X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_vs.py;h=5573984f89c93edc7330a518a5bd39912e37c00b;hb=refs%2Fheads%2F1.7;hp=4292d1d75243d798ea203c89e343e11cdb920a52;hpb=dfbec103d5234340d11f454c70c82891e5ac9344;p=nodemanager.git diff --git a/sliver_vs.py b/sliver_vs.py index 4292d1d..5573984 100644 --- a/sliver_vs.py +++ b/sliver_vs.py @@ -17,121 +17,269 @@ don't have to guess if there is a running process or not. """ import errno -import os +import traceback +import os, os.path +import time + import vserver -from bwlimit import bwmin, bwmax import accounts import logger import tools +from threading import BoundedSemaphore + +globalsem = BoundedSemaphore() +# special constant that tells vserver to keep its existing settings +KEEP_LIMIT = vserver.VC_LIM_KEEP -DEFAULTS = {'disk_max': 5000000, - 'net_min': bwmin, - 'net_max': bwmax, - 'net2_min': bwmin, - 'net2_max': bwmax, - 'net_share': 1, - 'enabled': 1, - 'cpu_min': 0, - 'cpu_share': 32, - 'keys': '', - 'initscript': ''} +# populate the sliver/vserver specific default allocations table, +# which is used to look for slice attributes +DEFAULT_ALLOCATION = {} +for rlimit in vserver.RLIMITS.keys(): + rlim = rlimit.lower() + DEFAULT_ALLOCATION["%s_min"%rlim]=KEEP_LIMIT + DEFAULT_ALLOCATION["%s_soft"%rlim]=KEEP_LIMIT + DEFAULT_ALLOCATION["%s_hard"%rlim]=KEEP_LIMIT -class Sliver_VS(vserver.VServer): - """This class wraps vserver.VServer to make its interface closer to what we need for the Node Manager.""" +class Sliver_VS(accounts.Account, vserver.VServer): + """This class wraps vserver.VServer to make its interface closer to what we need.""" SHELL = '/bin/vsh' TYPE = 'sliver.VServer' + _init_disk_info_sem = globalsem - def __init__(self, name): - vserver.VServer.__init__(self, name) - self.current_keys = '' - self.current_initscript = '' - self.disk_usage_initialized = False - self.rec = DEFAULTS.copy() + def __init__(self, rec): + logger.verbose ('initing Sliver_VS with name=%s'%rec['name']) + try: + vserver.VServer.__init__(self, rec['name'],logfile='/var/log/nm') + except Exception, err: + if not isinstance(err, vserver.NoSuchVServer): + # Probably a bad vserver or vserver configuration file + logger.log_exc(self.name) + logger.log('%s: recreating bad vserver' % rec['name']) + self.destroy(rec['name']) + self.create(rec['name'], rec['vref']) + vserver.VServer.__init__(self, rec['name'],logfile='/var/log/nm') + self.keys = '' + self.rspec = {} + self.initscript = '' + self.slice_id = rec['slice_id'] + self.disk_usage_initialized = False + self.initscriptchanged = False + self.configure(rec) @staticmethod - def create(name): logger.log_call('/usr/sbin/vuseradd', name) + def create(name, vref = None): + logger.verbose('Sliver_VS:create - name=%s'%name) + if vref is None: + vref='default' + try: + ### locating the right slicefamily + # this is a first draft, and more a proof of concept thing + # the idea is to parse vref for dash-separated wishes, + # and to project these against the defaults + # so e.g. if the default slice family (as found in /etc/planetlab/slicefamily) + # is planetlab-f8-i386, then here is what we get + # vref=x86_64 -> vuseradd -t planetlab-f8-x86_64 + # vref=centos5 -> vuseradd -t planetlab-centos5-i386 + # vref=centos5-onelab -> vuseradd -t onelab-centos5-i386 + # vref=planetflow -> vuseradd -t planetflow-f8-i386 + # vref=x86_64-planetflow -> vuseradd -t planetflow-f8-x86_64 + + # default + default=file("/etc/planetlab/slicefamily").read().strip() + (pldistro,fcdistro,arch) = default.split("-") + + known_archs = [ 'i386', 'x86_64' ] + known_fcdistros = [ 'f8', 'f9', 'centos5' ] + # from the slice attribute: cut dashes and try to figure the meaning + slice_wishes = vref.split("-") + for wish in slice_wishes: + if wish in known_archs: + arch=wish + elif wish in known_fcdistros: + fcdistro=wish + else: + pldistro=wish + + # rejoin the parts + refname="-".join( (pldistro,fcdistro,arch) ) + + # check the template exists -- there's probably a better way.. + if not os.path.isdir ("/vservers/.vref/%s"%refname): + logger.log("%s (%s) : vref %s not found, using default %s"%( + name,vref,refname,default)) + refname=default + # reset so arch is right + (pldistro,fcdistro,arch) = default.split("-") + # could check again, but as we have /etc/slicefamily + # there's probably no /vservers/.vref/default + + except IOError: + # have not found slicefamily + logger.log("%s (%s): legacy node - using fallback vrefname 'default'"%(name,vref)) + # for legacy nodes + refname="default" + arch="i386" + except: + logger.log("%s (%s) : unexpected error follows - using 'default'"%(name,vref)) + logger.log(traceback.format_exc()) + refname="default" + arch="i386" + + def personality (arch): + personality="linux32" + if arch.find("64")>=0: + personality="linux64" + return personality + + logger.log_call('/usr/sbin/vuseradd', '-t', refname, name) + # export slicename to the slice in /etc/slicename + file('/vservers/%s/etc/slicename' % name, 'w').write(name) + # set personality: only if needed (if arch's differ) + if tools.root_context_arch() != arch: + file('/etc/vservers/%s/personality' % name, 'w').write(personality(arch)) + logger.log('%s: set personality to %s'%(name,personality(arch))) @staticmethod def destroy(name): logger.log_call('/usr/sbin/vuserdel', name) - def configure(self, rec): - self.rec = DEFAULTS.copy() - self.rec.update(rec) - - self.set_resources() - - new_keys = self.rec['keys'] - if new_keys != self.current_keys: - accounts.install_keys(rec) - self.current_keys = new_keys - - new_initscript = self.rec['initscript'] - if new_initscript != self.current_initscript: - logger.log('%s: installing initscript' % self.name) - def install_initscript(): - flags = os.O_WRONLY | os.O_CREAT | os.O_TRUNC - fd = os.open('/etc/rc.vinit', flags, 0755) - os.write(fd, new_initscript) - os.close(fd) - try: self.chroot_call(install_initscript) - except: logger.log_exc() - self.current_initscript = new_initscript - - - def start(self): - if self.rec['enabled']: - logger.log('%s: starting' % self.name) + new_rspec = rec['_rspec'] + if new_rspec != self.rspec: + self.rspec = new_rspec + self.set_resources() + + new_initscript = rec['initscript'] + if new_initscript != self.initscript: + self.initscript = new_initscript + self.initscriptchanged = True + + accounts.Account.configure(self, rec) # install ssh keys + + def start(self, delay=0): + if self.rspec['enabled'] > 0: + logger.log('%s: starting in %d seconds' % (self.name, delay)) + time.sleep(delay) + # VServer.start calls fork() internally, + # so just close the nonstandard fds and fork once to avoid creating zombies child_pid = os.fork() if child_pid == 0: - # VServer.start calls fork() internally, so just close the nonstandard fds and fork once to avoid creating zombies + if self.initscriptchanged: + logger.log('%s: installing initscript' % self.name) + def install_initscript(): + flags = os.O_WRONLY | os.O_CREAT | os.O_TRUNC + fd = os.open('/etc/rc.vinit', flags, 0755) + os.write(fd, self.initscript) + os.close(fd) + try: + self.chroot_call(install_initscript) + except: logger.log_exc(self.name) tools.close_nonstandard_fds() - vserver.VServer.start(self, True) + vserver.VServer.start(self) os._exit(0) - else: os.waitpid(child_pid, 0) + else: + os.waitpid(child_pid, 0) + self.initscriptchanged = False else: logger.log('%s: not starting, is not enabled' % self.name) def stop(self): logger.log('%s: stopping' % self.name) vserver.VServer.stop(self) + def is_running(self): + return vserver.VServer.is_running(self) - def set_resources(self): - disk_max = int(self.rec['disk_max']) + def set_resources(self,setup=False): + disk_max = self.rspec['disk_max'] logger.log('%s: setting max disk usage to %d KiB' % (self.name, disk_max)) try: # if the sliver is over quota, .set_disk_limit will throw an exception if not self.disk_usage_initialized: self.vm_running = False - logger.log('%s: computing disk usage' % self.name) - self.init_disk_info() + Sliver_VS._init_disk_info_sem.acquire() + logger.log('%s: computing disk usage: beginning' % self.name) + try: self.init_disk_info() + finally: Sliver_VS._init_disk_info_sem.release() + logger.log('%s: computing disk usage: ended' % self.name) self.disk_usage_initialized = True - vserver.VServer.set_disklimit(self, disk_max_KiB) - except OSError: logger.log_exc() - - net_limits = (int(self.rec['net_min']), - int(self.rec['net_max']), - int(self.rec['net2_min']), - int(self.rec['net2_max']), - int(self.rec['net_share'])) - logger.log('%s: setting net limits to %s bps' % (self.name, net_limits[:-1])) - logger.log('%s: setting net share to %d' % (self.name, net_limits[-1])) - self.set_bwlimit(*net_limits) - - cpu_min = int(self.rec['cpu_min']) - cpu_share = int(self.rec['cpu_share']) - if bool(self.rec['enabled']): - if cpu_min > 0: - logger.log('%s: setting cpu share to %d%% guaranteed' % (self.name, cpu_min/10.0)) - self.set_sched_config(cpu_min, vserver.SCHED_CPU_GUARANTEED) + vserver.VServer.set_disklimit(self, max(disk_max, self.disk_blocks)) + except: + logger.log('%s: failed to set max disk usage' % self.name) + logger.log_exc(self.name) + + # get/set the min/soft/hard values for all of the vserver + # related RLIMITS. Note that vserver currently only + # implements support for hard limits. + for limit in vserver.RLIMITS.keys(): + type = limit.lower() + minimum = self.rspec['%s_min'%type] + soft = self.rspec['%s_soft'%type] + hard = self.rspec['%s_hard'%type] + update = self.set_rlimit(limit, hard, soft, minimum) + if update: + logger.log('%s: setting rlimit %s to (%d, %d, %d)' + % (self.name, type, hard, soft, minimum)) + + self.set_capabilities_config(self.rspec['capabilities']) + if self.rspec['capabilities']: + logger.log('%s: setting capabilities to %s' % (self.name, self.rspec['capabilities'])) + + cpu_pct = self.rspec['cpu_pct'] + cpu_share = self.rspec['cpu_share'] + + if setup: + for key in self.rspec.keys(): + if key.find('sysctl.') == 0: + sysctl=key.split('.') + try: + path="/proc/sys/%s" % ("/".join(sysctl[1:])) + logger.log("%s: opening %s"%(self.name,path)) + flags = os.O_WRONLY + fd = os.open(path, flags) + logger.log("%s: writing %s=%s"%(self.name,key,self.rspec[key])) + os.write(fd,self.rspec[key]) + os.close(fd) + except IOError, e: + logger.log("%s: could not set %s=%s"%(self.name,key,self.rspec[key])) + logger.log("%s: error = %s"%(self.name,e)) + + + if self.rspec['enabled'] > 0: + if cpu_pct > 0: + logger.log('%s: setting cpu reservation to %d%%' % (self.name, cpu_pct)) else: + cpu_pct = 0 + + if cpu_share > 0: logger.log('%s: setting cpu share to %d' % (self.name, cpu_share)) - self.set_sched_config(cpu_share, 0) - else: - # tell vsh to disable remote login by setting CPULIMIT to 0 + else: + cpu_share = 0 + + self.set_sched_config(cpu_pct, cpu_share) + # if IP address isn't set (even to 0.0.0.0), sliver won't be able to use network + if self.rspec['ip_addresses'] != '0.0.0.0': + logger.log('%s: setting IP address(es) to %s' % \ + (self.name, self.rspec['ip_addresses'])) + self.set_ipaddresses_config(self.rspec['ip_addresses']) + + if self.is_running(): + logger.log("%s: Setting name to %s" % (self.name, self.slice_id),2) + self.setname(self.slice_id) + + if False: # Does not work properly yet. + if self.have_limits_changed(): + logger.log('%s: limits have changed --- restarting' % self.name) + stopcount = 10 + while self.is_running() and stopcount > 0: + self.stop() + delay = 1 + time.sleep(delay) + stopcount = stopcount - 1 + self.start() + + else: # tell vsh to disable remote login by setting CPULIMIT to 0 logger.log('%s: disabling remote login' % self.name) self.set_sched_config(0, 0) self.stop()