from monitor import config
from monitor import parser as parsermodule
from monitor.common import *
+from monitor.const import MINUP
from monitor.model import *
from monitor.wrapper import plc
from monitor.wrapper import plccache
plc.nodeBootState(host, 'reinstall')
node_end_record(host)
+def check_node_and_pcu_status_for(loginbase):
+ """
+ this function checks whether all the nodes and associated pcus for a
+ given site are considered 'good'.
+
+ If so, the function returns True.
+ Otherwise, the function returns False.
+ """
+
+ results = []
+ for node in plccache.plcdb_lb2hn[loginbase]:
+
+ noderec = FindbadNodeRecord.findby_or_create(hostname=node['hostname'])
+ nodehist = HistoryNodeRecord.findby_or_create(hostname=node['hostname'])
+ nodebl = BlacklistRecord.get_by(hostname=node['hostname'])
+ pcuhist = HistoryPCURecord.get_by(plc_pcuid=noderec.plc_pcuid)
+
+ if (nodehist is not None and nodehist.status == 'good' and \
+ ((pcuhist is not None and pcuhist.status == 'good') or (pcuhist is None)) ):
+ if nodebl is None: # no entry in blacklist table
+ results.append(True)
+ elif nodebl is not None and nodebl.expired(): # expired entry in blacklist table
+ results.append(True)
+ else:
+ results.append(False) # entry that is not expired.
+ else:
+ results.append(False)
+
+ try:
+ print "test: %s" % results
+ # NOTE: incase results is empty, reduce does not work on an empty set.
+ return reduce(lambda x,y: x&y, results) and len(results) > MINUP
+ except:
+ return False
+
def main(hostnames, sitenames):
# commands:
i = 1
print "%s %s %s" % (i, sitehist.db.loginbase, sitehist.db.status)
- # determine if there are penalties within the last 30 days?
- # if so, add a 'pause_penalty' action.
- if sitehist.db.message_id != 0 and sitehist.db.message_status == 'open' and \
- sitehist.db.penalty_level > 0 and not found_within(recent_actions, 'pause_penalty', 30):
- # pause escalation
- print "Pausing penalties for %s" % site
- sitehist.pausePenalty()
- else:
+ if sitehist.db.status == 'down':
+ if sitehist.db.penalty_pause and \
+ changed_greaterthan(sitehist.db.penalty_pause_time, 30):
+
+ email_exception("", "clear pause penalty for site: %s" % sitehist.db.loginbase)
+ sitehist.closeTicket()
+ # NOTE: but preserve the penalty status.
+ sitehist.clearPenaltyPause()
- if sitehist.db.status == 'down':
- if not found_within(recent_actions, 'pause_penalty', 30) and \
- not found_within(recent_actions, 'increase_penalty', 7) and \
- changed_greaterthan(sitehist.db.last_changed, 7):
+ if sitehist.db.message_id != 0 and \
+ sitehist.db.message_status == 'open' and \
+ not sitehist.db.penalty_pause:
- # TODO: catch errors
- sitehist.increasePenalty()
- sitehist.applyPenalty()
- sitehist.sendMessage('increase_penalty')
+ email_exception("", "pause penalty for site: %s" % sitehist.db.loginbase)
+ sitehist.setPenaltyPause()
- print "send message for site %s penalty increase" % site
+ if not sitehist.db.penalty_pause and \
+ not found_within(recent_actions, 'increase_penalty', 7) and \
+ changed_greaterthan(sitehist.db.last_changed, 7):
- if sitehist.db.status == 'good':
- # clear penalty
- # NOTE: because 'all clear' should have an indefinite status, we
- # have a boolean value rather than a 'recent action'
- if sitehist.db.penalty_applied:
- # send message that penalties are cleared.
+ # TODO: catch errors
+ sitehist.increasePenalty()
+ sitehist.applyPenalty()
+ sitehist.sendMessage('increase_penalty')
- sitehist.clearPenalty()
- sitehist.applyPenalty()
- sitehist.sendMessage('clear_penalty')
- sitehist.closeTicket()
+ print "send message for site %s penalty increase" % site
- print "send message for site %s penalty cleared" % site
+ if sitehist.db.status == 'good':
+ # clear penalty
+ # NOTE: because 'all clear' should have an indefinite status, we
+ # have a boolean value rather than a 'recent action'
+ if sitehist.db.penalty_applied or sitehist.db.penalty_pause:
+ # send message that penalties are cleared.
+ sitehist.clearPenalty()
+ sitehist.applyPenalty()
+ sitehist.sendMessage('clear_penalty')
+ sitehist.closeTicket()
+
+ print "send message for site %s penalty cleared" % site
+
+ # check all nodes and pcus for this site; if they're all ok,
+ # close the ticket, else leave it open.
+ # NOTE: in the case where a PCU reboots and fails, a message is
+ # sent, but the PCU may appear to be ok according to tests.
+ # NOTE: Also, bootmanager sends messages regarding disks,
+ # configuration, etc. So, the conditions here are 'good'
+ # rather than 'not down' as it is in sitebad.
+ close_ticket = check_node_and_pcu_status_for(site)
+ if close_ticket:
+ sitehist.closeTicket()
site_count = site_count + 1