X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=sliver_vs.py;h=4245e531f116993d801d4e0b8ca8509a607c9ac4;hb=db4118a58cef01c0ad8ac5c76c752c6051e1575b;hp=f75bf0179e9318d1b6b685b8efca960fb7162e28;hpb=956140ef211c9353634c0b4be1c598bf1f3ddfe1;p=nodemanager.git diff --git a/sliver_vs.py b/sliver_vs.py index f75bf01..4245e53 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 @@ -17,18 +20,17 @@ don't have to guess if there is a running process or not. """ import errno +import traceback import os, os.path import time -import commands +from threading import BoundedSemaphore +# the util-vserver-pl module import vserver 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 @@ -47,115 +49,79 @@ class Sliver_VS(accounts.Account, vserver.VServer): SHELL = '/bin/vsh' TYPE = 'sliver.VServer' - _init_disk_info_sem = globalsem + _init_disk_info_sem = BoundedSemaphore() 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/nodemanager') 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/nodemanager') self.keys = '' self.rspec = {} - self.initscript = '' self.slice_id = rec['slice_id'] self.disk_usage_initialized = False - self.initscriptchanged = False + self.initscript = '' + self.enabled = True self.configure(rec) - _root_context_arch=None - @staticmethod - def root_context_arch(): - if not Sliver_VS._root_context_arch: - Sliver_VS._root_context_arch=commands.getoutput("uname -i") - return Sliver_VS._root_context_arch - - @staticmethod - def personality (arch): - personality="linux32" - if arch.find("64")>=0: - personality="linux64" - return personality - @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)) + # added by caglar + # band-aid for short period as old API doesn't have GetSliceFamily function + #return + vref = "planetlab-f8-i386" + + # 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("-") - - 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 os.path.isdir ("/vservers/.vref/%s"% vref): refname = vref - - 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 - # 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.verbose("%s (%s): legacy node - using fallback vrefname 'default'"%(name,vref)) - # for legacy nodes - refname="default" - arch="i386" + (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) + arch='i386' + + 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 Sliver_VS.root_context_arch() != arch: - file('/etc/vservers/%s/personality' % name, 'w').write(Sliver_VS.personality(arch)) - logger.log('%s: set personality to %s'%(name,Sliver_VS.personality(arch))) + if tools.root_context_arch() != arch: + file('/etc/vservers/%s/personality' % name, 'w').write(personality(arch)+"\n") + 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'] @@ -166,59 +132,86 @@ class Sliver_VS(accounts.Account, vserver.VServer): new_initscript = rec['initscript'] if new_initscript != self.initscript: self.initscript = new_initscript - self.initscriptchanged = True + # not used anymore, we always check against the installed script + #self.initscriptchanged = True + self.refresh_slice_vinit() accounts.Account.configure(self, rec) # install ssh keys + # unconditionnally install and enable the generic vinit script + # mimicking chkconfig for enabling the generic vinit script + # this is hardwired for runlevel 3 + def install_and_enable_vinit (self): + vinit_source="/usr/share/NodeManager/sliver-initscripts/vinit" + vinit_script="/vservers/%s/etc/rc.d/init.d/vinit"%self.name + rc3_link="/vservers/%s/etc/rc.d/rc3.d/S99vinit"%self.name + rc3_target="../init.d/vinit" + # install in sliver + body=file(vinit_source).read() + if tools.replace_file_with_string(vinit_script,body,chmod=0755): + logger.log("vsliver_vs: %s: installed generic vinit rc script"%self.name) + # create symlink for runlevel 3 + if not os.path.islink(rc3_link): + try: + logger.log("vsliver_vs: %s: creating runlevel3 symlink %s"%(self.name,rc3_link)) + os.symlink(rc3_target,rc3_link) + except: + logger.log_exc("vsliver_vs: %s: failed to create runlevel3 symlink %s"%rc3_link) + + # this one checks for the existence of the slice initscript + # install or remove the slice inistscript, as instructed by the initscript tag + def refresh_slice_vinit(self): + body=self.initscript + sliver_initscript="/vservers/%s/etc/rc.d/init.d/vinit.slice"%self.name + if tools.replace_file_with_string(sliver_initscript,body,remove_if_empty=True,chmod=0755): + if body: + logger.log("vsliver_vs: %s: Installed new initscript in %s"%(self.name,sliver_initscript)) + else: + logger.log("vsliver_vs: %s: Removed obsolete initscript %s"%(self.name,sliver_initscript)) + def start(self, delay=0): - if self.rspec['enabled'] > 0: - logger.log('%s: starting in %d seconds' % (self.name, delay)) + if self.rspec['enabled'] <= 0: + logger.log('sliver_vs: not starting %s, is not enabled'%self.name) + else: + 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 + # the generic /etc/init.d/vinit script is permanently refreshed, and enabled + self.install_and_enable_vinit() + # if a change has occured in the slice initscript, reflect this in /etc/init.d/vinit.slice + self.refresh_slice_vinit() child_pid = os.fork() if child_pid == 0: - 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) + # VServer.start calls fork() internally, + # so just close the nonstandard fds and fork once to avoid creating zombies tools.close_nonstandard_fds() vserver.VServer.start(self) os._exit(0) - else: + 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) + logger.log('sliver_vs: %s: stopping' % self.name) vserver.VServer.stop(self) - def is_running(self): + def is_running(self): return vserver.VServer.is_running(self) - def set_resources(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 Sliver_VS._init_disk_info_sem.acquire() - logger.log('%s: computing disk usage: beginning' % self.name) + logger.log('sliver_vs: %s: computing disk usage: beginning' % self.name) + # init_disk_info is inherited from VServer 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 @@ -230,41 +223,73 @@ 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'])) + 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' % \ + 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("%s: Setting name to %s" % (self.name, self.slice_id),2) - self.setname(self.slice_id) - + #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)) + try: + vserver_config_path = '/etc/vservers/%s'%self.name + if not os.path.exists (vserver_config_path): + os.makedirs (vserver_config_path) + file('%s/slice_id'%vserver_config_path, 'w').write("%d\n"%self.slice_id) + logger.log("sliver_vs: Recorded slice id %d for slice %s"%(self.slice_id,self.name)) + except IOError,e: + logger.log("sliver_vs: Could not record slice_id for slice %s. Error: %s"%(self.name,str(e))) + except Exception,e: + logger.log_exc("sliver_vs: Error recording slice id: %s"%str(e),name=self.name) + + + 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() @@ -274,6 +299,7 @@ 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()