# QUERY all nodes.
COMON_COTOPURL= "http://summer.cs.princeton.edu/status/tabulator.cgi?" + \
"table=table_nodeview&" + \
- "dumpcols='name,resptime,sshstatus,uptime,lastcotop'&" + \
+ "dumpcols='name,resptime,sshstatus,uptime,lastcotop,cpuspeed,memsize,disksize'&" + \
"formatcsv"
#"formatcsv&" + \
#"select='lastcotop!=0'"
import comon
import threadpool
import syncplcdb
+from nodequery import verify,query_to_dict,node_select
import plc
import auth
def collectPingAndSSH(nodename, cohash):
### RUN PING ######################
ping = soltesz.CMD()
- (oval,eval) = ping.run_noexcept("ping -c 1 -q %s | grep rtt" % nodename)
+ (oval,errval) = ping.run_noexcept("ping -c 1 -q %s | grep rtt" % nodename)
values = {}
else:
values['ping'] = "PING"
- #uptime = soltesz.SSH('root', nodename)
- #(oval,eval) = uptime.run_noexcept("uptime | awk '{print $3,$4}' | tr , ' '")
+ try:
+ for port in [22, 806]:
+ ssh = soltesz.SSH('root', nodename, port)
+
+ (oval, errval) = ssh.run_noexcept2(""" <<\EOF
+ echo "{"
+ echo ' "kernel":"'`uname -a`'",'
+ echo ' "bmlog":"'`ls /tmp/bm.log`'",'
+ echo ' "bootcd":"'`cat /mnt/cdrom/bootme/ID`'",'
+ echo ' "nm":"'`ps ax | grep nm.py | grep -v grep`'",'
+ echo ' "princeton_comon":"'`ls -d /vservers/princeton_comon`'",'
+
+ ID=`grep princeton_comon /etc/passwd | awk -F : '{if ( $3 > 500 ) { print $3}}'`
+
+ echo ' "princeton_comon_running":"'`ls -d /proc/virtual/$ID`'",'
+ echo ' "princeton_comon_procs":"'`vps ax | grep $ID | grep -v grep | wc -l`'",'
+ echo "}"
+EOF """)
+
+ if len(oval) > 0:
+ values.update(eval(oval))
+ values['sshport'] = port
+ break
+ else:
+ values.update({'kernel': "", 'bmlog' : "", 'bootcd' : '', 'nm' :
+ '', 'princeton_comon' : '', 'princeton_comon_running' : '',
+ 'princeton_comon_procs' : '', 'sshport' : None})
+ except:
+ import traceback; print traceback.print_exc()
+ sys.exit(1)
### RUN SSH ######################
b_getbootcd_id = True
- ssh = soltesz.SSH('root', nodename)
- oval = ""
- eval = ""
- (oval, eval) = ssh.run_noexcept('echo `uname -a ; ls /tmp/bm.log`')
- val = oval
+ #ssh = soltesz.SSH('root', nodename)
+ #oval = ""
+ #errval = ""
+ #(oval, errval) = ssh.run_noexcept('echo `uname -a ; ls /tmp/bm.log`')
+
+ oval = values['kernel']
if "2.6.17" in oval or "2.6.2" in oval:
values['ssh'] = 'SSH'
values['category'] = 'ALPHA'
- if "bm.log" in oval:
+ 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'
- if "bm.log" in oval:
+ if "bm.log" in values['bmlog']:
values['state'] = 'DEBUG'
else:
values['state'] = 'BOOT'
- elif "2.4" in oval:
+
+ # NOTE: on 2.6.8 kernels, with 4.2 bootstrapfs, the chroot command fails. I have no idea why.
+ elif "2.4" in oval or "2.6.8" in oval:
b_getbootcd_id = False
values['ssh'] = 'SSH'
values['category'] = 'OLDBOOTCD'
elif oval != "":
values['ssh'] = 'SSH'
values['category'] = 'UNKNOWN'
- if "bm.log" in oval:
+ if "bm.log" in values['bmlog']:
values['state'] = 'DEBUG'
else:
values['state'] = 'BOOT'
values['ssh'] = 'NOSSH'
values['category'] = 'ERROR'
values['state'] = 'DOWN'
- val = eval.strip()
+ val = errval.strip()
+ values['kernel'] = val
- values['kernel'] = val
+ #values['kernel'] = val
if b_getbootcd_id:
# try to get BootCD for all nodes that are not 2.4 nor inaccessible
- (oval, eval) = ssh.run_noexcept('cat /mnt/cdrom/bootme/ID')
- val = oval
- if "BootCD" in val:
- values['bootcd'] = val
- if "v2" in val and \
+ #(oval, errval) = ssh.run_noexcept('cat /mnt/cdrom/bootme/ID')
+ oval = values['bootcd']
+ if "BootCD" in oval:
+ values['bootcd'] = oval
+ if "v2" in oval and \
( nodename is not "planetlab1.cs.unc.edu" and \
nodename is not "planetlab2.cs.unc.edu" ):
values['category'] = 'OLDBOOTCD'
# TODO: get bm.log for debug nodes.
# 'zcat /tmp/bm.log'
+
+ #(oval, errval) = ssh.run_noexcept('ps ax | grep nm.py | grep -v grep')
+ oval = values['nm']
+ if "nm.py" in oval:
+ values['nm'] = "Y"
+ else:
+ values['nm'] = "N"
+
+ continue_slice_check = True
+ #(oval, errval) = ssh.run_noexcept('ls -d /vservers/princeton_comon')
+ oval = values['princeton_comon']
+ if "princeton_comon" in oval:
+ values['princeton_comon'] = "Y"
+ else:
+ values['princeton_comon'] = "N"
+ continue_slice_check = False
+
+ if continue_slice_check:
+ #(oval, errval) = ssh.run_noexcept('ID=`grep princeton_comon /etc/passwd | awk -F : "{if ( \\\$3 > 500 ) { print \\\$3}}"`; ls -d /proc/virtual/$ID')
+ oval = values['princeton_comon_running']
+ if len(oval) > len('/proc/virtual/'):
+ values['princeton_comon_running'] = "Y"
+ else:
+ values['princeton_comon_running'] = "N"
+ continue_slice_check = False
+ else:
+ values['princeton_comon_running'] = "-"
+
+ if continue_slice_check:
+ #(oval, errval) = ssh.run_noexcept('ID=`grep princeton_comon /etc/passwd | awk -F : "{if ( \\\$3 > 500 ) { print \\\$3}}"`; vps ax | grep $ID | grep -v grep | wc -l')
+ oval = values['princeton_comon_procs']
+ values['princeton_comon_procs'] = oval
+ else:
+ values['princeton_comon_procs'] = "-"
+
if nodename in cohash:
values['comonstats'] = cohash[nodename]
count += 1
print "%d %s %s" % (count, nodename, externalState['nodes'][nodename]['values'])
- soltesz.dbDump(config.dbname, externalState)
+ if count % 20 == 0:
+ soltesz.dbDump(config.dbname, externalState)
# this will be called when an exception occurs within a thread
def handle_exception(request, result):
print "All results collected."
break
+ soltesz.dbDump(config.dbname, externalState)
+
def main():
cotop_url = COMON_COTOPURL
# history information for all nodes
+ #cohash = {}
cohash = cotop.coget(cotop_url)
l_nodes = syncplcdb.create_plcdb()
if config.filename:
elif config.nodegroup:
ng = api.GetNodeGroups({'name' : config.nodegroup})
l_nodes = api.GetNodes(ng[0]['node_ids'])
-
+ elif config.site:
+ site = api.GetSites(config.site)
+ l_nodes = api.GetNodes(site[0]['node_ids'], ['hostname'])
+
l_nodes = [node['hostname'] for node in l_nodes]
+ # perform this query after the above options, so that the filter above
+ # does not break.
+ if config.nodeselect:
+ l_nodes = node_select(config.nodeselect)
+
print "fetching %s hosts" % len(l_nodes)
checkAndRecordState(l_nodes, cohash)
from config import config
from optparse import OptionParser
parser = OptionParser()
- parser.set_defaults(filename=None, node=None, nodegroup=None, increment=False, dbname="findbadnodes", cachenodes=False)
+ 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")
parser.add_option("", "--cachenodes", action="store_true",
help="Cache node lookup from PLC")
git://git.onelab.eu / monitor.git / blobdiff