import sys
import string
import time
+import config
+import util.file
# QUERY all nodes.
from nodequery import verify,query_to_dict,node_select
import traceback
import plc
-import auth
-api = plc.PLC(auth.auth, auth.plc)
+api = plc.getAuthAPI()
def collectPingAndSSH(nodename, cohash):
### RUN PING ######################
echo ' "bmlog":"'`ls /tmp/bm.log`'",'
echo ' "bootcd":"'`cat /mnt/cdrom/bootme/ID`'",'
echo ' "nm":"'`ps ax | grep nm.py | grep -v grep`'",'
+ echo ' "readonlyfs":"'`touch /var/log/monitor 2>&1 ; touch /vservers/monitor.log 2>&1`'",'
+ echo ' "dns":"'`host boot.planet-lab.org 2>&1`'",'
echo ' "princeton_comon":"'`ls -d /vservers/princeton_comon`'",'
ID=`grep princeton_comon /etc/passwd | awk -F : '{if ( $3 > 500 ) { print $3}}'`
values['sshport'] = port
break
else:
- values.update({'kernel': "", 'bmlog' : "", 'bootcd' : '', 'nm' :
- '', 'princeton_comon' : '', 'princeton_comon_running' : '',
- 'princeton_comon_procs' : '', 'sshport' : None})
+ values.update({'kernel': "", 'bmlog' : "", 'bootcd' : '',
+ 'nm' : '',
+ 'readonlyfs' : '',
+ 'dns' : '',
+ 'princeton_comon' : '',
+ 'princeton_comon_running' : '',
+ 'princeton_comon_procs' : '', 'sshport' : None})
except:
print traceback.print_exc()
+ from nodecommon import email_exception
+ email_exception()
sys.exit(1)
### RUN SSH ######################
oval = values['kernel']
if "2.6.17" in oval or "2.6.2" in oval:
values['ssh'] = 'SSH'
- values['category'] = 'ALPHA'
+ values['category'] = 'PROD'
if "bm.log" in values['bmlog']:
values['state'] = 'DEBUG'
else:
values['state'] = 'BOOT'
elif "2.6.12" in oval or "2.6.10" in oval:
values['ssh'] = 'SSH'
- values['category'] = 'PROD'
+ values['category'] = 'OLDPROD'
if "bm.log" in values['bmlog']:
values['state'] = 'DEBUG'
else:
except:
b_except = True
traceback.print_exc()
+ from nodecommon import email_exception
+ email_exception()
plc_lock.release()
if b_except: return (None, None)
values['pcu'] = "NOPCU"
site_id = d_node[0]['site_id']
last_contact = d_node[0]['last_contact']
- nodegroups = [ i['name'] for i in api.GetNodeGroups(d_node[0]['nodegroup_ids']) ]
+ nodegroups = [ i['groupname'] for i in api.GetNodeGroups(d_node[0]['nodegroup_ids']) ]
values['plcnode'] = {'status' : 'SUCCESS',
'pcu_ids': pcu,
'boot_state' : d_node[0]['boot_state'],
except:
b_except = True
traceback.print_exc()
+ from nodecommon import email_exception
+ email_exception()
plc_lock.release()
if b_except: return (None, None)
#cohash = {}
cohash = cotop.coget(cotop_url)
l_nodes = syncplcdb.create_plcdb()
- if config.filename:
- f_nodes = config.getListFromFile(config.filename)
+ if config.nodelist:
+ f_nodes = util.file.getListFromFile(config.nodelist)
l_nodes = filter(lambda x: x['hostname'] in f_nodes, l_nodes)
elif config.node:
f_nodes = [config.node]
l_nodes = filter(lambda x: x['hostname'] in f_nodes, l_nodes)
elif config.nodegroup:
- ng = api.GetNodeGroups({'name' : config.nodegroup})
+ ng = api.GetNodeGroups({'groupname' : config.nodegroup})
l_nodes = api.GetNodes(ng[0]['node_ids'])
elif config.site:
site = api.GetSites(config.site)
# perform this query after the above options, so that the filter above
# does not break.
if config.nodeselect:
- l_nodes = node_select(config.nodeselect)
+ fb = database.dbLoad("findbad")
+ l_nodes = node_select(config.nodeselect, fb['nodes'].keys(), fb)
print "fetching %s hosts" % len(l_nodes)
if __name__ == '__main__':
- from config import config
- from optparse import OptionParser
- parser = OptionParser()
- parser.set_defaults(filename=None, node=None, site=None, nodeselect=False, nodegroup=None,
- increment=False, dbname="findbadnodes", cachenodes=False)
- parser.add_option("", "--node", dest="node", metavar="hostname",
- help="Provide a single node to operate on")
- parser.add_option("-f", "--nodelist", dest="filename", metavar="FILE",
- help="Provide the input file for the node list")
- parser.add_option("", "--nodeselect", dest="nodeselect", metavar="query string",
- help="Provide a selection string to return a node list.")
- parser.add_option("", "--nodegroup", dest="nodegroup", metavar="FILE",
- help="Provide the nodegroup for the list of nodes.")
- parser.add_option("", "--site", dest="site", metavar="site name",
- help="Specify a site to view node status")
+ import parser as parsermodule
+
+ parser = parsermodule.getParser(['nodesets'])
+ parser.set_defaults( increment=False, dbname="findbad", cachenodes=False)
parser.add_option("", "--cachenodes", action="store_true",
help="Cache node lookup from PLC")
parser.add_option("", "--dbname", dest="dbname", metavar="FILE",
help="Specify the name of the database to which the information is saved")
parser.add_option("-i", "--increment", action="store_true", dest="increment",
help="Increment round number to force refresh or retry")
- config = config(parser)
- config.parse_args()
+
+ parser = parsermodule.getParser(['defaults'], parser)
+
+ cfg = parsermodule.parse_args(parser)
try:
main()
except Exception, err:
print traceback.print_exc()
+ from nodecommon import email_exception
+ email_exception()
print "Exception: %s" % err
print "Saving data... exitting."
database.dbDump(config.dbname, externalState)
git://git.onelab.eu / monitor.git / blobdiff