From: Stephen Soltesz <soltesz@cs.princeton.edu>
Date: Thu, 12 Feb 2009 21:19:36 +0000 (+0000)
Subject: added sitelist option for find* scripts.
X-Git-Tag: Monitor-2.0-1~13
X-Git-Url: http://git.onelab.eu/?a=commitdiff_plain;h=4df3989d4aee44cb6e36169ac69ba097436f7b15;p=monitor.git
added sitelist option for find* scripts.
clarified email message in emailZabbix.
---
diff --git a/automate/fetch.py b/automate/fetch.py
index 91b5715..8d01986 100755
--- a/automate/fetch.py
+++ b/automate/fetch.py
@@ -6,7 +6,7 @@ import os
from glob import glob
import vxargs
-import parser as parsermodule
+from monitor import parser as parsermodule
from automate import *
def build_vx_args(shell_cmd):
diff --git a/findbad.py b/findbad.py
index 78ba65e..9b348bb 100755
--- a/findbad.py
+++ b/findbad.py
@@ -122,6 +122,13 @@ def main():
elif config.site:
site = api.GetSites(config.site)
l_nodes = api.GetNodes(site[0]['node_ids'], ['hostname'])
+ elif config.sitelist:
+ site_list = config.sitelist.split(',')
+ sites = api.GetSites(site_list)
+ node_ids = []
+ for s in sites:
+ node_ids += s['node_ids']
+ l_nodes = api.GetNodes(node_ids, ['hostname'])
l_nodes = [node['hostname'] for node in l_nodes]
@@ -172,5 +179,3 @@ if __name__ == '__main__':
print "Saving data... exitting."
sys.exit(0)
print "sleeping"
- #print "final commit"
- #time.sleep(10)
diff --git a/findbadpcu.py b/findbadpcu.py
index 2d27f79..815a77e 100755
--- a/findbadpcu.py
+++ b/findbadpcu.py
@@ -91,9 +91,9 @@ def main():
if_new_set={'round' : global_round})
global_round = fbsync.round
+ api = plc.getAuthAPI()
if config.site is not None:
- api = plc.getAuthAPI()
site = api.GetSites(config.site)
l_nodes = api.GetNodes(site[0]['node_ids'], ['pcu_ids'])
pcus = []
@@ -101,6 +101,21 @@ def main():
pcus += node['pcu_ids']
# clear out dups.
l_pcus = [pcu for pcu in sets.Set(pcus)]
+ elif config.sitelist:
+ site_list = config.sitelist.split(',')
+
+ sites = api.GetSites(site_list)
+ node_ids = []
+ for s in sites:
+ node_ids += s['node_ids']
+
+ l_nodes = api.GetNodes(node_ids, ['pcu_ids'])
+ pcus = []
+ for node in l_nodes:
+ pcus += node['pcu_ids']
+ # clear out dups.
+ l_pcus = [pcu for pcu in sets.Set(pcus)]
+
elif config.pcuselect is not None:
n, pcus = pcu_select(config.pcuselect)
print pcus
diff --git a/monitor/database/zabbixapi/emailZabbix.py b/monitor/database/zabbixapi/emailZabbix.py
index 9f41da3..01f1b8f 100644
--- a/monitor/database/zabbixapi/emailZabbix.py
+++ b/monitor/database/zabbixapi/emailZabbix.py
@@ -123,15 +123,17 @@ You can acknowledge this notice by visiting the link below or by letting us know
Thank you for your help,
-- PlanetLab Central (%(support_email)s)
"""
- node_discovered_subject = """Discovered {HOSTNAME} and online"""
+ node_discovered_subject = """Discovered host {HOSTNAME} w/ ip {IPADDRESS} and online"""
node_discovered = """
Thank you for registering and installing this machine:
- {HOSTNAME}
+ hostname {HOSTNAME} : ip {IPADDRESS}
Our monitoring infrastructure has detected it, and in the future it will alert you in the event of unexpected downtime.
- {TRIGGER.NAME}: {STATUS}
+ Discovered : Online
+ {TRIGGER.URL}
+
Thank you for your help,
-- PlanetLab Central (%(support_email)s)
"""
diff --git a/monitor/parser.py b/monitor/parser.py
index e3365ee..ac26cd6 100644
--- a/monitor/parser.py
+++ b/monitor/parser.py
@@ -86,11 +86,13 @@ def parseSetNodeSets(parser=None):
if parser == None:
parser = OptionParser()
- parser.set_defaults(node=None, site=None, nodelist=None, nodeselect=None, nodegroup=None)
+ parser.set_defaults(node=None, site=None, sitelist=None, nodelist=None, nodeselect=None, nodegroup=None)
parser.add_option("", "--node", dest="node", metavar="hostname",
help="Provide a single node to operate on")
parser.add_option("", "--site", dest="site", metavar="site name",
help="Specify a single site to operate on")
+ parser.add_option("", "--sitelist", dest="sitelist", metavar="site name",
+ help="Specify a list of sites, separated by ','")
parser.add_option("", "--nodegroup", dest="nodegroup", metavar="GroupName",
help="Provide the nodegroup for the list of nodes.")
parser.add_option("", "--nodelist", dest="nodelist", metavar="FILE",
diff --git a/pcucontrol/reboot.py b/pcucontrol/reboot.py
index 035aa01..9d171a2 100755
--- a/pcucontrol/reboot.py
+++ b/pcucontrol/reboot.py
@@ -471,6 +471,8 @@ def reboot_test_new(nodename, values, verbose, dryrun):
# TODO: how to handle the weird, georgetown pcus, the drac faults, and ilo faults
except ExceptionPort, err:
rb_ret = str(err)
+ except NameError, err:
+ rb_ret = str(err)
return rb_ret
@@ -503,3 +505,6 @@ def main():
if __name__ == '__main__':
logger = logging.getLogger("monitor")
main()
+ f = open("/tmp/rebootlog", 'a')
+ f.write("reboot %s\n" % sys.argv)
+ f.close()
diff --git a/sitebad.py b/sitebad.py
index 5132233..f8524f0 100755
--- a/sitebad.py
+++ b/sitebad.py
@@ -19,14 +19,19 @@ from nodequery import verify,query_to_dict,node_select
from monitor.model import *
api = plc.getAuthAPI()
+def main():
+ main2(config)
-def main(config):
+def main2(config):
l_nodes = plccache.l_nodes
l_plcsites = plccache.l_sites
if config.site:
l_sites = [config.site]
+ elif config.sitelist:
+ site_list = config.sitelist.split(',')
+ l_sites = site_list
else:
l_sites = [site['login_base'] for site in l_plcsites]
@@ -108,13 +113,13 @@ if __name__ == '__main__':
parser.add_option("", "--site", dest="site", metavar="login_base",
help="Provide a single site to operate on")
- parser.add_option("", "--sitelist", dest="sitelist", metavar="file.list",
- help="Provide a list of files to operate on")
+ parser.add_option("", "--sitelist", dest="sitelist",
+ help="Provide a list of sites separated by ','")
config = parsermodule.parse_args(parser)
try:
- main(config)
+ main2(config)
except Exception, err:
import traceback
print traceback.print_exc()
diff --git a/web/MonitorWeb/monitorweb/controllers.py b/web/MonitorWeb/monitorweb/controllers.py
index f4832ba..bb0580b 100644
--- a/web/MonitorWeb/monitorweb/controllers.py
+++ b/web/MonitorWeb/monitorweb/controllers.py
@@ -203,26 +203,32 @@ class Root(controllers.RootController):
def nodeaction_handler(self, tg_exceptions=None):
"""Handle any kind of error."""
- refurl = request.headers.get("Referer",link("pcu"))
- print refurl
-
- # TODO: do this more intelligently...
- uri_fields = urllib.splitquery(refurl)
- if uri_fields[1] is not None:
- val = query_to_dict(uri_fields[1])
- if 'pcuid' in val:
- pcuid = val['pcuid']
- elif 'hostname' in val:
- pcuid = FindbadNodeRecord.get_latest_by(hostname=val['hostname']).first().plc_pcuid
+
+ if 'pcuid' in request.params:
+ pcuid = request.params['pcuid']
else:
- pcuid=None
+ refurl = request.headers.get("Referer",link("pcu"))
+ print refurl
+
+ # TODO: do this more intelligently...
+ uri_fields = urllib.splitquery(refurl)
+ if uri_fields[1] is not None:
+ val = query_to_dict(uri_fields[1])
+ if 'pcuid' in val:
+ pcuid = val['pcuid']
+ elif 'hostname' in val:
+ pcuid = FindbadNodeRecord.get_latest_by(hostname=val['hostname']).first().plc_pcuid
+ else:
+ pcuid=None
+ else:
+ pcuid=None
cherry_trail = cherrypy._cputil.get_object_trail()
for i in cherry_trail:
print "trail: ", i
print pcuid
- return self.pcuview(pcuid, **dict(exceptions=tg_exceptions))
+ return self.pcuview(None, pcuid, **dict(exceptions=tg_exceptions))
def nodeaction(self, **data):
for item in data.keys():
@@ -247,7 +253,7 @@ class Root(controllers.RootController):
print "REBOOT: %s" % hostname
ret = reboot.reboot_str(str(hostname))
print ret
- if ret: raise RuntimeError("Error using PCU: " + ret)
+ if ret: raise RuntimeError("Error using PCU: " + str(ret))
flash("Reboot appeared to work. All at most 5 minutes. Run ExternalScan to check current status.")
elif action == "ExternalScan":