import bwlimit
import database
-from sets import Set
-
priority = 20
# Defaults
# Set ENABLE to False to setup buckets, but not limit.
ENABLE = True
-datafile = "/var/lib/nodemanager/bwmon.dat"
+DB_FILE = "/var/lib/nodemanager/bwmon.pickle"
try:
sys.path.append("/etc/planetlab")
def sync(nmdbcopy):
"""
- Syncs tc, db, and bwmon.dat. Then, starts new slices, kills old ones, and updates byte accounts for each running slice. Sends emails and caps those that went over their limit.
+ Syncs tc, db, and bwmon.pickle.
+ Then, starts new slices, kills old ones, and updates byte accounts for each running slice.
+ Sends emails and caps those that went over their limit.
"""
# Defaults
- global datafile, \
+ global DB_FILE, \
period, \
default_MaxRate, \
default_Maxi2Rate, \
# All slices
names = []
- # Incase the limits have changed.
+ # In case the limits have changed.
default_MaxRate = int(bwlimit.get_bwcap() / 1000)
default_Maxi2Rate = int(bwlimit.bwmax / 1000)
default_MaxRate = 1000000
try:
- f = open(datafile, "r+")
- logger.log("bwmon: Loading %s" % datafile, 2)
+ f = open(DB_FILE, "r+")
+ logger.log("bwmon: Loading %s" % DB_FILE, 2)
(version, slices, deaddb) = pickle.load(f)
f.close()
# Check version of data file
if version != "$Id$":
- logger.log("bwmon: Not using old version '%s' data file %s" % (version, datafile))
+ logger.log("bwmon: Not using old version '%s' data file %s" % (version, DB_FILE))
raise Exception
except Exception:
version = "$Id$"
logger.log("bwmon: Found %s running HTBs" % kernelhtbs.keys().__len__(), 2)
# The dat file has HTBs for slices, but the HTBs aren't running
- nohtbslices = Set(slices.keys()) - Set(kernelhtbs.keys())
+ nohtbslices = set(slices.keys()) - set(kernelhtbs.keys())
logger.log( "bwmon: Found %s slices in dat but not running." % nohtbslices.__len__(), 2)
# Reset tc counts.
for nohtbslice in nohtbslices:
del slices[nohtbslice]
# The dat file doesnt have HTB for the slice but kern has HTB
- slicesnodat = Set(kernelhtbs.keys()) - Set(slices.keys())
+ slicesnodat = set(kernelhtbs.keys()) - set(slices.keys())
logger.log( "bwmon: Found %s slices with HTBs but not in dat" % slicesnodat.__len__(), 2)
for slicenodat in slicesnodat:
# But slice is running
# Get new slices.
# Slices in GetSlivers but not running HTBs
- newslicesxids = Set(live.keys()) - Set(kernelhtbs.keys())
+ newslicesxids = set(live.keys()) - set(kernelhtbs.keys())
logger.log("bwmon: Found %s new slices" % newslicesxids.__len__(), 2)
# Setup new slices
# aren't instantiated by PLC into the dead dict until
# recording period is over. This is to avoid the case where a slice is dynamically created
# and destroyed then recreated to get around byte limits.
- deadxids = Set(slices.keys()) - Set(live.keys())
+ deadxids = set(slices.keys()) - set(live.keys())
logger.log("bwmon: Found %s dead slices" % (deadxids.__len__() - 2), 2)
for deadxid in deadxids:
if deadxid == root_xid or deadxid == default_xid:
# Update byte counts
slice.update(kernelhtbs[xid], live[xid]['_rspec'])
- logger.log("bwmon: Saving %s slices in %s" % (slices.keys().__len__(),datafile), 2)
- f = open(datafile, "w")
+ logger.log("bwmon: Saving %s slices in %s" % (slices.keys().__len__(),DB_FILE), 2)
+ f = open(DB_FILE, "w")
pickle.dump((version, slices, deaddb), f)
f.close()
import sys
import resource
import glob
+import pickle
import logger
import tools
from plcapi import PLCAPI
import random
-id="$Id$"
-savedargv = sys.argv[:]
-
-# NOTE: modules listed here should also be loaded in this order
-# see the priority set in each module - lower comes first
-known_modules=['net','conf_files', 'sm', 'bwmon']
-
-plugin_path = "/usr/share/NodeManager/plugins"
-
-default_period=600
-default_random=301
-
-parser = optparse.OptionParser()
-parser.add_option('-d', '--daemon', action='store_true', dest='daemon', default=False, help='run daemonized')
-parser.add_option('-s', '--startup', action='store_true', dest='startup', default=False, help='run all sliver startup scripts')
-parser.add_option('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config', help='PLC configuration file')
-parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session', help='API session key (or file)')
-parser.add_option('-p', '--period', action='store', dest='period', default=default_period,
- help='Polling interval (sec) - default %d'%default_period)
-parser.add_option('-r', '--random', action='store', dest='random', default=default_random,
- help='Range for additional random polling interval (sec) -- default %d'%default_random)
-parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False, help='more verbose log')
-parser.add_option('-P', '--path', action='store', dest='path', default=plugin_path, help='Path to plugins directory')
-
-# NOTE: BUG the 'help' for this parser.add_option() wont list plugins from the --path argument
-parser.add_option('-m', '--module', action='store', dest='module', default='', help='run a single module among '+' '.join(known_modules))
-(options, args) = parser.parse_args()
-
-# Deal with plugins directory
-if os.path.exists(options.path):
- sys.path.append(options.path)
- plugins = [ os.path.split(os.path.splitext(x)[0])[1] for x in glob.glob( os.path.join(options.path,'*.py') ) ]
- known_modules += plugins
-
-modules = []
-
-def GetSlivers(config, plc):
- '''Run call backs defined in modules'''
- try:
- logger.log("nodemanager: Syncing w/ PLC")
- # retrieve GetSlivers from PLC
- data = plc.GetSlivers()
- # use the magic 'default' slice to retrieve system-wide defaults
- getPLCDefaults(data, config)
- # tweak the 'vref' attribute from GetSliceFamily
- setSliversVref (data)
- # always dump it for debug purposes
- # used to be done only in verbose; very helpful though, and tedious to obtain,
- # so let's dump this unconditionnally
- logger.log_slivers(data)
- logger.verbose("nodemanager: Sync w/ PLC done")
- except:
- logger.log_exc("nodemanager: failed in GetSlivers")
- # XXX So some modules can at least boostrap.
- logger.log("nodemanager: Can't contact PLC to GetSlivers(). Continuing.")
- data = {}
- # Invoke GetSlivers() functions from the callback modules
- for module in modules:
- logger.verbose('trigerring GetSlivers callback for module %s'%module.__name__)
- try:
- callback = getattr(module, 'GetSlivers')
- callback(data, config, plc)
+
+class NodeManager:
+
+ id="$Id$"
+
+ PLUGIN_PATH = "/usr/share/NodeManager/plugins"
+
+ DB_FILE = "/var/lib/nodemanager/getslivers.pickle"
+
+ # the modules in this directory that need to be run
+ # NOTE: modules listed here will also be loaded in this order
+ # once loaded, they get re-ordered after their priority (lower comes first)
+ # for determining the runtime order
+ core_modules=['net','conf_files', 'sm', 'bwmon']
+
+ default_period=600
+ default_random=301
+ default_priority=100
+
+ def __init__ (self):
+
+ parser = optparse.OptionParser()
+ parser.add_option('-d', '--daemon', action='store_true', dest='daemon', default=False, help='run daemonized')
+ parser.add_option('-s', '--startup', action='store_true', dest='startup', default=False, help='run all sliver startup scripts')
+ parser.add_option('-f', '--config', action='store', dest='config', default='/etc/planetlab/plc_config', help='PLC configuration file')
+ parser.add_option('-k', '--session', action='store', dest='session', default='/etc/planetlab/session', help='API session key (or file)')
+ parser.add_option('-p', '--period', action='store', dest='period', default=NodeManager.default_period,
+ help='Polling interval (sec) - default %d'%NodeManager.default_period)
+ parser.add_option('-r', '--random', action='store', dest='random', default=NodeManager.default_random,
+ help='Range for additional random polling interval (sec) -- default %d'%NodeManager.default_random)
+ parser.add_option('-v', '--verbose', action='store_true', dest='verbose', default=False, help='more verbose log')
+ parser.add_option('-P', '--path', action='store', dest='path', default=NodeManager.PLUGIN_PATH, help='Path to plugins directory')
+
+ # NOTE: BUG the 'help' for this parser.add_option() wont list plugins from the --path argument
+ parser.add_option('-m', '--module', action='store', dest='user_module', default='', help='run a single module')
+ (self.options, args) = parser.parse_args()
+
+ if len(args) != 0:
+ parser.print_help()
+ sys.exit(1)
+
+ # determine the modules to be run
+ self.modules = NodeManager.core_modules
+ # Deal with plugins directory
+ if os.path.exists(self.options.path):
+ sys.path.append(self.options.path)
+ plugins = [ os.path.split(os.path.splitext(x)[0])[1] for x in glob.glob( os.path.join(self.options.path,'*.py') ) ]
+ self.modules += plugins
+ if self.options.user_module:
+ assert self.options.user_module in self.modules
+ self.modules=[self.options.user_module]
+ logger.verbose('nodemanager: Running single module %s'%self.options.user_module)
+
+
+ def GetSlivers(self, config, plc):
+ """Run call backs defined in modules"""
+ try:
+ logger.log("nodemanager: Syncing w/ PLC")
+ # retrieve GetSlivers from PLC
+ data = plc.GetSlivers()
+ # use the magic 'default' slice to retrieve system-wide defaults
+ self.getPLCDefaults(data, config)
+ # tweak the 'vref' attribute from GetSliceFamily
+ self.setSliversVref (data)
+ # log it for debug purposes, no matter what verbose is
+ logger.log_slivers(data)
+ # dump it too, so it can be retrieved later in case of comm. failure
+ self.dumpSlivers(data)
+ logger.verbose("nodemanager: Sync w/ PLC done")
+ last_data=data
except:
- logger.log_exc("nodemanager: GetSlivers failed to run callback for module %r"%module)
-
-
-def getPLCDefaults(data, config):
- '''
- Get PLC wide defaults from _default system slice. Adds them to config class.
- '''
- for slice in data.get('slivers'):
- if slice['name'] == config.PLC_SLICE_PREFIX+"_default":
- attr_dict = {}
- for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value']
- if len(attr_dict):
- logger.verbose("nodemanager: Found default slice overrides.\n %s" % attr_dict)
- config.OVERRIDES = attr_dict
- return
- # NOTE: if an _default slice existed, it would have been found above and
- # the routine would return. Thus, if we've gotten here, then no default
- # slice is bound to this node.
- if 'OVERRIDES' in dir(config): del config.OVERRIDES
-
-
-def setSliversVref (data):
- '''
- Tweak the 'vref' attribute in all slivers based on the 'GetSliceFamily' key
- '''
- # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
- # It is safe to override the attributes with this, as this method has the right logic
- for sliver in data.get('slivers'):
+ logger.log_exc("nodemanager: failed in GetSlivers")
+ # XXX So some modules can at least boostrap.
+ logger.log("nodemanager: Can't contact PLC to GetSlivers(). Continuing.")
+ data = {}
+ # for modules that request it though the 'persistent_data' property
+ last_data=self.loadSlivers()
+ # Invoke GetSlivers() functions from the callback modules
+ for module in self.loaded_modules:
+ logger.verbose('triggering GetSlivers callback for module %s'%module.__name__)
+ try:
+ callback = getattr(module, 'GetSlivers')
+ module_data=data
+ if getattr(module,'persistent_data',False):
+ module_data=last_data
+ callback(data, config, plc)
+ except:
+ logger.log_exc("nodemanager: GetSlivers failed to run callback for module %r"%module)
+
+
+ def getPLCDefaults(self, data, config):
+ """
+ Get PLC wide defaults from _default system slice. Adds them to config class.
+ """
+ for slice in data.get('slivers'):
+ if slice['name'] == config.PLC_SLICE_PREFIX+"_default":
+ attr_dict = {}
+ for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value']
+ if len(attr_dict):
+ logger.verbose("nodemanager: Found default slice overrides.\n %s" % attr_dict)
+ config.OVERRIDES = attr_dict
+ return
+ # NOTE: if an _default slice existed, it would have been found above and
+ # the routine would return. Thus, if we've gotten here, then no default
+ # slice is bound to this node.
+ if 'OVERRIDES' in dir(config): del config.OVERRIDES
+
+
+ def setSliversVref (self, data):
+ """
+ Tweak the 'vref' attribute in all slivers based on the 'GetSliceFamily' key
+ """
+ # GetSlivers exposes the result of GetSliceFamily() as an separate key in data
+ # It is safe to override the attributes with this, as this method has the right logic
+ for sliver in data.get('slivers'):
+ try:
+ slicefamily=sliver.get('GetSliceFamily')
+ for att in sliver['attributes']:
+ if att['tagname']=='vref':
+ att['value']=slicefamily
+ continue
+ sliver['attributes'].append({ 'tagname':'vref','value':slicefamily})
+ except:
+ logger.log_exc("nodemanager: Could not overwrite 'vref' attribute from 'GetSliceFamily'",name=sliver['name'])
+
+ def dumpSlivers (self, slivers):
+ f = open(NodeManager.DB_FILE, "w")
+ logger.log ("nodemanager: saving successfully fetched GetSlivers in %s" % NodeManager.DB_FILE)
+ pickle.dump(slivers, f)
+ f.close()
+
+ def loadSlivers (self):
try:
- slicefamily=sliver.get('GetSliceFamily')
- for att in sliver['attributes']:
- if att['tagname']=='vref':
- att['value']=slicefamily
- continue
- sliver['attributes'].append({ 'tagname':'vref','value':slicefamily})
+ f = open(NodeManager.DB_FILE, "r+")
+ logger.log("nodemanager: restoring latest known GetSlivers from %s" % NodeManager.DB_FILE)
+ slivers = pickle.load(f)
+ f.close()
+ return slivers
except:
- logger.log_exc("nodemanager: Could not overwrite 'vref' attribute from 'GetSliceFamily'",name=sliver['name'])
+ logger.log("Could not restore GetSlivers from %s" % NodeManager.DB_FILE)
+ return {}
-
-def run():
- try:
- if options.daemon: tools.daemon()
-
- # set log level
- if (options.verbose):
- logger.set_level(logger.LOG_VERBOSE)
-
- # Load /etc/planetlab/plc_config
- config = Config(options.config)
-
+ def run(self):
try:
- other_pid = tools.pid_file()
- if other_pid != None:
- print """There might be another instance of the node manager running as pid %d. If this is not the case, please remove the pid file %s""" % (other_pid, tools.PID_FILE)
- return
- except OSError, err:
- print "Warning while writing PID file:", err
-
- # Load and start modules
- if options.module:
- assert options.module in known_modules
- running_modules=[options.module]
- logger.verbose('nodemanager: Running single module %s'%options.module)
- else:
- running_modules=known_modules
- for module in running_modules:
- try:
- m = __import__(module)
- m.start(options, config)
- modules.append(m)
- except ImportError, err:
- print "Warning while loading module %s:" % module, err
-
- default_priority=100
- # sort on priority (lower first)
- def sort_module_priority (m1,m2):
- return getattr(m1,'priority',default_priority) - getattr(m2,'priority',default_priority)
- modules.sort(sort_module_priority)
-
- logger.log('ordered modules:')
- for module in modules: logger.log ('%s: %s'%(getattr(module,'priority',default_priority),module.__name__))
-
- # Load /etc/planetlab/session
- if os.path.exists(options.session):
- session = file(options.session).read().strip()
- else:
- session = None
-
- # Initialize XML-RPC client
- iperiod=int(options.period)
- irandom=int(options.random)
- plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
-
- #check auth
- logger.log("nodemanager: Checking Auth.")
- while plc.check_authentication() != True:
+ if self.options.daemon: tools.daemon()
+
+ # set log level
+ if (self.options.verbose):
+ logger.set_level(logger.LOG_VERBOSE)
+
+ # Load /etc/planetlab/plc_config
+ config = Config(self.options.config)
+
try:
- plc.update_session()
- logger.log("nodemanager: Authentication Failure. Retrying")
- except Exception,e:
- logger.log("nodemanager: Retry Failed. (%r); Waiting.."%e)
- time.sleep(iperiod)
- logger.log("nodemanager: Authentication Succeeded!")
-
-
- while True:
- # Main nodemanager Loop
- logger.log('nodemanager: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom))
- GetSlivers(config, plc)
- delay=iperiod + random.randrange(0,irandom)
- logger.log('nodemanager: mainloop - sleeping for %d s'%delay)
- time.sleep(delay)
- except: logger.log_exc("nodemanager: failed in run")
-
+ other_pid = tools.pid_file()
+ if other_pid != None:
+ print """There might be another instance of the node manager running as pid %d. If this is not the case, please remove the pid file %s""" % (other_pid, tools.PID_FILE)
+ return
+ except OSError, err:
+ print "Warning while writing PID file:", err
+
+ # load modules
+ self.loaded_modules = []
+ for module in self.modules:
+ try:
+ m = __import__(module)
+ m.start(self.options, config)
+ self.loaded_modules.append(m)
+ except ImportError, err:
+ print "Warning while loading module %s:" % module, err
+
+ # sort on priority (lower first)
+ def sort_module_priority (m1,m2):
+ return getattr(m1,'priority',NodeManager.default_priority) - getattr(m2,'priority',NodeManager.default_priority)
+ self.loaded_modules.sort(sort_module_priority)
+
+ logger.log('ordered modules:')
+ for module in self.loaded_modules:
+ logger.log ('%s: %s'%(getattr(module,'priority',NodeManager.default_priority),module.__name__))
+
+ # Load /etc/planetlab/session
+ if os.path.exists(self.options.session):
+ session = file(self.options.session).read().strip()
+ else:
+ session = None
+
+
+ # get random periods
+ iperiod=int(self.options.period)
+ irandom=int(self.options.random)
+
+ # Initialize XML-RPC client
+ plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2)
+
+ #check auth
+ logger.log("nodemanager: Checking Auth.")
+ while plc.check_authentication() != True:
+ try:
+ plc.update_session()
+ logger.log("nodemanager: Authentication Failure. Retrying")
+ except Exception,e:
+ logger.log("nodemanager: Retry Failed. (%r); Waiting.."%e)
+ time.sleep(iperiod)
+ logger.log("nodemanager: Authentication Succeeded!")
+
+
+ while True:
+ # Main nodemanager Loop
+ logger.log('nodemanager: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom))
+ self.GetSlivers(config, plc)
+ delay=iperiod + random.randrange(0,irandom)
+ logger.log('nodemanager: mainloop - sleeping for %d s'%delay)
+ time.sleep(delay)
+ except: logger.log_exc("nodemanager: failed in run")
+def run():
+ logger.log("======================================== Entering nodemanager.py "+NodeManager.id)
+ NodeManager().run()
+
if __name__ == '__main__':
- logger.log("======================================== Entering nodemanager.py "+id)
run()
else:
# This is for debugging purposes. Open a copy of Python and import nm
git://git.onelab.eu / nodemanager.git / commitdiff