X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=nm.py;h=abc05bbff85ef7a556ef935274596a8bc533fb2d;hb=74a8cfb9d2eac39ae02e5323e1fb5b1d33297981;hp=83a78f2fa57d69971a2605bb1f0279a2b587fa2a;hpb=b994d0a94352d9c4f3fc430c317f7a37dd3414d4;p=nodemanager.git diff --git a/nm.py b/nm.py index 83a78f2..abc05bb 100755 --- a/nm.py +++ b/nm.py @@ -1,5 +1,7 @@ #!/usr/bin/python - +# +# $Id$ +# $URL$ # # Useful information can be found at https://svn.planet-lab.org/wiki/NodeManager # @@ -28,17 +30,27 @@ import random id="$Id$" savedargv = sys.argv[:] -known_modules=['conf_files', 'sm', 'bwmon', 'vsys', 'codemux'] +# NOTE: modules listed here should also be loaded in this order +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=600, help='Polling interval (sec)') -parser.add_option('-r', '--random', action='store', dest='random', default=301, help='Range for additional random polling interval (sec)') +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='/usr/share/NodeManager/plugins', help='Path to plugins directory') +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() @@ -49,15 +61,23 @@ if os.path.exists(options.path): modules = [] -def GetSlivers(plc, config): +def GetSlivers(config, plc): '''Run call backs defined in modules''' try: - logger.log("Syncing w/ PLC") + logger.log("nm: Syncing w/ PLC") + # retrieve GetSlivers from PLC data = plc.GetSlivers() - if (options.verbose): logger.log_slivers(data) + # 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("nm: Sync w/ PLC done") except: - logger.log_exc() + logger.log_exc("nm: failed in GetSlivers") # XXX So some modules can at least boostrap. logger.log("nm: Can't contact PLC to GetSlivers(). Continuing.") data = {} @@ -65,8 +85,8 @@ def GetSlivers(plc, config): for module in modules: try: callback = getattr(module, 'GetSlivers') - callback(plc, data, config) - except: logger.log_exc() + callback(data, config, plc) + except: logger.log_exc("nm: GetSlivers failed to run callback for module %r"%module) def getPLCDefaults(data, config): @@ -78,12 +98,33 @@ def getPLCDefaults(data, config): attr_dict = {} for attr in slice.get('attributes'): attr_dict[attr['tagname']] = attr['value'] if len(attr_dict): - logger.verbose("Found default slice overrides.\n %s" % attr_dict) + logger.verbose("nm: Found default slice overrides.\n %s" % attr_dict) config.OVERRIDES = attr_dict - return + 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'): + 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("nm: Could not overwrite 'vref' attribute from 'GetSliceFamily'",name=sliver['name']) + + def run(): try: if options.daemon: tools.daemon() @@ -107,7 +148,7 @@ def run(): if options.module: assert options.module in known_modules running_modules=[options.module] - logger.verbose('Running single module %s'%options.module) + logger.verbose('nm: Running single module %s'%options.module) else: running_modules=known_modules for module in running_modules: @@ -129,25 +170,30 @@ def run(): irandom=int(options.random) plc = PLCAPI(config.plc_api_uri, config.cacert, session, timeout=iperiod/2) + #check auth + logger.log("nm: Checking Auth.") + while plc.check_authentication() != True: + try: + plc.update_session() + logger.log("nm: Authentication Failure. Retrying") + except: + logger.log("nm: Retry Failed. Waiting") + time.sleep(iperiod) + logger.log("nm: Authentication Succeeded!") + + while True: # Main NM Loop - logger.verbose('mainloop - nm:getSlivers - period=%d random=%d'%(iperiod,irandom)) - GetSlivers(plc, config) + logger.verbose('nm: mainloop - calling GetSlivers - period=%d random=%d'%(iperiod,irandom)) + GetSlivers(config, plc) delay=iperiod + random.randrange(0,irandom) - logger.verbose('mainloop - sleeping for %d s'%delay) + logger.verbose('nm: mainloop - sleeping for %d s'%delay) time.sleep(delay) - except: logger.log_exc() + except: logger.log_exc("nm: failed in run") if __name__ == '__main__': - logger.log("Entering nm.py "+id) - stacklim = 512*1024 # 0.5 MiB - curlim = resource.getrlimit(resource.RLIMIT_STACK)[0] # soft limit - if curlim > stacklim: - resource.setrlimit(resource.RLIMIT_STACK, (stacklim, stacklim)) - # for some reason, doesn't take effect properly without the exec() - python = '/usr/bin/python' - os.execv(python, [python] + savedargv) + logger.log("======================================== Entering nm.py "+id) run() else: # This is for debugging purposes. Open a copy of Python and import nm