www/database.php

[monitor.git] / findbad.py

diff --git a/findbad.py b/findbad.py
index 5b04398..47459ad 100755 (executable)
--- a/findbad.py
+++ b/findbad.py
@@ -4,6 +4,8 @@ import os
 import sys
 import string
 import time
+import config
+import util.file
 
 
 # QUERY all nodes.
@@ -21,19 +23,19 @@ externalState = {'round': round, 'nodes': {}}
 count = 0
 
 
-import soltesz
+import database
+import moncommands 
 import comon
 import threadpool
 import syncplcdb
 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 ######################
-       ping = soltesz.CMD()
+       ping = moncommands.CMD()
        (oval,errval) = ping.run_noexcept("ping -c 1 -q %s | grep rtt" % nodename)
 
        values = {}
@@ -46,7 +48,7 @@ def collectPingAndSSH(nodename, cohash):
 
        try:
                for port in [22, 806]: 
-                       ssh = soltesz.SSH('root', nodename, port)
+                       ssh = moncommands.SSH('root', nodename, port)
 
                        (oval, errval) = ssh.run_noexcept2(""" <<\EOF
                                echo "{"
@@ -77,7 +79,7 @@ EOF                   """)
 
        ### RUN SSH ######################
        b_getbootcd_id = True
-       #ssh = soltesz.SSH('root', nodename)
+       #ssh = moncommands.SSH('root', nodename)
        #oval = ""
        #errval = ""
        #(oval, errval) = ssh.run_noexcept('echo `uname -a ; ls /tmp/bm.log`')
@@ -266,7 +268,7 @@ def recordPingAndSSH(request, result):
                count += 1
                print "%d %s %s" % (count, nodename, externalState['nodes'][nodename]['values'])
                if count % 20 == 0:
-                       soltesz.dbDump(config.dbname, externalState)
+                       database.dbDump(config.dbname, externalState)
 
 # this will be called when an exception occurs within a thread
 def handle_exception(request, result):
@@ -301,10 +303,16 @@ def checkAndRecordState(l_nodes, cohash):
                        pass
 
        # WAIT while all the work requests are processed.
+       begin = time.time()
        while 1:
                try:
                        time.sleep(1)
                        tp.poll()
+                       # if more than two hours
+                       if time.time() - begin > (60*60*1.5):
+                               print "findbad.py has run out of time!!!!!!"
+                               database.dbDump(config.dbname, externalState)
+                               os._exit(1)
                except KeyboardInterrupt:
                        print "Interrupted!"
                        break
@@ -312,14 +320,14 @@ def checkAndRecordState(l_nodes, cohash):
                        print "All results collected."
                        break
 
-       soltesz.dbDump(config.dbname, externalState)
+       database.dbDump(config.dbname, externalState)
 
 
 
 def main():
        global externalState
 
-       externalState = soltesz.if_cached_else(1, config.dbname, lambda : externalState) 
+       externalState = database.if_cached_else(1, config.dbname, lambda : externalState) 
 
        if config.increment:
                # update global round number to force refreshes across all nodes
@@ -334,8 +342,8 @@ def main():
        #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]
@@ -352,7 +360,8 @@ def main():
        # 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)
 
@@ -362,30 +371,21 @@ def main():
 
 
 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()
@@ -393,5 +393,5 @@ if __name__ == '__main__':
                print traceback.print_exc()
                print "Exception: %s" % err
                print "Saving data... exitting."
-               soltesz.dbDump(config.dbname, externalState)
+               database.dbDump(config.dbname, externalState)
                sys.exit(0)