2 # Thierry Parmentelat <thierry.parmentelat@inria.fr>
3 # Copyright (C) 2010 INRIA
5 # #################### history
7 # This is a complete rewrite of TestResources/Tracker/Pool
8 # we don't use trackers anymore and just probe/sense the running
9 # boxes to figure out where we are
10 # in order to implement some fairness in the round-robin allocation scheme
11 # we need an indication of the 'age' of each running entity,
12 # hence the 'timestamp-*' steps in TestPlc
14 # this should be much more flexible:
15 # * supports several plc boxes
16 # * supports several qemu guests per host
17 # * no need to worry about tracker being in sync or not
19 # #################### howto use
21 # each site is to write its own LocalSubstrate.py,
22 # (see e.g. LocalSubstrate.inria.py)
23 # LocalSubstrate.py is expected to be in /root on the testmaster box
26 # . the vserver-capable boxes used for hosting myplcs
27 # . and their admissible load (max # of myplcs)
28 # . the pool of DNS-names and IP-addresses available for myplcs
30 # . the kvm-qemu capable boxes to host qemu instances
31 # . and their admissible load (max # of myplcs)
32 # . the pool of DNS-names and IP-addresses available for nodes
34 # #################### implem. note
36 # this model relies on 'sensing' the substrate,
37 # i.e. probing all the boxes for their running instances of vservers and qemu
38 # this is how we get rid of tracker inconsistencies
39 # however there is a 'black hole' between the time where a given address is
40 # allocated and when it actually gets used/pingable
41 # this is why we still need a shared knowledge among running tests
42 # in a file named /root/starting
43 # this is connected to the Pool class
45 # ####################
54 from optparse import OptionParser
57 from TestSsh import TestSsh
58 from TestMapper import TestMapper
60 def header (message,banner=True):
61 if not message: return
62 if banner: print "===============",
66 def timestamp_sort(o1,o2): return o1.timestamp-o2.timestamp
70 # allows to pick an available IP among a pool
71 # input is expressed as a list of tuples (hostname,ip,user_data)
72 # that can be searched iteratively for a free slot
74 # pool = [ (hostname1,user_data1),
75 # (hostname2,user_data2),
76 # (hostname3,user_data2),
77 # (hostname4,user_data4) ]
78 # assuming that ip1 and ip3 are taken (pingable), then we'd get
80 # pool.next_free() -> entry2
81 # pool.next_free() -> entry4
82 # pool.next_free() -> None
83 # that is, even if ip2 is not busy/pingable when the second next_free() is issued
86 def __init__ (self,hostname,userdata):
87 self.hostname=hostname
88 self.userdata=userdata
89 # slot holds 'busy' or 'free' or 'mine' or 'starting' or None
90 # 'mine' is for our own stuff, 'starting' from the concurrent tests
95 return "Pooled %s (%s) -> %s"%(self.hostname,self.userdata, self.status)
98 if self.status==None: return '?'
99 elif self.status=='busy': return '*'
100 elif self.status=='free': return '.'
101 elif self.status=='mine': return 'M'
102 elif self.status=='starting': return 'S'
105 if self.ip: return self.ip
106 ip=socket.gethostbyname(self.hostname)
112 def __init__ (self, tuples,message):
113 self.pool= [ PoolItem (h,u) for (h,u) in tuples ]
117 for i in self.pool: print i.line()
121 for i in self.pool: line += ' ' + i.char()
124 def _item (self, hostname):
126 if i.hostname==hostname: return i
127 raise Exception ("Could not locate hostname %s in pool %s"%(hostname,self.message))
129 def retrieve_userdata (self, hostname):
130 return self._item(hostname).userdata
132 def get_ip (self, hostname):
133 try: return self._item(hostname).get_ip()
134 except: return socket.gethostbyname(hostname)
136 def set_mine (self, hostname):
138 self._item(hostname).status='mine'
140 print 'WARNING: host %s not found in IP pool %s'%(hostname,self.message)
142 def next_free (self):
144 if i.status == 'free':
146 return (i.hostname,i.userdata)
149 # the place were other test instances tell about their not-yet-started
150 # instances, that go undetected through sensing
151 starting='/root/starting'
152 def add_starting (self, name):
153 try: items=[line.strip() for line in file(Pool.starting).readlines()]
155 if not name in items:
156 file(Pool.starting,'a').write(name+'\n')
158 if i.hostname==name: i.status='mine'
160 # we load this after actual sensing;
161 def load_starting (self):
162 try: items=[line.strip() for line in file(Pool.starting).readlines()]
165 if i.hostname in items:
166 if i.status=='free' : i.status='starting'
168 def release_my_starting (self):
171 self.del_starting(i.hostname)
174 def del_starting (self, name):
175 try: items=[line.strip() for line in file(Pool.starting).readlines()]
178 f=file(Pool.starting,'w')
180 if item != name: f.write(item+'\n')
185 for item in self.pool:
186 if item.status is not None:
188 if self.check_ping (item.hostname):
194 print 'Sensing IP pool',self.message,
198 print 'After starting: IP pool'
201 # OS-dependent ping option (support for macos, for convenience)
202 ping_timeout_option = None
203 # returns True when a given hostname/ip responds to ping
204 def check_ping (self,hostname):
205 if not Pool.ping_timeout_option:
206 (status,osname) = commands.getstatusoutput("uname -s")
208 raise Exception, "TestPool: Cannot figure your OS name"
209 if osname == "Linux":
210 Pool.ping_timeout_option="-w"
211 elif osname == "Darwin":
212 Pool.ping_timeout_option="-t"
214 command="ping -c 1 %s 1 %s"%(Pool.ping_timeout_option,hostname)
215 (status,output) = commands.getstatusoutput(command)
216 if status==0: print '*',
222 def __init__ (self,hostname):
223 self.hostname=hostname
224 def short_hostname (self):
225 return self.hostname.split('.')[0]
226 def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
228 self.test_ssh().run("shutdown -r now",message="Rebooting %s"%self.hostname)
230 def run(self,argv,message=None,trash_err=False,dry_run=False):
238 return subprocess.call(argv)
240 return subprocess.call(argv,stderr=file('/dev/null','w'))
242 def run_ssh (self, argv, message, trash_err=False):
243 ssh_argv = self.test_ssh().actual_argv(argv)
244 result=self.run (ssh_argv, message, trash_err)
246 print "WARNING: failed to run %s on %s"%(" ".join(argv),self.hostname)
249 def backquote (self, argv, trash_err=False):
251 result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
253 result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
256 def backquote_ssh (self, argv, trash_err=False):
257 # first probe the ssh link
258 probe_argv=self.test_ssh().actual_argv(['hostname'])
259 hostname=self.backquote ( probe_argv, trash_err=True )
261 print "root@%s unreachable"%self.hostname
264 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
266 ############################################################
268 def __init__ (self, buildname, pid, buildbox):
269 self.buildname=buildname
270 self.buildbox=buildbox
273 def add_pid(self,pid):
274 self.pids.append(pid)
277 return "== %s == (pids=%r)"%(self.buildname,self.pids)
279 class BuildBox (Box):
280 def __init__ (self,hostname):
281 Box.__init__(self,hostname)
282 self.build_instances=[]
284 def add_build (self,buildname,pid):
285 for build in self.build_instances:
286 if build.buildname==buildname:
289 self.build_instances.append(BuildInstance(buildname, pid, self))
292 if not self.build_instances:
293 header ('No build process on %s (%s)'%(self.hostname,self.uptime()))
295 header ("Builds on %s (%s)"%(self.hostname,self.uptime()))
296 for b in self.build_instances:
297 header (b.line(),banner=False)
300 if hasattr(self,'_uptime') and self._uptime: return self._uptime
301 return '*undef* uptime'
303 # inspect box and find currently running builds
304 matcher_exclude=re.compile(".*builds\.sh.*")
305 matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
306 def sense(self,reboot=False,verbose=True):
312 self._uptime=self.backquote_ssh(command,trash_err=True).strip()
313 if not self._uptime: self._uptime='unreachable'
314 pids=self.backquote_ssh(['pgrep','build'],trash_err=True)
316 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
317 ps_lines=self.backquote_ssh (command).split('\n')
318 for line in ps_lines:
319 if not line.strip() or line.find('PID')>=0: continue
320 if BuildBox.matcher_exclude.match(line): 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())
634 def short_hostname (self, hostname):
635 if hostname.find('.')>=0: return hostname.split('.')[0]
637 # return True if actual sensing takes place
638 def sense (self,force=False):
639 if self._sensed and not force: return False
640 print 'Sensing local substrate...',
641 for b in self.all_boxes: b.sense()
646 def add_dummy_plc (self, plc_boxname, plcname):
647 for pb in self.plc_boxes:
648 if pb.hostname==plc_boxname:
649 pb.add_dummy(plcname)
650 def add_dummy_qemu (self, qemu_boxname, qemuname):
651 for qb in self.qemu_boxes:
652 if qb.hostname==qemu_boxname:
653 qb.add_dummy(qemuname)
656 def provision (self,plcs,options):
658 # attach each plc to a plc box and an IP address
659 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
660 # attach each node/qemu to a qemu box with an IP address
661 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
662 # update the SFA spec accordingly
663 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
666 print '* Could not provision this test on current substrate','--',e,'--','exiting'
667 traceback.print_exc()
670 # it is expected that a couple of options like ips_bplc and ips_vplc
671 # are set or unset together
673 def check_options (x,y):
674 if not x and not y: return True
675 return len(x)==len(y)
677 # find an available plc box (or make space)
678 # and a free IP address (using options if present)
679 def provision_plc (self, plc, options):
681 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
683 #### let's find an IP address for that plc
687 # we don't check anything here,
688 # it is the caller's responsability to cleanup and make sure this makes sense
689 plc_boxname = options.ips_bplc.pop()
690 vplc_hostname=options.ips_vplc.pop()
692 if self.sense(): self.list_all()
695 # try to find an available IP
696 self.vplc_pool.sense()
697 couple=self.vplc_pool.next_free()
699 (vplc_hostname,unused)=couple
700 #### we need to find one plc box that still has a slot
702 # use the box that has max free spots for load balancing
703 for pb in self.plc_boxes:
706 plc_boxname=pb.hostname
708 # if there's no available slot in the plc_boxes, or we need a free IP address
709 # make space by killing the oldest running instance
710 if not plc_boxname or not vplc_hostname:
711 # find the oldest of all our instances
712 all_plc_instances=reduce(lambda x, y: x+y,
713 [ pb.plc_instances for pb in self.plc_boxes ],
715 all_plc_instances.sort(timestamp_sort)
717 plc_instance_to_kill=all_plc_instances[0]
720 if not plc_boxname: msg += " PLC boxes are full"
721 if not vplc_hostname: msg += " vplc IP pool exhausted"
722 raise Exception,"Could not make space for a PLC instance:"+msg
723 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
724 freed_vplc_hostname=plc_instance_to_kill.vplcname()
725 message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
727 plc_instance_to_kill.kill()
728 # use this new plcbox if that was the problem
730 plc_boxname=freed_plc_boxname
731 # ditto for the IP address
732 if not vplc_hostname:
733 vplc_hostname=freed_vplc_hostname
734 # record in pool as mine
735 self.vplc_pool.set_mine(vplc_hostname)
738 self.add_dummy_plc(plc_boxname,plc['name'])
739 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
740 self.vplc_pool.add_starting(vplc_hostname)
742 #### compute a helpful vserver name
743 # remove domain in hostname
744 vplc_short = self.short_hostname(vplc_hostname)
745 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_short)
746 plc_name = "%s_%s"%(plc['name'],vplc_short)
748 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
749 (plc['name'],plc_boxname,vplc_hostname,vservername))
751 #### apply in the plc_spec
753 # label=options.personality.replace("linux","")
754 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
755 # 'name':'%s-'+label,
757 'vservername':vservername,
759 'PLC_DB_HOST':vplc_hostname,
760 'PLC_API_HOST':vplc_hostname,
761 'PLC_BOOT_HOST':vplc_hostname,
762 'PLC_WWW_HOST':vplc_hostname,
763 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
764 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
769 # mappers only work on a list of plcs
770 return TestMapper([plc],options).map(mapper)[0]
773 def provision_qemus (self, plc, options):
775 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
777 test_mapper = TestMapper ([plc], options)
778 nodenames = test_mapper.node_names()
780 for nodename in nodenames:
782 if options.ips_vnode:
783 # as above, it's a rerun, take it for granted
784 qemu_boxname=options.ips_bnode.pop()
785 vnode_hostname=options.ips_vnode.pop()
787 if self.sense(): self.list_all()
790 # try to find an available IP
791 self.vnode_pool.sense()
792 couple=self.vnode_pool.next_free()
794 (vnode_hostname,unused)=couple
795 # find a physical box
797 # use the box that has max free spots for load balancing
798 for qb in self.qemu_boxes:
801 qemu_boxname=qb.hostname
803 # if we miss the box or the IP, kill the oldest instance
804 if not qemu_boxname or not vnode_hostname:
805 # find the oldest of all our instances
806 all_qemu_instances=reduce(lambda x, y: x+y,
807 [ qb.qemu_instances for qb in self.qemu_boxes ],
809 all_qemu_instances.sort(timestamp_sort)
811 qemu_instance_to_kill=all_qemu_instances[0]
814 if not qemu_boxname: msg += " QEMU boxes are full"
815 if not vnode_hostname: msg += " vnode IP pool exhausted"
816 raise Exception,"Could not make space for a QEMU instance:"+msg
817 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
818 freed_vnode_hostname=self.short_hostname(qemu_instance_to_kill.nodename)
820 message='killing oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
822 qemu_instance_to_kill.kill()
823 # use these freed resources where needed
825 qemu_boxname=freed_qemu_boxname
826 if not vnode_hostname:
827 vnode_hostname=freed_vnode_hostname
828 self.vnode_pool.set_mine(vnode_hostname)
830 self.add_dummy_qemu (qemu_boxname,nodename)
831 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
832 ip=self.vnode_pool.get_ip (vnode_hostname)
833 self.vnode_pool.add_starting(vnode_hostname)
835 vnode_fqdn = self.fqdn(vnode_hostname)
836 nodemap={'host_box':qemu_boxname,
837 'node_fields:hostname':vnode_fqdn,
838 'interface_fields:ip':ip,
839 'interface_fields:mac':mac,
841 nodemap.update(self.network_settings())
842 maps.append ( (nodename, nodemap) )
844 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
845 (nodename,qemu_boxname,vnode_hostname,mac))
847 return test_mapper.map({'node':maps})[0]
849 def localize_sfa_rspec (self,plc,options):
851 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
852 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
853 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
854 plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
855 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
856 for site in plc['sites']:
857 for node in site['nodes']:
858 plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
861 #################### release:
862 def release (self,options):
863 self.vplc_pool.release_my_starting()
864 self.vnode_pool.release_my_starting()
867 #################### show results for interactive mode
870 for b in self.all_boxes: b.list()
872 def get_box (self,box):
873 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes:
874 if b.short_hostname()==box:
876 print "Could not find box %s"%box
879 def list_box(self,box):
885 # can be run as a utility to manage the local infrastructure
887 parser=OptionParser()
888 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
890 (options,args)=parser.parse_args()
892 self.options.verbose=True