X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_vs.py;h=b9ebf1107c3446df3f2a6eaea957e09b1dbef2c5;hb=1783d50ab175efdd5cf0a536cd0075003d26583e;hp=3060b9f5652d5fc0be1a532c7fe4e119bf1ddc20;hpb=be69c6fca3b19f87c483091cdfaa99d465d0c1c1;p=nodemanager.git diff --git a/sliver_vs.py b/sliver_vs.py index 3060b9f..b9ebf11 100644 --- a/sliver_vs.py +++ b/sliver_vs.py @@ -1,3 +1,6 @@ +# $Id$ +# $URL$ + """VServer slivers. There are a couple of tricky things going on here. First, the kernel @@ -17,8 +20,10 @@ don't have to guess if there is a running process or not. """ import errno +import traceback import os, os.path import time + import vserver import accounts @@ -48,84 +53,73 @@ class Sliver_VS(accounts.Account, vserver.VServer): _init_disk_info_sem = globalsem def __init__(self, rec): - logger.verbose ('initing Sliver_VS with name=%s'%rec['name']) + name=rec['name'] + logger.verbose ('sliver_vs: %s init'%name) try: - vserver.VServer.__init__(self, rec['name'],logfile='/var/log/nm') + logger.log("sliver_vs: %s: first chance..."%name) + vserver.VServer.__init__(self, 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') + logger.log_exc("sliver_vs:__init__ (first chance) %s",name=name) + logger.log('sliver_vs: %s: recreating bad vserver' % name) + self.destroy(name) + self.create(name, rec['vref']) + logger.log("sliver_vs: %s: second chance..."%name) + vserver.VServer.__init__(self, 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.enabled = True self.configure(rec) @staticmethod def create(name, vref = None): - logger.verbose('Sliver_VS:create - name=%s'%name) + logger.verbose('sliver_vs: %s: create'%name) if vref is None: - vref='default' + logger.log("sliver_vs: %s: ERROR - no vref attached, this is unexpected"%(name)) + return + # used to look in /etc/planetlab/family, + # now relies on the 'GetSliceFamily' extra attribute in GetSlivers() + # which for legacy is still exposed here as the 'vref' key + + # check the template exists -- there's probably a better way.. + if not os.path.isdir ("/vservers/.vref/%s"%vref): + logger.log ("sliver_vs: %s: ERROR Could not create sliver - vreference image %s not found"%(name,vref)) + return + + # guess arch 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("-") - # from the slice attribute: cut dashes and try to figure the meaning - slice_wishes = vref.split("-") - for wish in slice_wishes: - if wish == "i386" or wish == "x86_64": - arch=wish - elif wish == "f8" or wish == "centos5" : - fcdistro=wish - else: - pldistro=wish - - # rejoin the parts - refname="-".join( (pldistro,fcdistro,arch) ) - - # check the templates exists -- there's probably a better way.. - if not os.path.isdir ("/vservers/.vref/%s"%refname): - logger.verbose("%s (%s) : vref %s not found, using default %s"%( - name,vref,refname,default)) - refname=default - # could check again, but as we have /etc/slicefamily - # there's probably no /vservers/.vref/default - - except IOError: - # have not found slicefamily - logger.verbose("%s (%s): legacy node - using fallback vrefname 'default'"%(name,vref)) - # for legacy nodes - refname="default" + (x,y,arch)=vref.split('-') + # mh, this of course applies when 'vref' is e.g. 'netflow' + # and that's not quite right except: - import traceback - logger.log("%s (%s) : unexpected error follows - using 'default'"%( - name,vref)) - logger.log(traceback.format_exc()) - refname="default" + arch='i386' - logger.log_call('/usr/sbin/vuseradd', '-t', refname, name) - open('/vservers/%s/etc/slicename' % name, 'w').write(name) + def personality (arch): + personality="linux32" + if arch.find("64")>=0: + personality="linux64" + return personality + +# logger.log_call(['/usr/sbin/vuseradd', '-t', vref, name, ], timeout=15*60) + logger.log_call(['/bin/bash','-x','/usr/sbin/vuseradd', '-t', vref, name, ], timeout=15*60) + # export slicename to the slice in /etc/slicename + file('/vservers/%s/etc/slicename' % name, 'w').write(name) + file('/vservers/%s/etc/slicefamily' % name, 'w').write(vref) + # 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('sliver_vs: %s: set personality to %s'%(name,personality(arch))) @staticmethod - def destroy(name): logger.log_call('/usr/sbin/vuserdel', name) + def destroy(name): +# logger.log_call(['/usr/sbin/vuserdel', name, ]) + logger.log_call(['/bin/bash','-x','/usr/sbin/vuserdel', name, ]) def configure(self, rec): new_rspec = rec['_rspec'] @@ -136,53 +130,58 @@ class Sliver_VS(accounts.Account, vserver.VServer): new_initscript = rec['initscript'] if new_initscript != self.initscript: self.initscript = new_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) - self.initscriptchanged = True - except: logger.log_exc(self.name) + 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)) + logger.log('sliver_vs: %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('sliver_vs: %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("sliver_vs: start",name=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: logger.log('%s: not starting, is not enabled' % self.name) - self.initscriptchanged = False + else: + os.waitpid(child_pid, 0) + self.initscriptchanged = False + else: logger.log('sliver_vs: not starting %s, is not enabled'%self.name) def stop(self): - logger.log('%s: stopping' % self.name) + logger.log('sliver_vs: %s: stopping' % self.name) vserver.VServer.stop(self) - def set_resources(self): + def is_running(self): + return vserver.VServer.is_running(self) + + 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)) + logger.log('sliver_vs: %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: beginning' % self.name) Sliver_VS._init_disk_info_sem.acquire() + logger.log('sliver_vs: %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) + logger.log('sliver_vs: %s: computing disk usage: ended' % self.name) self.disk_usage_initialized = True 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) + logger.log_exc('sliver_vs: failed to set max disk usage',name=self.name) # get/set the min/soft/hard values for all of the vserver # related RLIMITS. Note that vserver currently only @@ -194,48 +193,64 @@ class Sliver_VS(accounts.Account, vserver.VServer): 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)' + logger.log('sliver_vs: %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'])) - - if False: # this code was commented out before - # N.B. net_*_rate are in kbps because of XML-RPC maxint - # limitations, convert to bps which is what bwlimit.py expects. - net_limits = (self.rspec['net_min_rate'] * 1000, - self.rspec['net_max_rate'] * 1000, - self.rspec['net_i2_min_rate'] * 1000, - self.rspec['net_i2_max_rate'] * 1000, - self.rspec['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) + logger.log('sliver_vs: %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("sliver_vs: %s: opening %s"%(self.name,path)) + flags = os.O_WRONLY + fd = os.open(path, flags) + logger.log("sliver_vs: %s: writing %s=%s"%(self.name,key,self.rspec[key])) + os.write(fd,self.rspec[key]) + os.close(fd) + except IOError, e: + logger.log("sliver_vs: %s: could not set %s=%s"%(self.name,key,self.rspec[key])) + logger.log("sliver_vs: %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)) + logger.log('sliver_vs: %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)) + logger.log('sliver_vs: %s: setting cpu share to %d' % (self.name, cpu_share)) 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'])) + logger.log('sliver_vs: %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("sliver_vs: %s: Setting name to %s" % (self.name, self.slice_id)) + #self.setname(self.slice_id) + logger.log("sliver_vs: %s: Storing slice id of %s for PlanetFlow" % (self.name, self.slice_id)) + file('/etc/vservers/%s/slice_id' % self.name, 'w').write("%d"%self.slice_id) + + if self.enabled == False: + self.enabled = True + self.start() + if False: # Does not work properly yet. if self.have_limits_changed(): - logger.log('%s: limits have changed --- restarting' % self.name) + logger.log('sliver_vs: %s: limits have changed --- restarting' % self.name) stopcount = 10 while self.is_running() and stopcount > 0: self.stop() @@ -245,6 +260,7 @@ class Sliver_VS(accounts.Account, vserver.VServer): self.start() else: # tell vsh to disable remote login by setting CPULIMIT to 0 - logger.log('%s: disabling remote login' % self.name) + logger.log('sliver_vs: %s: disabling remote login' % self.name) self.set_sched_config(0, 0) + self.enabled = False self.stop()