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):
67 if not o1.timestamp: return -1
68 elif not o2.timestamp: return 1
69 else: return o2.timestamp-o1.timestamp
73 # allows to pick an available IP among a pool
74 # input is expressed as a list of tuples (hostname,ip,user_data)
75 # that can be searched iteratively for a free slot
77 # pool = [ (hostname1,user_data1),
78 # (hostname2,user_data2),
79 # (hostname3,user_data2),
80 # (hostname4,user_data4) ]
81 # assuming that ip1 and ip3 are taken (pingable), then we'd get
83 # pool.next_free() -> entry2
84 # pool.next_free() -> entry4
85 # pool.next_free() -> None
86 # that is, even if ip2 is not busy/pingable when the second next_free() is issued
89 def __init__ (self,hostname,userdata):
90 self.hostname=hostname
91 self.userdata=userdata
92 # slot holds 'busy' or 'free' or 'mine' or 'starting' or None
93 # 'mine' is for our own stuff, 'starting' from the concurrent tests
98 return "Pooled %s (%s) -> %s"%(self.hostname,self.userdata, self.status)
101 if self.status==None: return '?'
102 elif self.status=='busy': return '*'
103 elif self.status=='free': return '.'
104 elif self.status=='mine': return 'M'
105 elif self.status=='starting': return 'S'
108 if self.ip: return self.ip
109 ip=socket.gethostbyname(self.hostname)
115 def __init__ (self, tuples,message):
116 self.pool= [ PoolItem (h,u) for (h,u) in tuples ]
120 for i in self.pool: print i.line()
124 for i in self.pool: line += ' ' + i.char()
127 def _item (self, hostname):
129 if i.hostname==hostname: return i
130 raise Exception ("Could not locate hostname %s in pool %s"%(hostname,self.message))
132 def retrieve_userdata (self, hostname):
133 return self._item(hostname).userdata
135 def get_ip (self, hostname):
136 try: return self._item(hostname).get_ip()
137 except: return socket.gethostbyname(hostname)
139 def set_mine (self, hostname):
140 self._item(hostname).status='mine'
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 simple_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 return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
253 return subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
255 def backquote_ssh (self, argv, trash_err=False):
256 # first probe the ssh link
257 probe_argv=self.test_ssh().actual_argv(['hostname'])
258 hostname=self.backquote ( probe_argv, trash_err=True )
260 print "root@%s unreachable"%self.hostname
263 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
265 ############################################################
267 def __init__ (self, buildname, pid, buildbox):
268 self.buildname=buildname
269 self.buildbox=buildbox
272 def add_pid(self,pid):
273 self.pids.append(pid)
276 return "== %s == (pids=%r)"%(self.buildname,self.pids)
278 class BuildBox (Box):
279 def __init__ (self,hostname):
280 Box.__init__(self,hostname)
281 self.build_instances=[]
283 def add_build (self,buildname,pid):
284 for build in self.build_instances:
285 if build.buildname==buildname:
288 self.build_instances.append(BuildInstance(buildname, pid, self))
291 if not self.build_instances:
292 header ('No build process on %s (%s)'%(self.hostname,self.uptime()))
294 header ("Builds on %s (%s)"%(self.hostname,self.uptime()))
295 for b in self.build_instances:
296 header (b.line(),banner=False)
299 if hasattr(self,'_uptime') and self._uptime: return self._uptime
300 return '*undef* uptime'
302 # inspect box and find currently running builds
303 matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
304 def sense(self,reboot=False,verbose=True):
310 self._uptime=self.backquote_ssh(command,trash_err=True).strip()
311 if not self._uptime: self._uptime='unreachable'
312 pids=self.backquote_ssh(['pgrep','build'],trash_err=True)
314 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
315 ps_lines=self.backquote_ssh (command).split('\n')
316 for line in ps_lines:
317 if not line.strip() or line.find('PID')>=0: continue
318 m=BuildBox.matcher.match(line)
319 if m: self.add_build (m.group('buildname'),m.group('pid'))
320 else: header('command %r returned line that failed to match'%command)
322 ############################################################
324 def __init__ (self, vservername, ctxid, plcbox):
325 self.vservername=vservername
331 def set_timestamp (self,timestamp): self.timestamp=timestamp
332 def set_now (self): self.timestamp=int(time.time())
333 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
336 msg="== %s == (ctx=%s)"%(self.vservername,self.ctxid)
337 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
338 else: msg += " *unknown timestamp*"
339 if self.ctxid==0: msg+=" not (yet?) running"
343 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
344 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
345 self.plc_box.forget(self)
348 def __init__ (self, hostname, max_plcs):
349 Box.__init__(self,hostname)
350 self.plc_instances=[]
351 self.max_plcs=max_plcs
353 def add_vserver (self,vservername,ctxid):
354 for plc in self.plc_instances:
355 if plc.vservername==vservername:
356 header("WARNING, duplicate myplc %s running on %s"%\
357 (vservername,self.hostname),banner=False)
359 self.plc_instances.append(PlcInstance(vservername,ctxid,self))
361 def forget (self, plc_instance):
362 self.plc_instances.remove(plc_instance)
364 # fill one slot even though this one is not started yet
365 def add_dummy (self, plcname):
366 dummy=PlcInstance('dummy_'+plcname,0,self)
368 self.plc_instances.append(dummy)
371 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_plcs,self.free_spots(),self.uname())
375 if not self.plc_instances:
376 header ('No vserver running on %s'%(self.line()))
378 header ("Active plc VMs on %s"%self.line())
379 for p in self.plc_instances:
380 header (p.line(),banner=False)
382 def free_spots (self):
383 return self.max_plcs - len(self.plc_instances)
386 if hasattr(self,'_uname') and self._uname: return self._uname
387 return '*undef* uname'
389 def plc_instance_by_vservername (self, vservername):
390 for p in self.plc_instances:
391 if p.vservername==vservername: return p
394 def sense (self, reboot=False, soft=False):
396 # remove mark for all running servers to avoid resurrection
397 stop_command=['rm','-rf','/etc/vservers/*/apps/init/mark']
398 self.run_ssh(stop_command,"Removing all vserver marks on %s"%self.hostname)
403 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers")
406 self._uname=self.backquote_ssh(['uname','-r']).strip()
407 # try to find fullname (vserver_stat truncates to a ridiculously short name)
408 # fetch the contexts for all vservers on that box
409 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
410 context_map=self.backquote_ssh (map_command)
411 # at this point we have a set of lines like
412 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
414 for map_line in context_map.split("\n"):
415 if not map_line: continue
416 [path,xid] = map_line.split(':')
417 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
418 # at this point ctx_id maps context id to vservername
420 command=['vserver-stat']
421 vserver_stat = self.backquote_ssh (command)
422 for vserver_line in vserver_stat.split("\n"):
423 if not vserver_line: continue
424 context=vserver_line.split()[0]
425 if context=="CTX": continue
426 longname=ctx_dict[context]
427 self.add_vserver(longname,context)
428 # print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
431 running_ctx_ids = [ i.ctxid for i in self.plc_instances ]
432 command= ['grep','.']
433 command += ['/vservers/%s/timestamp'%b for b in running_ctx_ids]
434 command += ['/dev/null']
435 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
436 for ts_line in ts_lines:
437 if not ts_line.strip(): continue
438 # expect /vservers/<vservername>/timestamp:<timestamp>
440 (_,__,vservername,tail)=ts_line.split('/')
441 (_,timestamp)=tail.split(':')
442 timestamp=int(timestamp)
443 q=self.plc_instance_by_vservername(vservername)
445 print 'WARNING unattached plc instance',ts_line
446 print 'was expeting to find',vservername,'in',[i.vservername for i in self.plc_instances]
448 q.set_timestamp(timestamp)
449 except: print 'WARNING, could not parse ts line',ts_line
454 ############################################################
456 def __init__ (self, nodename, pid, qemubox):
457 self.nodename=nodename
459 self.qemu_box=qemubox
464 def set_buildname (self,buildname): self.buildname=buildname
465 def set_timestamp (self,timestamp): self.timestamp=timestamp
466 def set_now (self): self.timestamp=int(time.time())
467 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
470 msg = "== %s == (pid=%s)"%(self.nodename,self.pid)
471 if self.buildname: msg += " <--> %s"%self.buildname
472 else: msg += " *unknown build*"
473 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
474 else: msg += " *unknown timestamp*"
475 if self.pid: msg += "pid=%s"%self.pid
476 else: msg += " not (yet?) running"
480 if self.pid==0: print "cannot kill qemu %s with pid==0"%self.nodename
481 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
482 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
483 self.qemu_box.forget(self)
487 def __init__ (self, hostname, max_qemus):
488 Box.__init__(self,hostname)
489 self.qemu_instances=[]
490 self.max_qemus=max_qemus
492 def add_node (self,nodename,pid):
493 for qemu in self.qemu_instances:
494 if qemu.nodename==nodename:
495 header("WARNING, duplicate qemu %s running on %s"%\
496 (nodename,self.hostname), banner=False)
498 self.qemu_instances.append(QemuInstance(nodename,pid,self))
500 def forget (self, qemu_instance):
501 self.qemu_instances.remove(qemu_instance)
503 # fill one slot even though this one is not started yet
504 def add_dummy (self, nodename):
505 dummy=QemuInstance('dummy_'+nodename,0,self)
507 self.qemu_instances.append(dummy)
510 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_qemus,self.free_spots(),self.driver())
514 if not self.qemu_instances:
515 header ('No qemu process on %s'%(self.line()))
517 header ("Active qemu processes on %s"%(self.line()))
518 for q in self.qemu_instances:
519 header (q.line(),banner=False)
521 def free_spots (self):
522 return self.max_qemus - len(self.qemu_instances)
525 if hasattr(self,'_driver') and self._driver: return self._driver
526 return '*undef* driver'
528 def qemu_instance_by_pid (self,pid):
529 for q in self.qemu_instances:
530 if q.pid==pid: return q
533 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
534 for q in self.qemu_instances:
535 if q.nodename==nodename and q.buildname==buildname:
539 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
540 def sense(self, reboot=False, soft=False):
545 self.run_ssh(box,['pkill','qemu'],"Killing qemu instances")
548 modules=self.backquote_ssh(['lsmod']).split('\n')
549 self._driver='*NO kqemu/kmv_intel MODULE LOADED*'
550 for module in modules:
551 if module.find('kqemu')==0:
552 self._driver='kqemu module loaded'
553 # kvm might be loaded without vkm_intel (we dont have AMD)
554 elif module.find('kvm_intel')==0:
555 self._driver='kvm_intel module loaded'
556 ########## find out running pids
557 pids=self.backquote_ssh(['pgrep','qemu'])
559 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
560 ps_lines = self.backquote_ssh (command).split("\n")
561 for line in ps_lines:
562 if not line.strip() or line.find('PID') >=0 : continue
563 m=QemuBox.matcher.match(line)
564 if m: self.add_node (m.group('nodename'),m.group('pid'))
565 else: header('command %r returned line that failed to match'%command)
566 ########## retrieve alive instances and map to build
568 command=['grep','.','*/*/qemu.pid','/dev/null']
569 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
570 for pid_line in pid_lines:
571 if not pid_line.strip(): continue
572 # expect <build>/<nodename>/qemu.pid:<pid>pid
574 (buildname,nodename,tail)=pid_line.split('/')
575 (_,pid)=tail.split(':')
576 q=self.qemu_instance_by_pid (pid)
578 q.set_buildname(buildname)
579 live_builds.append(buildname)
580 except: print 'WARNING, could not parse pid line',pid_line
581 # retrieve timestamps
582 command= ['grep','.']
583 command += ['%s/*/timestamp'%b for b in live_builds]
584 command += ['/dev/null']
585 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
586 for ts_line in ts_lines:
587 if not ts_line.strip(): continue
588 # expect <build>/<nodename>/timestamp:<timestamp>
590 (buildname,nodename,tail)=ts_line.split('/')
591 nodename=nodename.replace('qemu-','')
592 (_,timestamp)=tail.split(':')
593 timestamp=int(timestamp)
594 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
596 print 'WARNING unattached qemu instance',ts_line,nodename,buildname
598 q.set_timestamp(timestamp)
599 except: print 'WARNING, could not parse ts line',ts_line
601 ############################################################
610 self.options=Options()
611 self.options.dry_run=False
612 self.options.verbose=False
613 self.options.probe=True
614 self.options.soft=True
615 self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
616 self.plc_boxes = [ PlcBox (h,m) for (h,m) in self.plc_boxes_spec ()]
617 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
618 self.all_boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
621 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs")
622 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes")
624 # def build_box_names (self):
625 # return [ h for h in self.build_boxes_spec() ]
626 # def plc_boxes (self):
627 # return [ h for (h,m) in self.plc_boxes_spec() ]
628 # def qemu_boxes (self):
629 # return [ h for (h,m) in self.qemu_boxes_spec() ]
631 # return True if actual sensing takes place
632 def sense (self,force=False):
633 if self._sensed and not force: return False
634 print 'Sensing local substrate...',
635 for b in self.all_boxes: b.sense()
640 def add_dummy_plc (self, plc_boxname, plcname):
641 for pb in self.plc_boxes:
642 if pb.hostname==plc_boxname:
643 pb.add_dummy(plcname)
644 def add_dummy_qemu (self, qemu_boxname, qemuname):
645 for qb in self.qemu_boxes:
646 if qb.hostname==qemu_boxname:
647 qb.add_dummy(qemuname)
650 def provision (self,plcs,options):
652 # attach each plc to a plc box and an IP address
653 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
654 # attach each node/qemu to a qemu box with an IP address
655 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
656 # update the SFA spec accordingly
657 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
660 print '* Could not provision this test on current substrate','--',e,'--','exiting'
661 traceback.print_exc()
664 # it is expected that a couple of options like ips_bplc and ips_vplc
665 # are set or unset together
667 def check_options (x,y):
668 if not x and not y: return True
669 return len(x)==len(y)
671 # find an available plc box (or make space)
672 # and a free IP address (using options if present)
673 def provision_plc (self, plc, options):
675 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
677 #### let's find an IP address for that plc
681 # we don't check anything here,
682 # it is the caller's responsability to cleanup and make sure this makes sense
683 plc_boxname = options.ips_bplc.pop()
684 vplc_hostname=options.ips_vplc.pop()
686 if self.sense(): self.list_all()
689 # try to find an available IP
690 self.vplc_pool.sense()
691 couple=self.vplc_pool.next_free()
693 (vplc_hostname,unused)=couple
694 #### we need to find one plc box that still has a slot
696 # use the box that has max free spots for load balancing
697 for pb in self.plc_boxes:
700 plc_boxname=pb.hostname
702 # if there's no available slot in the plc_boxes, or we need a free IP address
703 # make space by killing the oldest running instance
704 if not plc_boxname or not vplc_hostname:
705 # find the oldest of all our instances
706 all_plc_instances=reduce(lambda x, y: x+y,
707 [ pb.plc_instances for pb in self.plc_boxes ],
709 all_plc_instances.sort(timestamp_sort)
711 plc_instance_to_kill=all_plc_instances[0]
714 if not plc_boxname: msg += " PLC boxes are full"
715 if not vplc_hostname: msg += " vplc IP pool exhausted"
716 raise Exception,"Could not make space for a PLC instance:"+msg
717 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
718 freed_vplc_hostname=plc_instance_to_kill.vservername
719 plc_instance_to_kill.kill()
720 print 'killed oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
721 plc_instance_to_kill.freed_plc_boxname)
722 # use this new plcbox if that was the problem
724 plc_boxname=freed_plc_boxname
725 # ditto for the IP address
726 if not vplc_hostname:
727 vplc_hostname=freed_vplc_hostname
728 # record in pool as mine
729 self.vplc_pool.set_mine(vplc_hostname)
732 self.add_dummy_plc(plc_boxname,plc['name'])
733 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
734 self.vplc_pool.add_starting(vplc_hostname)
736 #### compute a helpful vserver name
737 # remove domain in hostname
738 vplc_simple = vplc_hostname.split('.')[0]
739 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_simple)
740 plc_name = "%s_%s"%(plc['name'],vplc_simple)
742 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
743 (plc['name'],plc_boxname,vplc_hostname,vservername))
745 #### apply in the plc_spec
747 # label=options.personality.replace("linux","")
748 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
749 # 'name':'%s-'+label,
751 'vservername':vservername,
753 'PLC_DB_HOST':vplc_hostname,
754 'PLC_API_HOST':vplc_hostname,
755 'PLC_BOOT_HOST':vplc_hostname,
756 'PLC_WWW_HOST':vplc_hostname,
757 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
758 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
763 # mappers only work on a list of plcs
764 return TestMapper([plc],options).map(mapper)[0]
767 def provision_qemus (self, plc, options):
769 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
771 test_mapper = TestMapper ([plc], options)
772 nodenames = test_mapper.node_names()
774 for nodename in nodenames:
776 if options.ips_vnode:
777 # as above, it's a rerun, take it for granted
778 qemu_boxname=options.ips_bnode.pop()
779 vnode_hostname=options.ips_vnode.pop()
781 if self.sense(): self.list_all()
784 # try to find an available IP
785 self.vnode_pool.sense()
786 couple=self.vnode_pool.next_free()
788 (vnode_hostname,unused)=couple
789 # find a physical box
791 # use the box that has max free spots for load balancing
792 for qb in self.qemu_boxes:
795 qemu_boxname=qb.hostname
797 # if we miss the box or the IP, kill the oldest instance
798 if not qemu_boxname or not vnode_hostname:
799 # find the oldest of all our instances
800 all_qemu_instances=reduce(lambda x, y: x+y,
801 [ qb.qemu_instances for qb in self.qemu_boxes ],
803 all_qemu_instances.sort(timestamp_sort)
805 qemu_instance_to_kill=all_qemu_instances[0]
808 if not qemu_boxname: msg += " QEMU boxes are full"
809 if not vnode_hostname: msg += " vnode IP pool exhausted"
810 raise Exception,"Could not make space for a QEMU instance:"+msg
811 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
812 freed_vnode_hostname=qemu_instance_to_kill.nodename
814 qemu_instance_to_kill.kill()
815 print 'killed oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
816 qemu_instance_to_kill.qemu_boxname.hostname)
817 # use these freed resources where needed
819 qemu_boxname=freed_qemu_boxname
820 if not vnode_hostname:
821 vnode_hostname=freed_vnode_hostname
822 self.vnode_pool.set_mine(vnode_hostname)
824 self.add_dummy_qemu (qemu_boxname,nodename)
825 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
826 ip=self.vnode_pool.get_ip (vnode_hostname)
827 self.vnode_pool.add_starting(vnode_hostname)
829 vnode_fqdn = vnode_hostname
830 if vnode_fqdn.find('.')<0:
831 vnode_fqdn += "."+self.domain()
832 nodemap={'host_box':qemu_boxname,
833 'node_fields:hostname':vnode_fqdn,
834 'interface_fields:ip':ip,
835 'interface_fields:mac':mac,
837 nodemap.update(self.network_settings())
838 maps.append ( (nodename, nodemap) )
840 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
841 (nodename,qemu_boxname,vnode_hostname,mac))
843 return test_mapper.map({'node':maps})[0]
845 def localize_sfa_rspec (self,plc,options):
847 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
848 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
849 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
850 plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
851 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
852 for site in plc['sites']:
853 for node in site['nodes']:
854 plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
857 #################### release:
858 def release (self,options):
859 self.vplc_pool.release_my_starting()
860 self.vnode_pool.release_my_starting()
863 #################### show results for interactive mode
866 for b in self.all_boxes: b.list()
868 def get_box (self,box):
869 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes:
870 if b.simple_hostname()==box:
872 print "Could not find box %s"%box
875 def list_box(self,box):
881 # can be run as a utility to manage the local infrastructure
883 parser=OptionParser()
884 (options,args)=parser.parse_args()