# where to send notifications upon load_starting
self.substrate=substrate
- def list (self):
+ def list (self, verbose=False):
for i in self.pool_items: print i.line()
def line (self):
return
self.build_instances.append(BuildInstance(buildname, pid, self))
- def list(self):
+ def list(self, verbose=False):
if not self.build_instances:
header ('No build process on %s (%s)'%(self.hostname,self.uptime()))
else:
def reboot (self, options):
if not options.soft:
- self.reboot(options)
+ Box.reboot(self,options)
else:
command=['pkill','vbuild']
self.run_ssh(command,"Terminating vbuild processes",dry_run=options.dry_run)
# inspect box and find currently running builds
matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
- matcher_building_vm=re.compile("\s*(?P<pid>[0-9]+).*init-vserver.*-i\s+eth.\s+(?P<buildname>[^\s]+)\s*\Z")
+ matcher_building_vm=re.compile("\s*(?P<pid>[0-9]+).*init-vserver.*\s+(?P<buildname>[^\s]+)\s*\Z")
def sense(self, options):
- print 'b',
+ print 'bb',
self.sense_uptime()
pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
if not pids: return
class PlcLxcInstance (PlcInstance):
# does lxc have a context id of any kind ?
- def __init__ (self, plcbox, lxcname):
+ def __init__ (self, plcbox, lxcname, pid):
PlcInstance.__init__(self, plcbox)
self.lxcname = lxcname
+ self.pid = pid
- def kill (self):
- print "TODO PlcLxcInstance.kill"
+ def vplcname (self):
+ return self.lxcname.split('-')[-1]
+ def buildname (self):
+ return self.lxcname.rsplit('-',2)[0]
def line (self):
- return "TODO PlcLxcInstance.line"
+ msg="== %s =="%(self.vplcname())
+ msg += " [=%s]"%self.lxcname
+ if self.pid==-1: msg+=" not (yet?) running"
+ else: msg+=" (pid=%s)"%self.pid
+ if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
+ else: msg += " *unknown timestamp*"
+ return msg
+
+ def kill (self):
+ command="rsync lxc-driver.sh %s:/root"%self.plc_box.hostname
+ commands.getstatusoutput(command)
+ msg="lxc container stopping %s on %s"%(self.lxcname,self.plc_box.hostname)
+ self.plc_box.run_ssh(['/root/lxc-driver.sh','-c','stop_lxc','-n',self.lxcname],msg)
+ self.plc_box.forget(self)
##########
class PlcBox (Box):
self.plc_instances=[]
self.max_plcs=max_plcs
-class PlcVsBox (PlcBox):
-
- def add_vserver (self,vservername,ctxid):
- for plc in self.plc_instances:
- if plc.vservername==vservername:
- header("WARNING, duplicate myplc %s running on %s"%\
- (vservername,self.hostname),banner=False)
- return
- self.plc_instances.append(PlcVsInstance(self,vservername,ctxid))
-
- def forget (self, plc_instance):
- self.plc_instances.remove(plc_instance)
+ def free_slots (self):
+ return self.max_plcs - len(self.plc_instances)
# fill one slot even though this one is not started yet
def add_dummy (self, plcname):
dummy.set_now()
self.plc_instances.append(dummy)
- def line(self):
- msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_plcs,self.free_slots(),self.uname())
- return msg
-
- def list(self):
+ def forget (self, plc_instance):
+ self.plc_instances.remove(plc_instance)
+
+ def reboot (self, options):
+ if not options.soft:
+ Box.reboot(self,options)
+ else:
+ self.soft_reboot (options)
+
+ def list(self, verbose=False):
if not self.plc_instances:
- header ('No vserver running on %s'%(self.line()))
+ header ('No plc running on %s'%(self.line()))
else:
header ("Active plc VMs on %s"%self.line())
self.plc_instances.sort(timestamp_sort)
for p in self.plc_instances:
header (p.line(),banner=False)
- def free_slots (self):
- return self.max_plcs - len(self.plc_instances)
+ def get_uname(self):
+ self._uname=self.backquote_ssh(['uname','-r']).strip()
+ # expecting sense () to have filled self._uname
def uname(self):
if hasattr(self,'_uname') and self._uname: return self._uname
return '*undef* uname'
+class PlcVsBox (PlcBox):
+
+ def add_vserver (self,vservername,ctxid):
+ for plc in self.plc_instances:
+ if plc.vservername==vservername:
+ header("WARNING, duplicate myplc %s running on %s"%\
+ (vservername,self.hostname),banner=False)
+ return
+ self.plc_instances.append(PlcVsInstance(self,vservername,ctxid))
+
+ def line(self):
+ msg="%s [max=%d,free=%d, VS-based] (%s)"%(self.hostname, self.max_plcs,self.free_slots(),self.uname())
+ return msg
+
def plc_instance_by_vservername (self, vservername):
for p in self.plc_instances:
if p.vservername==vservername: return p
return None
- def reboot (self, options):
- if not options.soft:
- self.reboot(options)
- else:
- self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers",
- dry_run=options.dry_run)
+ def soft_reboot (self, options):
+ self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers on %s"%(self.hostname,),
+ dry_run=options.dry_run)
def sense (self, options):
- print 'p',
- self._uname=self.backquote_ssh(['uname','-r']).strip()
+ print 'vp',
+ self.get_uname()
# try to find fullname (vserver_stat truncates to a ridiculously short name)
# fetch the contexts for all vservers on that box
map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
class PlcLxcBox (PlcBox):
- def add_dummy (self, plcname):
- print "TODO PlcLxcBox.add_dummy"
+ def add_lxc (self,lxcname,pid):
+ for plc in self.plc_instances:
+ if plc.lxcname==lxcname:
+ header("WARNING, duplicate myplc %s running on %s"%\
+ (lxcname,self.hostname),banner=False)
+ return
+ self.plc_instances.append(PlcLxcInstance(self,lxcname,pid))
- def free_slots (self):
- print "TODO PlcLxcBox.free_slots"
- def list (self):
- print "TODO PlcLxcBox.list"
+ # a line describing the box
+ def line(self):
+ msg="%s [max=%d,free=%d, LXC-based] (%s)"%(self.hostname, self.max_plcs,self.free_slots(),self.uname())
+ return msg
+
+ def plc_instance_by_lxcname (self, lxcname):
+ for p in self.plc_instances:
+ if p.lxcname==lxcname: return p
+ return None
+
+ # essentially shutdown all running containers
+ def soft_reboot (self, options):
+ command="rsync lxc-driver.sh %s:/root"%self.hostname
+ commands.getstatusoutput(command)
+ self.run_ssh(['/root/lxc-driver.sh','-c','stop_all'],"Stopping all running lxc containers on %s"%(self.hostname,),
+ dry_run=options.dry_run)
- def reboot (self, options):
- print "TODO PlcLxcBox.reboot"
+ # sense is expected to fill self.plc_instances with PlcLxcInstance's
+ # to describe the currently running VM's
+ # as well as to call self.get_uname() once
def sense (self, options):
- print "TODO PlcLxcBox.sense"
-
+ print "xp",
+ self.get_uname()
+ command="rsync lxc-driver.sh %s:/root"%self.hostname
+ commands.getstatusoutput(command)
+ command=['/root/lxc-driver.sh','-c','sense_all']
+ lxc_stat = self.backquote_ssh (command)
+ for lxc_line in lxc_stat.split("\n"):
+ if not lxc_line: continue
+ lxcname=lxc_line.split(";")[0]
+ pid=lxc_line.split(";")[1]
+ timestamp=lxc_line.split(";")[2]
+ self.add_lxc(lxcname,pid)
+ timestamp=int(timestamp)
+ p=self.plc_instance_by_lxcname(lxcname)
+ if not p:
+ print 'WARNING zombie plc',self.hostname,lxcname
+ print '... was expecting',lxcname,'in',[i.lxcname for i in self.plc_instances]
+ continue
+ p.set_timestamp(timestamp)
############################################################
class QemuInstance:
self.qemu_instances.append(dummy)
def line (self):
- msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_qemus,self.free_slots(),self.driver())
+ msg="%s [max=%d,free=%d] (%s)"%(self.hostname, self.max_qemus,self.free_slots(),self.driver())
return msg
- def list(self):
+ def list(self, verbose=False):
if not self.qemu_instances:
header ('No qemu process on %s'%(self.line()))
else:
def reboot (self, options):
if not options.soft:
- self.reboot(options)
+ Box.reboot(self,options)
else:
self.run_ssh(['pkill','qemu'],"Killing qemu instances",
dry_run=options.dry_run)
matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
def sense(self, options):
- print 'q',
+ print 'qn',
modules=self.backquote_ssh(['lsmod']).split('\n')
- self._driver='*NO kqemu/kmv_intel MODULE LOADED*'
+ self._driver='*NO kqemu/kvm_intel MODULE LOADED*'
for module in modules:
if module.find('kqemu')==0:
self._driver='kqemu module loaded'
- # kvm might be loaded without vkm_intel (we dont have AMD)
+ # kvm might be loaded without kvm_intel (we dont have AMD)
elif module.find('kvm_intel')==0:
self._driver='kvm_intel module loaded'
########## find out running pids
header(">>%s<<"%line)
########## retrieve alive instances and map to build
live_builds=[]
- command=['grep','.','*/*/qemu.pid','/dev/null']
+ command=['grep','.','/vservers/*/*/qemu.pid','/dev/null']
pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
for pid_line in pid_lines:
if not pid_line.strip(): continue
# expect <build>/<nodename>/qemu.pid:<pid>pid
try:
- (buildname,nodename,tail)=pid_line.split('/')
+ (_,__,buildname,nodename,tail)=pid_line.split('/')
(_,pid)=tail.split(':')
q=self.qemu_instance_by_pid (pid)
if not q: continue
# retrieve timestamps
if not live_builds: return
command= ['grep','.']
- command += ['%s/*/timestamp'%b for b in live_builds]
+ command += ['/vservers/%s/*/timestamp'%b for b in live_builds]
command += ['/dev/null']
ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
for ts_line in ts_lines:
if not ts_line.strip(): continue
# expect <build>/<nodename>/timestamp:<timestamp>
try:
- (buildname,nodename,tail)=ts_line.split('/')
+ (_,__,buildname,nodename,tail)=ts_line.split('/')
nodename=nodename.replace('qemu-','')
(_,timestamp)=tail.split(':')
timestamp=int(timestamp)
def set_now (self): self.timestamp=int(time.time())
def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
+ def is_running (self): return len(self.pids) != 0
def add_pid (self,pid):
self.pids.append(pid)
matcher_proc=re.compile (".*/proc/(?P<pid>[0-9]+)/cwd.*/root/(?P<buildname>[^/]+)$")
matcher_grep=re.compile ("/root/(?P<buildname>[^/]+)/logs/trace.*:TRACE:\s*(?P<plcindex>[0-9]+).*step=(?P<step>\S+).*")
def sense (self, options):
- print 't',
+ print 'tm',
self.sense_uptime()
self.starting_ips=[x for x in self.backquote_ssh(['cat',Starting.location], trash_err=True).strip().split('\n') if x]
def line (self):
return "%s (%s)"%(self.hostname,self.uptime())
- def list (self):
- if not self.test_instances:
- header ("No known tests on %s"%self.line())
+ def list (self, verbose=False):
+ # verbose shows all tests
+ if verbose:
+ instances = self.test_instances
+ msg="knwown tests"
else:
- header ("Known tests on %s"%self.line())
- self.test_instances.sort(timestamp_sort)
- for i in self.test_instances: print i.line()
+ instances = [ i for i in self.test_instances if i.is_running() ]
+ msg="known running tests"
+
+ if not instances:
+ header ("No %s on %s"%(msg,self.line()))
+ else:
+ header ("%s on %s"%(msg,self.line()))
+ instances.sort(timestamp_sort)
+ for i in instances: print i.line()
+ # show 'starting' regardless of verbose
if self.starting_ips:
header ("Starting IP addresses on %s"%self.line())
self.starting_ips.sort()
for starting in self.starting_ips: print starting
+ else:
+ header ("Empty 'starting' on %s"%self.line())
############################################################
class Options: pass
class Substrate:
- def __init__ (self, use_plc_vs_boxes=True, use_plc_lxc_boxes=False):
+ def __init__ (self, plcs_on_vs=True, plcs_on_lxc=False):
self.options=Options()
self.options.dry_run=False
self.options.verbose=False
self.options.soft=False
self.test_box = TestBox (self.test_box_spec())
self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
- self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_vs_boxes_spec ()]
- self.plc_lxc_boxes = [ PlcLxcBox (h,m) for (h,m) in self.plc_lxc_boxes_spec ()]
+ # for compat with older LocalSubstrate
+ try:
+ self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_vs_boxes_spec ()]
+ self.plc_lxc_boxes = [ PlcLxcBox (h,m) for (h,m) in self.plc_lxc_boxes_spec ()]
+ except:
+ self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_boxes_spec ()]
+ self.plc_lxc_boxes = [ ]
self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
self._sensed=False
self.vplc_pool = Pool (self.vplc_ips(),"for vplcs",self)
self.vnode_pool = Pool (self.vnode_ips(),"for vnodes",self)
- self.rescope (use_plc_vs_boxes, use_plc_lxc_boxes)
+ self.rescope (plcs_on_vs=plcs_on_vs, plcs_on_lxc=plcs_on_lxc)
- def rescope(self, plcs_on_vs, plcs_on_lxc):
+ # which plc boxes are we interested in ?
+ def rescope (self, plcs_on_vs, plcs_on_lxc):
self.plc_boxes=[]
if plcs_on_vs: self.plc_boxes += self.plc_vs_boxes
if plcs_on_lxc: self.plc_boxes += self.plc_lxc_boxes
self.default_boxes = self.plc_boxes + self.qemu_boxes
self.all_boxes = self.build_boxes + [ self.test_box ] + self.plc_boxes + self.qemu_boxes
+ def summary_line (self):
+ msg = "["
+ msg += " %d vp"%len(self.plc_vs_boxes)
+ msg += " %d xp"%len(self.plc_lxc_boxes)
+ msg += " %d tried plc boxes"%len(self.plc_boxes)
+ msg += "]"
+ return msg
+
def fqdn (self, hostname):
if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
return hostname
self._sensed=True
return True
- def list (self):
+ def list (self, verbose=False):
for b in self.default_boxes:
b.list()
except:
msg=""
if not plc_boxname: msg += " PLC boxes are full"
- if not vplc_hostname: msg += " vplc IP pool exhausted"
- raise Exception,"Could not make space for a PLC instance:"+msg
+ if not vplc_hostname: msg += " vplc IP pool exhausted"
+ msg += " %s"%self.summary_line()
+ raise Exception,"Cannot make space for a PLC instance:"+msg
freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
freed_vplc_hostname=plc_instance_to_kill.vplcname()
message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
msg=""
if not qemu_boxname: msg += " QEMU boxes are full"
if not vnode_hostname: msg += " vnode IP pool exhausted"
- raise Exception,"Could not make space for a QEMU instance:"+msg
+ msg += " %s"%self.summary_line()
+ raise Exception,"Cannot make space for a QEMU instance:"+msg
freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
freed_vnode_hostname=short_hostname(qemu_instance_to_kill.nodename)
# kill it
nodemap={'host_box':qemu_boxname,
'node_fields:hostname':vnode_fqdn,
'interface_fields:ip':ip,
+ 'ipaddress_fields:ip_addr':ip,
'interface_fields:mac':mac,
}
nodemap.update(self.network_settings())
for box in box_or_names:
if not isinstance(box,Box): box=self.get_box(box)
if not box: continue
- box.list()
+ box.list(self.options.verbose)
def reboot_boxes(self,box_or_names):
for box in box_or_names:
box.reboot(self.options)
####################
- # can be run as a utility to manage the local infrastructure
+ # can be run as a utility to probe/display/manage the local infrastructure
def main (self):
parser=OptionParser()
parser.add_option ('-r',"--reboot",action='store_true',dest='reboot',default=False,
help='add build boxes')
parser.add_option ('-p',"--plc",action='store_true',dest='plcs',default=False,
help='add plc boxes')
- parser.add_option ('-X', "--lxc",action='store_true',dest='plcs_use_lxc',
- help='use lxc-enabled plc boxes instead of vs-enabled ones')
parser.add_option ('-q',"--qemu",action='store_true',dest='qemus',default=False,
help='add qemu boxes')
parser.add_option ('-a',"--all",action='store_true',dest='all',default=False,
help='dry run mode')
(self.options,args)=parser.parse_args()
- if self.options.plcs_use_lxc:
- self.rescope (plcs_on_vs=False, plcs_on_lxc=True)
+ self.rescope (plcs_on_vs=True, plcs_on_lxc=True)
boxes=args
if self.options.testbox: boxes += [self.test_box]
if self.options.qemus: boxes += self.qemu_boxes
if self.options.all: boxes += self.all_boxes
- # default scope is -b -p -q
+ # default scope is -b -p -q -t
if not boxes:
- boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
+ boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes + [self.test_box]
if self.options.reboot: self.reboot_boxes (boxes)
else: self.list_boxes (boxes)
git://git.onelab.eu / tests.git / blobdiff