2 # Thierry Parmentelat <thierry.parmentelat@inria.fr>
3 # Copyright (C) 2010 INRIA
5 # #################### history
7 # see also Substrate.readme
9 # This is a complete rewrite of TestResources/Tracker/Pool
10 # we don't use trackers anymore and just probe/sense the running
11 # boxes to figure out where we are
12 # in order to implement some fairness in the round-robin allocation scheme
13 # we need an indication of the 'age' of each running entity,
14 # hence the 'timestamp-*' steps in TestPlc
16 # this should be much more flexible:
17 # * supports several plc boxes
18 # * supports several qemu guests per host
19 # * no need to worry about tracker being in sync or not
21 # #################### howto use
23 # each site is to write its own LocalSubstrate.py,
24 # (see e.g. LocalSubstrate.inria.py)
25 # LocalSubstrate.py is expected to be in /root on the testmaster box
28 # . the vserver-capable boxes used for hosting myplcs
29 # . and their admissible load (max # of myplcs)
30 # . the pool of DNS-names and IP-addresses available for myplcs
32 # . the kvm-qemu capable boxes to host qemu instances
33 # . and their admissible load (max # of myplcs)
34 # . the pool of DNS-names and IP-addresses available for nodes
36 # #################### implem. note
38 # this model relies on 'sensing' the substrate,
39 # i.e. probing all the boxes for their running instances of vservers and qemu
40 # this is how we get rid of tracker inconsistencies
41 # however there is a 'black hole' between the time where a given address is
42 # allocated and when it actually gets used/pingable
43 # this is why we still need a shared knowledge among running tests
44 # in a file named /root/starting
45 # this is connected to the Pool class
47 # ####################
56 from optparse import OptionParser
59 from TestSsh import TestSsh
60 from TestMapper import TestMapper
62 def header (message,banner=True):
63 if not message: return
64 if banner: print "===============",
68 def timestamp_sort(o1,o2): return o1.timestamp-o2.timestamp
72 # allows to pick an available IP among a pool
73 # input is expressed as a list of tuples (hostname,ip,user_data)
74 # that can be searched iteratively for a free slot
76 # pool = [ (hostname1,user_data1),
77 # (hostname2,user_data2),
78 # (hostname3,user_data2),
79 # (hostname4,user_data4) ]
80 # assuming that ip1 and ip3 are taken (pingable), then we'd get
82 # pool.next_free() -> entry2
83 # pool.next_free() -> entry4
84 # pool.next_free() -> None
85 # that is, even if ip2 is not busy/pingable when the second next_free() is issued
88 def __init__ (self,hostname,userdata):
89 self.hostname=hostname
90 self.userdata=userdata
91 # slot holds 'busy' or 'free' or 'mine' or 'starting' or None
92 # 'mine' is for our own stuff, 'starting' from the concurrent tests
97 return "Pooled %s (%s) -> %s"%(self.hostname,self.userdata, self.status)
100 if self.status==None: return '?'
101 elif self.status=='busy': return '*'
102 elif self.status=='free': return '.'
103 elif self.status=='mine': return 'M'
104 elif self.status=='starting': return 'S'
107 if self.ip: return self.ip
108 ip=socket.gethostbyname(self.hostname)
114 def __init__ (self, tuples,message):
115 self.pool= [ PoolItem (h,u) for (h,u) in tuples ]
119 for i in self.pool: print i.line()
123 for i in self.pool: line += ' ' + i.char()
126 def _item (self, hostname):
128 if i.hostname==hostname: return i
129 raise Exception ("Could not locate hostname %s in pool %s"%(hostname,self.message))
131 def retrieve_userdata (self, hostname):
132 return self._item(hostname).userdata
134 def get_ip (self, hostname):
135 try: return self._item(hostname).get_ip()
136 except: return socket.gethostbyname(hostname)
138 def set_mine (self, hostname):
140 self._item(hostname).status='mine'
142 print 'WARNING: host %s not found in IP pool %s'%(hostname,self.message)
144 def next_free (self):
146 if i.status == 'free':
148 return (i.hostname,i.userdata)
151 # the place were other test instances tell about their not-yet-started
152 # instances, that go undetected through sensing
153 starting='/root/starting'
154 def add_starting (self, name):
155 try: items=[line.strip() for line in file(Pool.starting).readlines()]
157 if not name in items:
158 file(Pool.starting,'a').write(name+'\n')
160 if i.hostname==name: i.status='mine'
162 # we load this after actual sensing;
163 def load_starting (self):
164 try: items=[line.strip() for line in file(Pool.starting).readlines()]
167 if i.hostname in items:
168 if i.status=='free' : i.status='starting'
170 def release_my_starting (self):
173 self.del_starting(i.hostname)
176 def del_starting (self, name):
177 try: items=[line.strip() for line in file(Pool.starting).readlines()]
180 f=file(Pool.starting,'w')
182 if item != name: f.write(item+'\n')
187 for item in self.pool:
188 if item.status is not None:
190 if self.check_ping (item.hostname):
196 print 'Sensing IP pool',self.message,
200 print 'After starting: IP pool'
203 # OS-dependent ping option (support for macos, for convenience)
204 ping_timeout_option = None
205 # returns True when a given hostname/ip responds to ping
206 def check_ping (self,hostname):
207 if not Pool.ping_timeout_option:
208 (status,osname) = commands.getstatusoutput("uname -s")
210 raise Exception, "TestPool: Cannot figure your OS name"
211 if osname == "Linux":
212 Pool.ping_timeout_option="-w"
213 elif osname == "Darwin":
214 Pool.ping_timeout_option="-t"
216 command="ping -c 1 %s 1 %s"%(Pool.ping_timeout_option,hostname)
217 (status,output) = commands.getstatusoutput(command)
218 if status==0: print '*',
224 def __init__ (self,hostname):
225 self.hostname=hostname
226 def short_hostname (self):
227 return self.hostname.split('.')[0]
228 def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
230 self.test_ssh().run("shutdown -r now",message="Rebooting %s"%self.hostname)
232 def run(self,argv,message=None,trash_err=False,dry_run=False):
240 return subprocess.call(argv)
242 return subprocess.call(argv,stderr=file('/dev/null','w'))
244 def run_ssh (self, argv, message, trash_err=False):
245 ssh_argv = self.test_ssh().actual_argv(argv)
246 result=self.run (ssh_argv, message, trash_err)
248 print "WARNING: failed to run %s on %s"%(" ".join(argv),self.hostname)
251 def backquote (self, argv, trash_err=False):
253 result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
255 result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
258 def backquote_ssh (self, argv, trash_err=False):
259 # first probe the ssh link
260 probe_argv=self.test_ssh().actual_argv(['hostname'])
261 hostname=self.backquote ( probe_argv, trash_err=True )
263 print "root@%s unreachable"%self.hostname
266 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
268 ############################################################
270 def __init__ (self, buildname, pid, buildbox):
271 self.buildname=buildname
272 self.buildbox=buildbox
275 def add_pid(self,pid):
276 self.pids.append(pid)
279 return "== %s == (pids=%r)"%(self.buildname,self.pids)
281 class BuildBox (Box):
282 def __init__ (self,hostname):
283 Box.__init__(self,hostname)
284 self.build_instances=[]
286 def add_build (self,buildname,pid):
287 for build in self.build_instances:
288 if build.buildname==buildname:
291 self.build_instances.append(BuildInstance(buildname, pid, self))
294 if not self.build_instances:
295 header ('No build process on %s (%s)'%(self.hostname,self.uptime()))
297 header ("Builds on %s (%s)"%(self.hostname,self.uptime()))
298 for b in self.build_instances:
299 header (b.line(),banner=False)
302 if hasattr(self,'_uptime') and self._uptime: return self._uptime
303 return '*undef* uptime'
305 # inspect box and find currently running builds
306 matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
307 def sense(self,reboot=False,verbose=True):
313 self._uptime=self.backquote_ssh(command,trash_err=True).strip()
314 if not self._uptime: self._uptime='unreachable'
315 pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
317 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
318 ps_lines=self.backquote_ssh (command).split('\n')
319 for line in ps_lines:
320 if not line.strip() or line.find('PID')>=0: continue
321 m=BuildBox.matcher.match(line)
323 date=time.strftime('%Y-%m-%d',time.localtime(time.time()))
324 buildname=m.group('buildname').replace('@DATE@',date)
325 self.add_build (buildname,m.group('pid'))
326 else: header('command %r returned line that failed to match'%command)
328 ############################################################
330 def __init__ (self, vservername, ctxid, plcbox):
331 self.vservername=vservername
337 def set_timestamp (self,timestamp): self.timestamp=timestamp
338 def set_now (self): self.timestamp=int(time.time())
339 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
342 return self.vservername.split('-')[-1]
345 msg="== %s == (ctx=%s)"%(self.vservername,self.ctxid)
346 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
347 else: msg += " *unknown timestamp*"
348 if self.ctxid==0: msg+=" not (yet?) running"
352 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
353 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
354 self.plc_box.forget(self)
357 def __init__ (self, hostname, max_plcs):
358 Box.__init__(self,hostname)
359 self.plc_instances=[]
360 self.max_plcs=max_plcs
362 def add_vserver (self,vservername,ctxid):
363 for plc in self.plc_instances:
364 if plc.vservername==vservername:
365 header("WARNING, duplicate myplc %s running on %s"%\
366 (vservername,self.hostname),banner=False)
368 self.plc_instances.append(PlcInstance(vservername,ctxid,self))
370 def forget (self, plc_instance):
371 self.plc_instances.remove(plc_instance)
373 # fill one slot even though this one is not started yet
374 def add_dummy (self, plcname):
375 dummy=PlcInstance('dummy_'+plcname,0,self)
377 self.plc_instances.append(dummy)
380 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_plcs,self.free_spots(),self.uname())
384 if not self.plc_instances:
385 header ('No vserver running on %s'%(self.line()))
387 header ("Active plc VMs on %s"%self.line())
388 for p in self.plc_instances:
389 header (p.line(),banner=False)
391 def free_spots (self):
392 return self.max_plcs - len(self.plc_instances)
395 if hasattr(self,'_uname') and self._uname: return self._uname
396 return '*undef* uname'
398 def plc_instance_by_vservername (self, vservername):
399 for p in self.plc_instances:
400 if p.vservername==vservername: return p
403 def sense (self, reboot=False, soft=False):
405 # remove mark for all running servers to avoid resurrection
406 stop_command=['rm','-rf','/etc/vservers/*/apps/init/mark']
407 self.run_ssh(stop_command,"Removing all vserver marks on %s"%self.hostname)
412 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers")
415 self._uname=self.backquote_ssh(['uname','-r']).strip()
416 # try to find fullname (vserver_stat truncates to a ridiculously short name)
417 # fetch the contexts for all vservers on that box
418 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
419 context_map=self.backquote_ssh (map_command)
420 # at this point we have a set of lines like
421 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
423 for map_line in context_map.split("\n"):
424 if not map_line: continue
425 [path,xid] = map_line.split(':')
426 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
427 # at this point ctx_id maps context id to vservername
429 command=['vserver-stat']
430 vserver_stat = self.backquote_ssh (command)
431 for vserver_line in vserver_stat.split("\n"):
432 if not vserver_line: continue
433 context=vserver_line.split()[0]
434 if context=="CTX": continue
435 longname=ctx_dict[context]
436 self.add_vserver(longname,context)
437 # print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
440 running_vsnames = [ i.vservername for i in self.plc_instances ]
441 command= ['grep','.']
442 command += ['/vservers/%s/timestamp'%vs for vs in running_vsnames]
443 command += ['/dev/null']
444 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
445 for ts_line in ts_lines:
446 if not ts_line.strip(): continue
447 # expect /vservers/<vservername>/timestamp:<timestamp>
449 (_,__,vservername,tail)=ts_line.split('/')
450 (_,timestamp)=tail.split(':')
451 timestamp=int(timestamp)
452 p=self.plc_instance_by_vservername(vservername)
454 print 'WARNING unattached plc instance',ts_line
455 print 'was expecting to find',vservername,'in',[i.vservername for i in self.plc_instances]
457 p.set_timestamp(timestamp)
458 except: print 'WARNING, could not parse ts line',ts_line
463 ############################################################
465 def __init__ (self, nodename, pid, qemubox):
466 self.nodename=nodename
468 self.qemu_box=qemubox
473 def set_buildname (self,buildname): self.buildname=buildname
474 def set_timestamp (self,timestamp): self.timestamp=timestamp
475 def set_now (self): self.timestamp=int(time.time())
476 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
479 msg = "== %s == (pid=%s)"%(self.nodename,self.pid)
480 if self.buildname: msg += " <--> %s"%self.buildname
481 else: msg += " *unknown build*"
482 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
483 else: msg += " *unknown timestamp*"
484 if self.pid: msg += " pid=%s"%self.pid
485 else: msg += " not (yet?) running"
490 print "cannot kill qemu %s with pid==0"%self.nodename
492 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
493 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
494 self.qemu_box.forget(self)
498 def __init__ (self, hostname, max_qemus):
499 Box.__init__(self,hostname)
500 self.qemu_instances=[]
501 self.max_qemus=max_qemus
503 def add_node (self,nodename,pid):
504 for qemu in self.qemu_instances:
505 if qemu.nodename==nodename:
506 header("WARNING, duplicate qemu %s running on %s"%\
507 (nodename,self.hostname), banner=False)
509 self.qemu_instances.append(QemuInstance(nodename,pid,self))
511 def forget (self, qemu_instance):
512 self.qemu_instances.remove(qemu_instance)
514 # fill one slot even though this one is not started yet
515 def add_dummy (self, nodename):
516 dummy=QemuInstance('dummy_'+nodename,0,self)
518 self.qemu_instances.append(dummy)
521 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_qemus,self.free_spots(),self.driver())
525 if not self.qemu_instances:
526 header ('No qemu process on %s'%(self.line()))
528 header ("Active qemu processes on %s"%(self.line()))
529 for q in self.qemu_instances:
530 header (q.line(),banner=False)
532 def free_spots (self):
533 return self.max_qemus - len(self.qemu_instances)
536 if hasattr(self,'_driver') and self._driver: return self._driver
537 return '*undef* driver'
539 def qemu_instance_by_pid (self,pid):
540 for q in self.qemu_instances:
541 if q.pid==pid: return q
544 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
545 for q in self.qemu_instances:
546 if q.nodename==nodename and q.buildname==buildname:
550 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
551 def sense(self, reboot=False, soft=False):
556 self.run_ssh(box,['pkill','qemu'],"Killing qemu instances")
559 modules=self.backquote_ssh(['lsmod']).split('\n')
560 self._driver='*NO kqemu/kmv_intel MODULE LOADED*'
561 for module in modules:
562 if module.find('kqemu')==0:
563 self._driver='kqemu module loaded'
564 # kvm might be loaded without vkm_intel (we dont have AMD)
565 elif module.find('kvm_intel')==0:
566 self._driver='kvm_intel module loaded'
567 ########## find out running pids
568 pids=self.backquote_ssh(['pgrep','qemu'])
570 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
571 ps_lines = self.backquote_ssh (command).split("\n")
572 for line in ps_lines:
573 if not line.strip() or line.find('PID') >=0 : continue
574 m=QemuBox.matcher.match(line)
575 if m: self.add_node (m.group('nodename'),m.group('pid'))
576 else: header('command %r returned line that failed to match'%command)
577 ########## retrieve alive instances and map to build
579 command=['grep','.','*/*/qemu.pid','/dev/null']
580 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
581 for pid_line in pid_lines:
582 if not pid_line.strip(): continue
583 # expect <build>/<nodename>/qemu.pid:<pid>pid
585 (buildname,nodename,tail)=pid_line.split('/')
586 (_,pid)=tail.split(':')
587 q=self.qemu_instance_by_pid (pid)
589 q.set_buildname(buildname)
590 live_builds.append(buildname)
591 except: print 'WARNING, could not parse pid line',pid_line
592 # retrieve timestamps
593 command= ['grep','.']
594 command += ['%s/*/timestamp'%b for b in live_builds]
595 command += ['/dev/null']
596 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
597 for ts_line in ts_lines:
598 if not ts_line.strip(): continue
599 # expect <build>/<nodename>/timestamp:<timestamp>
601 (buildname,nodename,tail)=ts_line.split('/')
602 nodename=nodename.replace('qemu-','')
603 (_,timestamp)=tail.split(':')
604 timestamp=int(timestamp)
605 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
607 print 'WARNING unattached qemu instance',ts_line,nodename,buildname
609 q.set_timestamp(timestamp)
610 except: print 'WARNING, could not parse ts line',ts_line
612 ############################################################
618 self.options=Options()
619 self.options.dry_run=False
620 self.options.verbose=False
621 self.options.probe=True
622 self.options.soft=True
623 self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
624 self.plc_boxes = [ PlcBox (h,m) for (h,m) in self.plc_boxes_spec ()]
625 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
626 self.all_boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
629 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs")
630 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes")
632 def fqdn (self, hostname):
633 if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
635 def short_hostname (self, hostname):
636 if hostname.find('.')>=0: return hostname.split('.')[0]
639 # return True if actual sensing takes place
640 def sense (self,force=False):
641 if self._sensed and not force: return False
642 print 'Sensing local substrate...',
643 for b in self.all_boxes: b.sense()
648 def add_dummy_plc (self, plc_boxname, plcname):
649 for pb in self.plc_boxes:
650 if pb.hostname==plc_boxname:
651 pb.add_dummy(plcname)
652 def add_dummy_qemu (self, qemu_boxname, qemuname):
653 for qb in self.qemu_boxes:
654 if qb.hostname==qemu_boxname:
655 qb.add_dummy(qemuname)
658 def provision (self,plcs,options):
660 # attach each plc to a plc box and an IP address
661 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
662 # attach each node/qemu to a qemu box with an IP address
663 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
664 # update the SFA spec accordingly
665 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
668 print '* Could not provision this test on current substrate','--',e,'--','exiting'
669 traceback.print_exc()
672 # it is expected that a couple of options like ips_bplc and ips_vplc
673 # are set or unset together
675 def check_options (x,y):
676 if not x and not y: return True
677 return len(x)==len(y)
679 # find an available plc box (or make space)
680 # and a free IP address (using options if present)
681 def provision_plc (self, plc, options):
683 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
685 #### let's find an IP address for that plc
689 # we don't check anything here,
690 # it is the caller's responsability to cleanup and make sure this makes sense
691 plc_boxname = options.ips_bplc.pop()
692 vplc_hostname=options.ips_vplc.pop()
694 if self.sense(): self.list_all()
697 # try to find an available IP
698 self.vplc_pool.sense()
699 couple=self.vplc_pool.next_free()
701 (vplc_hostname,unused)=couple
702 #### we need to find one plc box that still has a slot
704 # use the box that has max free spots for load balancing
705 for pb in self.plc_boxes:
708 plc_boxname=pb.hostname
710 # if there's no available slot in the plc_boxes, or we need a free IP address
711 # make space by killing the oldest running instance
712 if not plc_boxname or not vplc_hostname:
713 # find the oldest of all our instances
714 all_plc_instances=reduce(lambda x, y: x+y,
715 [ pb.plc_instances for pb in self.plc_boxes ],
717 all_plc_instances.sort(timestamp_sort)
719 plc_instance_to_kill=all_plc_instances[0]
722 if not plc_boxname: msg += " PLC boxes are full"
723 if not vplc_hostname: msg += " vplc IP pool exhausted"
724 raise Exception,"Could not make space for a PLC instance:"+msg
725 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
726 freed_vplc_hostname=plc_instance_to_kill.vplcname()
727 message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
729 plc_instance_to_kill.kill()
730 # use this new plcbox if that was the problem
732 plc_boxname=freed_plc_boxname
733 # ditto for the IP address
734 if not vplc_hostname:
735 vplc_hostname=freed_vplc_hostname
736 # record in pool as mine
737 self.vplc_pool.set_mine(vplc_hostname)
740 self.add_dummy_plc(plc_boxname,plc['name'])
741 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
742 self.vplc_pool.add_starting(vplc_hostname)
744 #### compute a helpful vserver name
745 # remove domain in hostname
746 vplc_short = self.short_hostname(vplc_hostname)
747 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_short)
748 plc_name = "%s_%s"%(plc['name'],vplc_short)
750 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
751 (plc['name'],plc_boxname,vplc_hostname,vservername))
753 #### apply in the plc_spec
755 # label=options.personality.replace("linux","")
756 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
757 # 'name':'%s-'+label,
759 'vservername':vservername,
761 'PLC_DB_HOST':vplc_hostname,
762 'PLC_API_HOST':vplc_hostname,
763 'PLC_BOOT_HOST':vplc_hostname,
764 'PLC_WWW_HOST':vplc_hostname,
765 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
766 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
771 # mappers only work on a list of plcs
772 return TestMapper([plc],options).map(mapper)[0]
775 def provision_qemus (self, plc, options):
777 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
779 test_mapper = TestMapper ([plc], options)
780 nodenames = test_mapper.node_names()
782 for nodename in nodenames:
784 if options.ips_vnode:
785 # as above, it's a rerun, take it for granted
786 qemu_boxname=options.ips_bnode.pop()
787 vnode_hostname=options.ips_vnode.pop()
789 if self.sense(): self.list_all()
792 # try to find an available IP
793 self.vnode_pool.sense()
794 couple=self.vnode_pool.next_free()
796 (vnode_hostname,unused)=couple
797 # find a physical box
799 # use the box that has max free spots for load balancing
800 for qb in self.qemu_boxes:
803 qemu_boxname=qb.hostname
805 # if we miss the box or the IP, kill the oldest instance
806 if not qemu_boxname or not vnode_hostname:
807 # find the oldest of all our instances
808 all_qemu_instances=reduce(lambda x, y: x+y,
809 [ qb.qemu_instances for qb in self.qemu_boxes ],
811 all_qemu_instances.sort(timestamp_sort)
813 qemu_instance_to_kill=all_qemu_instances[0]
816 if not qemu_boxname: msg += " QEMU boxes are full"
817 if not vnode_hostname: msg += " vnode IP pool exhausted"
818 raise Exception,"Could not make space for a QEMU instance:"+msg
819 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
820 freed_vnode_hostname=self.short_hostname(qemu_instance_to_kill.nodename)
822 message='killing oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
824 qemu_instance_to_kill.kill()
825 # use these freed resources where needed
827 qemu_boxname=freed_qemu_boxname
828 if not vnode_hostname:
829 vnode_hostname=freed_vnode_hostname
830 self.vnode_pool.set_mine(vnode_hostname)
832 self.add_dummy_qemu (qemu_boxname,nodename)
833 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
834 ip=self.vnode_pool.get_ip (vnode_hostname)
835 self.vnode_pool.add_starting(vnode_hostname)
837 vnode_fqdn = self.fqdn(vnode_hostname)
838 nodemap={'host_box':qemu_boxname,
839 'node_fields:hostname':vnode_fqdn,
840 'interface_fields:ip':ip,
841 'interface_fields:mac':mac,
843 nodemap.update(self.network_settings())
844 maps.append ( (nodename, nodemap) )
846 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
847 (nodename,qemu_boxname,vnode_hostname,mac))
849 return test_mapper.map({'node':maps})[0]
851 def localize_sfa_rspec (self,plc,options):
853 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
854 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
855 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
856 plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
857 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
858 for site in plc['sites']:
859 for node in site['nodes']:
860 plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
863 #################### release:
864 def release (self,options):
865 self.vplc_pool.release_my_starting()
866 self.vnode_pool.release_my_starting()
869 #################### show results for interactive mode
872 for b in self.all_boxes: b.list()
874 def get_box (self,box):
875 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes:
876 if b.short_hostname()==box:
878 print "Could not find box %s"%box
881 def list_box(self,box):
887 # can be run as a utility to manage the local infrastructure
889 parser=OptionParser()
890 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
892 (options,args)=parser.parse_args()
894 self.options.verbose=True