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 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=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
305 def sense(self,reboot=False,verbose=True):
311 self._uptime=self.backquote_ssh(command,trash_err=True).strip()
312 if not self._uptime: self._uptime='unreachable'
313 pids=self.backquote_ssh(['pgrep','build'],trash_err=True)
315 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
316 ps_lines=self.backquote_ssh (command).split('\n')
317 for line in ps_lines:
318 if not line.strip() or line.find('PID')>=0: continue
319 m=BuildBox.matcher.match(line)
320 if m: self.add_build (m.group('buildname'),m.group('pid'))
321 else: header('command %r returned line that failed to match'%command)
323 ############################################################
325 def __init__ (self, vservername, ctxid, plcbox):
326 self.vservername=vservername
332 def set_timestamp (self,timestamp): self.timestamp=timestamp
333 def set_now (self): self.timestamp=int(time.time())
334 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
337 msg="== %s == (ctx=%s)"%(self.vservername,self.ctxid)
338 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
339 else: msg += " *unknown timestamp*"
340 if self.ctxid==0: msg+=" not (yet?) running"
344 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
345 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
346 self.plc_box.forget(self)
349 def __init__ (self, hostname, max_plcs):
350 Box.__init__(self,hostname)
351 self.plc_instances=[]
352 self.max_plcs=max_plcs
354 def add_vserver (self,vservername,ctxid):
355 for plc in self.plc_instances:
356 if plc.vservername==vservername:
357 header("WARNING, duplicate myplc %s running on %s"%\
358 (vservername,self.hostname),banner=False)
360 self.plc_instances.append(PlcInstance(vservername,ctxid,self))
362 def forget (self, plc_instance):
363 self.plc_instances.remove(plc_instance)
365 # fill one slot even though this one is not started yet
366 def add_dummy (self, plcname):
367 dummy=PlcInstance('dummy_'+plcname,0,self)
369 self.plc_instances.append(dummy)
372 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_plcs,self.free_spots(),self.uname())
376 if not self.plc_instances:
377 header ('No vserver running on %s'%(self.line()))
379 header ("Active plc VMs on %s"%self.line())
380 for p in self.plc_instances:
381 header (p.line(),banner=False)
383 def free_spots (self):
384 return self.max_plcs - len(self.plc_instances)
387 if hasattr(self,'_uname') and self._uname: return self._uname
388 return '*undef* uname'
390 def plc_instance_by_vservername (self, vservername):
391 for p in self.plc_instances:
392 if p.vservername==vservername: return p
395 def sense (self, reboot=False, soft=False):
397 # remove mark for all running servers to avoid resurrection
398 stop_command=['rm','-rf','/etc/vservers/*/apps/init/mark']
399 self.run_ssh(stop_command,"Removing all vserver marks on %s"%self.hostname)
404 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers")
407 self._uname=self.backquote_ssh(['uname','-r']).strip()
408 # try to find fullname (vserver_stat truncates to a ridiculously short name)
409 # fetch the contexts for all vservers on that box
410 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
411 context_map=self.backquote_ssh (map_command)
412 # at this point we have a set of lines like
413 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
415 for map_line in context_map.split("\n"):
416 if not map_line: continue
417 [path,xid] = map_line.split(':')
418 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
419 # at this point ctx_id maps context id to vservername
421 command=['vserver-stat']
422 vserver_stat = self.backquote_ssh (command)
423 for vserver_line in vserver_stat.split("\n"):
424 if not vserver_line: continue
425 context=vserver_line.split()[0]
426 if context=="CTX": continue
427 longname=ctx_dict[context]
428 self.add_vserver(longname,context)
429 # print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
432 running_vsnames = [ i.vservername for i in self.plc_instances ]
433 command= ['grep','.']
434 command += ['/vservers/%s/timestamp'%vs for vs in running_vsnames]
435 command += ['/dev/null']
436 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
437 for ts_line in ts_lines:
438 if not ts_line.strip(): continue
439 # expect /vservers/<vservername>/timestamp:<timestamp>
441 (_,__,vservername,tail)=ts_line.split('/')
442 (_,timestamp)=tail.split(':')
443 timestamp=int(timestamp)
444 p=self.plc_instance_by_vservername(vservername)
446 print 'WARNING unattached plc instance',ts_line
447 print 'was expecting to find',vservername,'in',[i.vservername for i in self.plc_instances]
449 p.set_timestamp(timestamp)
450 except: print 'WARNING, could not parse ts line',ts_line
455 ############################################################
457 def __init__ (self, nodename, pid, qemubox):
458 self.nodename=nodename
460 self.qemu_box=qemubox
465 def set_buildname (self,buildname): self.buildname=buildname
466 def set_timestamp (self,timestamp): self.timestamp=timestamp
467 def set_now (self): self.timestamp=int(time.time())
468 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
471 msg = "== %s == (pid=%s)"%(self.nodename,self.pid)
472 if self.buildname: msg += " <--> %s"%self.buildname
473 else: msg += " *unknown build*"
474 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
475 else: msg += " *unknown timestamp*"
476 if self.pid: msg += " pid=%s"%self.pid
477 else: msg += " not (yet?) running"
481 if self.pid==0: print "cannot kill qemu %s with pid==0"%self.nodename
482 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
483 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
484 self.qemu_box.forget(self)
488 def __init__ (self, hostname, max_qemus):
489 Box.__init__(self,hostname)
490 self.qemu_instances=[]
491 self.max_qemus=max_qemus
493 def add_node (self,nodename,pid):
494 for qemu in self.qemu_instances:
495 if qemu.nodename==nodename:
496 header("WARNING, duplicate qemu %s running on %s"%\
497 (nodename,self.hostname), banner=False)
499 self.qemu_instances.append(QemuInstance(nodename,pid,self))
501 def forget (self, qemu_instance):
502 self.qemu_instances.remove(qemu_instance)
504 # fill one slot even though this one is not started yet
505 def add_dummy (self, nodename):
506 dummy=QemuInstance('dummy_'+nodename,0,self)
508 self.qemu_instances.append(dummy)
511 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_qemus,self.free_spots(),self.driver())
515 if not self.qemu_instances:
516 header ('No qemu process on %s'%(self.line()))
518 header ("Active qemu processes on %s"%(self.line()))
519 for q in self.qemu_instances:
520 header (q.line(),banner=False)
522 def free_spots (self):
523 return self.max_qemus - len(self.qemu_instances)
526 if hasattr(self,'_driver') and self._driver: return self._driver
527 return '*undef* driver'
529 def qemu_instance_by_pid (self,pid):
530 for q in self.qemu_instances:
531 if q.pid==pid: return q
534 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
535 for q in self.qemu_instances:
536 if q.nodename==nodename and q.buildname==buildname:
540 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
541 def sense(self, reboot=False, soft=False):
546 self.run_ssh(box,['pkill','qemu'],"Killing qemu instances")
549 modules=self.backquote_ssh(['lsmod']).split('\n')
550 self._driver='*NO kqemu/kmv_intel MODULE LOADED*'
551 for module in modules:
552 if module.find('kqemu')==0:
553 self._driver='kqemu module loaded'
554 # kvm might be loaded without vkm_intel (we dont have AMD)
555 elif module.find('kvm_intel')==0:
556 self._driver='kvm_intel module loaded'
557 ########## find out running pids
558 pids=self.backquote_ssh(['pgrep','qemu'])
560 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
561 ps_lines = self.backquote_ssh (command).split("\n")
562 for line in ps_lines:
563 if not line.strip() or line.find('PID') >=0 : continue
564 m=QemuBox.matcher.match(line)
565 if m: self.add_node (m.group('nodename'),m.group('pid'))
566 else: header('command %r returned line that failed to match'%command)
567 ########## retrieve alive instances and map to build
569 command=['grep','.','*/*/qemu.pid','/dev/null']
570 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
571 for pid_line in pid_lines:
572 if not pid_line.strip(): continue
573 # expect <build>/<nodename>/qemu.pid:<pid>pid
575 (buildname,nodename,tail)=pid_line.split('/')
576 (_,pid)=tail.split(':')
577 q=self.qemu_instance_by_pid (pid)
579 q.set_buildname(buildname)
580 live_builds.append(buildname)
581 except: print 'WARNING, could not parse pid line',pid_line
582 # retrieve timestamps
583 command= ['grep','.']
584 command += ['%s/*/timestamp'%b for b in live_builds]
585 command += ['/dev/null']
586 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
587 for ts_line in ts_lines:
588 if not ts_line.strip(): continue
589 # expect <build>/<nodename>/timestamp:<timestamp>
591 (buildname,nodename,tail)=ts_line.split('/')
592 nodename=nodename.replace('qemu-','')
593 (_,timestamp)=tail.split(':')
594 timestamp=int(timestamp)
595 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
597 print 'WARNING unattached qemu instance',ts_line,nodename,buildname
599 q.set_timestamp(timestamp)
600 except: print 'WARNING, could not parse ts line',ts_line
602 ############################################################
611 self.options=Options()
612 self.options.dry_run=False
613 self.options.verbose=False
614 self.options.probe=True
615 self.options.soft=True
616 self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
617 self.plc_boxes = [ PlcBox (h,m) for (h,m) in self.plc_boxes_spec ()]
618 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
619 self.all_boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
622 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs")
623 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes")
625 # def build_box_names (self):
626 # return [ h for h in self.build_boxes_spec() ]
627 # def plc_boxes (self):
628 # return [ h for (h,m) in self.plc_boxes_spec() ]
629 # def qemu_boxes (self):
630 # return [ h for (h,m) in self.qemu_boxes_spec() ]
632 # return True if actual sensing takes place
633 def sense (self,force=False):
634 if self._sensed and not force: return False
635 print 'Sensing local substrate...',
636 for b in self.all_boxes: b.sense()
641 def add_dummy_plc (self, plc_boxname, plcname):
642 for pb in self.plc_boxes:
643 if pb.hostname==plc_boxname:
644 pb.add_dummy(plcname)
645 def add_dummy_qemu (self, qemu_boxname, qemuname):
646 for qb in self.qemu_boxes:
647 if qb.hostname==qemu_boxname:
648 qb.add_dummy(qemuname)
651 def provision (self,plcs,options):
653 # attach each plc to a plc box and an IP address
654 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
655 # attach each node/qemu to a qemu box with an IP address
656 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
657 # update the SFA spec accordingly
658 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
661 print '* Could not provision this test on current substrate','--',e,'--','exiting'
662 traceback.print_exc()
665 # it is expected that a couple of options like ips_bplc and ips_vplc
666 # are set or unset together
668 def check_options (x,y):
669 if not x and not y: return True
670 return len(x)==len(y)
672 # find an available plc box (or make space)
673 # and a free IP address (using options if present)
674 def provision_plc (self, plc, options):
676 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
678 #### let's find an IP address for that plc
682 # we don't check anything here,
683 # it is the caller's responsability to cleanup and make sure this makes sense
684 plc_boxname = options.ips_bplc.pop()
685 vplc_hostname=options.ips_vplc.pop()
687 if self.sense(): self.list_all()
690 # try to find an available IP
691 self.vplc_pool.sense()
692 couple=self.vplc_pool.next_free()
694 (vplc_hostname,unused)=couple
695 #### we need to find one plc box that still has a slot
697 # use the box that has max free spots for load balancing
698 for pb in self.plc_boxes:
701 plc_boxname=pb.hostname
703 # if there's no available slot in the plc_boxes, or we need a free IP address
704 # make space by killing the oldest running instance
705 if not plc_boxname or not vplc_hostname:
706 # find the oldest of all our instances
707 all_plc_instances=reduce(lambda x, y: x+y,
708 [ pb.plc_instances for pb in self.plc_boxes ],
710 all_plc_instances.sort(timestamp_sort)
712 plc_instance_to_kill=all_plc_instances[0]
715 if not plc_boxname: msg += " PLC boxes are full"
716 if not vplc_hostname: msg += " vplc IP pool exhausted"
717 raise Exception,"Could not make space for a PLC instance:"+msg
718 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
719 freed_vplc_hostname=plc_instance_to_kill.vservername
720 plc_instance_to_kill.kill()
721 print 'killed oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
722 plc_instance_to_kill.freed_plc_boxname)
723 # use this new plcbox if that was the problem
725 plc_boxname=freed_plc_boxname
726 # ditto for the IP address
727 if not vplc_hostname:
728 vplc_hostname=freed_vplc_hostname
729 # record in pool as mine
730 self.vplc_pool.set_mine(vplc_hostname)
733 self.add_dummy_plc(plc_boxname,plc['name'])
734 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
735 self.vplc_pool.add_starting(vplc_hostname)
737 #### compute a helpful vserver name
738 # remove domain in hostname
739 vplc_simple = vplc_hostname.split('.')[0]
740 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_simple)
741 plc_name = "%s_%s"%(plc['name'],vplc_simple)
743 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
744 (plc['name'],plc_boxname,vplc_hostname,vservername))
746 #### apply in the plc_spec
748 # label=options.personality.replace("linux","")
749 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
750 # 'name':'%s-'+label,
752 'vservername':vservername,
754 'PLC_DB_HOST':vplc_hostname,
755 'PLC_API_HOST':vplc_hostname,
756 'PLC_BOOT_HOST':vplc_hostname,
757 'PLC_WWW_HOST':vplc_hostname,
758 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
759 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
764 # mappers only work on a list of plcs
765 return TestMapper([plc],options).map(mapper)[0]
768 def provision_qemus (self, plc, options):
770 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
772 test_mapper = TestMapper ([plc], options)
773 nodenames = test_mapper.node_names()
775 for nodename in nodenames:
777 if options.ips_vnode:
778 # as above, it's a rerun, take it for granted
779 qemu_boxname=options.ips_bnode.pop()
780 vnode_hostname=options.ips_vnode.pop()
782 if self.sense(): self.list_all()
785 # try to find an available IP
786 self.vnode_pool.sense()
787 couple=self.vnode_pool.next_free()
789 (vnode_hostname,unused)=couple
790 # find a physical box
792 # use the box that has max free spots for load balancing
793 for qb in self.qemu_boxes:
796 qemu_boxname=qb.hostname
798 # if we miss the box or the IP, kill the oldest instance
799 if not qemu_boxname or not vnode_hostname:
800 # find the oldest of all our instances
801 all_qemu_instances=reduce(lambda x, y: x+y,
802 [ qb.qemu_instances for qb in self.qemu_boxes ],
804 all_qemu_instances.sort(timestamp_sort)
806 qemu_instance_to_kill=all_qemu_instances[0]
809 if not qemu_boxname: msg += " QEMU boxes are full"
810 if not vnode_hostname: msg += " vnode IP pool exhausted"
811 raise Exception,"Could not make space for a QEMU instance:"+msg
812 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
813 freed_vnode_hostname=qemu_instance_to_kill.nodename
815 qemu_instance_to_kill.kill()
816 print 'killed oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
817 qemu_instance_to_kill.qemu_boxname.hostname)
818 # use these freed resources where needed
820 qemu_boxname=freed_qemu_boxname
821 if not vnode_hostname:
822 vnode_hostname=freed_vnode_hostname
823 self.vnode_pool.set_mine(vnode_hostname)
825 self.add_dummy_qemu (qemu_boxname,nodename)
826 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
827 ip=self.vnode_pool.get_ip (vnode_hostname)
828 self.vnode_pool.add_starting(vnode_hostname)
830 vnode_fqdn = vnode_hostname
831 if vnode_fqdn.find('.')<0:
832 vnode_fqdn += "."+self.domain()
833 nodemap={'host_box':qemu_boxname,
834 'node_fields:hostname':vnode_fqdn,
835 'interface_fields:ip':ip,
836 'interface_fields:mac':mac,
838 nodemap.update(self.network_settings())
839 maps.append ( (nodename, nodemap) )
841 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
842 (nodename,qemu_boxname,vnode_hostname,mac))
844 return test_mapper.map({'node':maps})[0]
846 def localize_sfa_rspec (self,plc,options):
848 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
849 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
850 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
851 plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
852 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
853 for site in plc['sites']:
854 for node in site['nodes']:
855 plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
858 #################### release:
859 def release (self,options):
860 self.vplc_pool.release_my_starting()
861 self.vnode_pool.release_my_starting()
864 #################### show results for interactive mode
867 for b in self.all_boxes: b.list()
869 def get_box (self,box):
870 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes:
871 if b.simple_hostname()==box:
873 print "Could not find box %s"%box
876 def list_box(self,box):
882 # can be run as a utility to manage the local infrastructure
884 parser=OptionParser()
885 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
887 (options,args)=parser.parse_args()
889 self.options.verbose=True