad("findbad")
+ #fbnodes = FindbadNodeRecord.select(FindbadNodeRecord.q.hostname, orderBy='date_checked',distinct=True).reversed()
+ #fb = database.dbLoad("findbad")
+ fb = None
fbpcu = database.dbLoad("findbadpcus")
+ reboot.fb = fbpcu
if config.nodelist:
- nodelist = config.getListFromFile(config.nodelist)
+ nodelist = util.file.getListFromFile(config.nodelist)
else:
- nodelist = fb['nodes'].keys()
+ # NOTE: list of nodes should come from findbad db. Otherwise, we
+ # don't know for sure that there's a record in the db..
+ plcnodes = database.dbLoad("l_plcnodes")
+ nodelist = [ node['hostname'] for node in plcnodes ]
+ #nodelist = ['planetlab-1.cs.princeton.edu']
pculist = None
if config.select is not None and config.pcuselect is not None:
- nodelist = node_select(config.select, nodelist)
+ nodelist = node_select(config.select, nodelist, fb)
nodelist, pculist = pcu_select(config.pcuselect, nodelist)
elif config.select is not None:
- nodelist = node_select(config.select, nodelist)
+ nodelist = node_select(config.select, nodelist, fb)
elif config.pcuselect is not None:
nodelist, pculist = pcu_select(config.pcuselect, nodelist)
-
if pculist:
for pcu in pculist:
print pcu
for node in nodelist:
config.node = node
- if node not in fb['nodes']:
+ if node not in nodelist:
continue
- fb_nodeinfo = fb['nodes'][node]['values']
+ try:
+ # Find the most recent record
+ fb_noderec = FindbadNodeRecord.query.filter(FindbadNodeRecord.hostname==node).order_by(FindbadNodeRecord.date_checked.desc()).first()
+ except:
+ print traceback.print_exc()
+ pass #fb_nodeinfo = fb['nodes'][node]['values']
if config.list:
print node
if config.daysdown:
daysdown_print_nodeinfo(fb_nodeinfo, node)
else:
+ fb_nodeinfo = fb_noderec.to_dict()
if config.select:
if config.fields:
fields = config.fields.split(",")