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:
191 if self.check_ping (item.hostname):
199 print 'Sensing IP pool',self.message,
203 print 'After starting: IP pool'
206 # OS-dependent ping option (support for macos, for convenience)
207 ping_timeout_option = None
208 # returns True when a given hostname/ip responds to ping
209 def check_ping (self,hostname):
210 if not Pool.ping_timeout_option:
211 (status,osname) = commands.getstatusoutput("uname -s")
213 raise Exception, "TestPool: Cannot figure your OS name"
214 if osname == "Linux":
215 Pool.ping_timeout_option="-w"
216 elif osname == "Darwin":
217 Pool.ping_timeout_option="-t"
219 command="ping -c 1 %s 1 %s"%(Pool.ping_timeout_option,hostname)
220 (status,output) = commands.getstatusoutput(command)
225 def __init__ (self,hostname):
226 self.hostname=hostname
227 def short_hostname (self):
228 return self.hostname.split('.')[0]
229 def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
230 def reboot (self, options):
231 self.test_ssh().run("shutdown -r now",message="Rebooting %s"%self.hostname,
232 dry_run=options.dry_run)
234 def run(self,argv,message=None,trash_err=False,dry_run=False):
242 return subprocess.call(argv)
244 return subprocess.call(argv,stderr=file('/dev/null','w'))
246 def run_ssh (self, argv, message, trash_err=False, dry_run=False):
247 ssh_argv = self.test_ssh().actual_argv(argv)
248 result=self.run (ssh_argv, message, trash_err, dry_run=dry_run)
250 print "WARNING: failed to run %s on %s"%(" ".join(argv),self.hostname)
253 def backquote (self, argv, trash_err=False):
255 result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
257 result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
260 def backquote_ssh (self, argv, trash_err=False):
261 # first probe the ssh link
262 probe_argv=self.test_ssh().actual_argv(['hostname'])
263 hostname=self.backquote ( probe_argv, trash_err=True )
265 print "root@%s unreachable"%self.hostname
268 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
270 ############################################################
272 def __init__ (self, buildname, pid, buildbox):
273 self.buildname=buildname
274 self.buildbox=buildbox
277 def add_pid(self,pid):
278 self.pids.append(pid)
281 return "== %s == (pids=%r)"%(self.buildname,self.pids)
283 class BuildBox (Box):
284 def __init__ (self,hostname):
285 Box.__init__(self,hostname)
286 self.build_instances=[]
288 def add_build (self,buildname,pid):
289 for build in self.build_instances:
290 if build.buildname==buildname:
293 self.build_instances.append(BuildInstance(buildname, pid, self))
296 if not self.build_instances:
297 header ('No build process on %s (%s)'%(self.hostname,self.uptime()))
299 header ("Builds on %s (%s)"%(self.hostname,self.uptime()))
300 for b in self.build_instances:
301 header (b.line(),banner=False)
304 if hasattr(self,'_uptime') and self._uptime: return self._uptime
305 return '*undef* uptime'
307 # inspect box and find currently running builds
308 matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
309 def sense(self,options):
314 command=['pkill','vbuild']
315 self.run_ssh(command,"Terminating vbuild processes",dry_run=options.dry_run)
319 self._uptime=self.backquote_ssh(command,trash_err=True).strip()
320 if not self._uptime: self._uptime='unreachable'
321 pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
323 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
324 ps_lines=self.backquote_ssh (command).split('\n')
325 for line in ps_lines:
326 if not line.strip() or line.find('PID')>=0: continue
327 m=BuildBox.matcher.match(line)
329 date=time.strftime('%Y-%m-%d',time.localtime(time.time()))
330 buildname=m.group('buildname').replace('@DATE@',date)
331 self.add_build (buildname,m.group('pid'))
332 else: header('command %r returned line that failed to match'%command)
334 ############################################################
336 def __init__ (self, vservername, ctxid, plcbox):
337 self.vservername=vservername
343 def set_timestamp (self,timestamp): self.timestamp=timestamp
344 def set_now (self): self.timestamp=int(time.time())
345 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
348 return self.vservername.split('-')[-1]
349 def buildname (self):
350 return self.vservername.rsplit('-',2)[0]
353 msg="== %s =="%(self.vplcname())
354 msg += " [=%s]"%self.vservername
355 if self.ctxid==0: msg+=" not (yet?) running"
356 else: msg+=" (ctx=%s)"%self.ctxid
357 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
358 else: msg += " *unknown timestamp*"
362 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
363 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
364 self.plc_box.forget(self)
367 def __init__ (self, hostname, max_plcs):
368 Box.__init__(self,hostname)
369 self.plc_instances=[]
370 self.max_plcs=max_plcs
372 def add_vserver (self,vservername,ctxid):
373 for plc in self.plc_instances:
374 if plc.vservername==vservername:
375 header("WARNING, duplicate myplc %s running on %s"%\
376 (vservername,self.hostname),banner=False)
378 self.plc_instances.append(PlcInstance(vservername,ctxid,self))
380 def forget (self, plc_instance):
381 self.plc_instances.remove(plc_instance)
383 # fill one slot even though this one is not started yet
384 def add_dummy (self, plcname):
385 dummy=PlcInstance('dummy_'+plcname,0,self)
387 self.plc_instances.append(dummy)
390 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_plcs,self.free_spots(),self.uname())
394 if not self.plc_instances:
395 header ('No vserver running on %s'%(self.line()))
397 header ("Active plc VMs on %s"%self.line())
398 self.plc_instances.sort(timestamp_sort)
399 for p in self.plc_instances:
400 header (p.line(),banner=False)
402 def free_spots (self):
403 return self.max_plcs - len(self.plc_instances)
406 if hasattr(self,'_uname') and self._uname: return self._uname
407 return '*undef* uname'
409 def plc_instance_by_vservername (self, vservername):
410 for p in self.plc_instances:
411 if p.vservername==vservername: return p
414 def sense (self, options):
420 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers",
421 dry_run=options.dry_run)
424 self._uname=self.backquote_ssh(['uname','-r']).strip()
425 # try to find fullname (vserver_stat truncates to a ridiculously short name)
426 # fetch the contexts for all vservers on that box
427 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
428 context_map=self.backquote_ssh (map_command)
429 # at this point we have a set of lines like
430 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
432 for map_line in context_map.split("\n"):
433 if not map_line: continue
434 [path,xid] = map_line.split(':')
435 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
436 # at this point ctx_id maps context id to vservername
438 command=['vserver-stat']
439 vserver_stat = self.backquote_ssh (command)
440 for vserver_line in vserver_stat.split("\n"):
441 if not vserver_line: continue
442 context=vserver_line.split()[0]
443 if context=="CTX": continue
444 longname=ctx_dict[context]
445 self.add_vserver(longname,context)
446 # print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
449 running_vsnames = [ i.vservername for i in self.plc_instances ]
450 command= ['grep','.']
451 command += ['/vservers/%s/timestamp'%vs for vs in running_vsnames]
452 command += ['/dev/null']
453 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
454 for ts_line in ts_lines:
455 if not ts_line.strip(): continue
456 # expect /vservers/<vservername>/timestamp:<timestamp>
458 (_,__,vservername,tail)=ts_line.split('/')
459 (_,timestamp)=tail.split(':')
460 timestamp=int(timestamp)
461 p=self.plc_instance_by_vservername(vservername)
463 print 'WARNING unattached plc instance',ts_line
464 print 'was expecting to find',vservername,'in',[i.vservername for i in self.plc_instances]
466 p.set_timestamp(timestamp)
467 except: print 'WARNING, could not parse ts line',ts_line
472 ############################################################
474 def __init__ (self, nodename, pid, qemubox):
475 self.nodename=nodename
477 self.qemu_box=qemubox
482 def set_buildname (self,buildname): self.buildname=buildname
483 def set_timestamp (self,timestamp): self.timestamp=timestamp
484 def set_now (self): self.timestamp=int(time.time())
485 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
488 msg = "== %s =="%(self.nodename)
489 msg += " [=%s]"%self.buildname
490 if self.pid: msg += " (pid=%s)"%self.pid
491 else: msg += " not (yet?) running"
492 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
493 else: msg += " *unknown timestamp*"
498 print "cannot kill qemu %s with pid==0"%self.nodename
500 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
501 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
502 self.qemu_box.forget(self)
506 def __init__ (self, hostname, max_qemus):
507 Box.__init__(self,hostname)
508 self.qemu_instances=[]
509 self.max_qemus=max_qemus
511 def add_node (self,nodename,pid):
512 for qemu in self.qemu_instances:
513 if qemu.nodename==nodename:
514 header("WARNING, duplicate qemu %s running on %s"%\
515 (nodename,self.hostname), banner=False)
517 self.qemu_instances.append(QemuInstance(nodename,pid,self))
519 def forget (self, qemu_instance):
520 self.qemu_instances.remove(qemu_instance)
522 # fill one slot even though this one is not started yet
523 def add_dummy (self, nodename):
524 dummy=QemuInstance('dummy_'+nodename,0,self)
526 self.qemu_instances.append(dummy)
529 msg="%s [max=%d,%d free] (%s)"%(self.hostname, self.max_qemus,self.free_spots(),self.driver())
533 if not self.qemu_instances:
534 header ('No qemu process on %s'%(self.line()))
536 header ("Active qemu processes on %s"%(self.line()))
537 self.qemu_instances.sort(timestamp_sort)
538 for q in self.qemu_instances:
539 header (q.line(),banner=False)
541 def free_spots (self):
542 return self.max_qemus - len(self.qemu_instances)
545 if hasattr(self,'_driver') and self._driver: return self._driver
546 return '*undef* driver'
548 def qemu_instance_by_pid (self,pid):
549 for q in self.qemu_instances:
550 if q.pid==pid: return q
553 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
554 for q in self.qemu_instances:
555 if q.nodename==nodename and q.buildname==buildname:
559 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
560 def sense(self, options):
565 self.run_ssh(['pkill','qemu'],"Killing qemu instances",
566 dry_run=options.dry_run)
569 modules=self.backquote_ssh(['lsmod']).split('\n')
570 self._driver='*NO kqemu/kmv_intel MODULE LOADED*'
571 for module in modules:
572 if module.find('kqemu')==0:
573 self._driver='kqemu module loaded'
574 # kvm might be loaded without vkm_intel (we dont have AMD)
575 elif module.find('kvm_intel')==0:
576 self._driver='kvm_intel module loaded'
577 ########## find out running pids
578 pids=self.backquote_ssh(['pgrep','qemu'])
580 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
581 ps_lines = self.backquote_ssh (command).split("\n")
582 for line in ps_lines:
583 if not line.strip() or line.find('PID') >=0 : continue
584 m=QemuBox.matcher.match(line)
585 if m: self.add_node (m.group('nodename'),m.group('pid'))
586 else: header('command %r returned line that failed to match'%command)
587 ########## retrieve alive instances and map to build
589 command=['grep','.','*/*/qemu.pid','/dev/null']
590 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
591 for pid_line in pid_lines:
592 if not pid_line.strip(): continue
593 # expect <build>/<nodename>/qemu.pid:<pid>pid
595 (buildname,nodename,tail)=pid_line.split('/')
596 (_,pid)=tail.split(':')
597 q=self.qemu_instance_by_pid (pid)
599 q.set_buildname(buildname)
600 live_builds.append(buildname)
601 except: print 'WARNING, could not parse pid line',pid_line
602 # retrieve timestamps
603 command= ['grep','.']
604 command += ['%s/*/timestamp'%b for b in live_builds]
605 command += ['/dev/null']
606 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
607 for ts_line in ts_lines:
608 if not ts_line.strip(): continue
609 # expect <build>/<nodename>/timestamp:<timestamp>
611 (buildname,nodename,tail)=ts_line.split('/')
612 nodename=nodename.replace('qemu-','')
613 (_,timestamp)=tail.split(':')
614 timestamp=int(timestamp)
615 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
617 print 'WARNING unattached qemu instance',ts_line,nodename,buildname
619 q.set_timestamp(timestamp)
620 except: print 'WARNING, could not parse ts line',ts_line
622 ############################################################
628 self.options=Options()
629 self.options.dry_run=False
630 self.options.verbose=False
631 self.options.reboot=False
632 self.options.soft=False
633 self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
634 self.plc_boxes = [ PlcBox (h,m) for (h,m) in self.plc_boxes_spec ()]
635 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
636 self.all_boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes
639 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs")
640 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes")
642 def fqdn (self, hostname):
643 if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
645 def short_hostname (self, hostname):
646 if hostname.find('.')>=0: return hostname.split('.')[0]
649 # return True if actual sensing takes place
650 def sense (self,force=False):
651 if self._sensed and not force: return False
652 print 'Sensing local substrate...',
653 for b in self.all_boxes: b.sense(self.options)
658 def add_dummy_plc (self, plc_boxname, plcname):
659 for pb in self.plc_boxes:
660 if pb.hostname==plc_boxname:
661 pb.add_dummy(plcname)
662 def add_dummy_qemu (self, qemu_boxname, qemuname):
663 for qb in self.qemu_boxes:
664 if qb.hostname==qemu_boxname:
665 qb.add_dummy(qemuname)
668 def provision (self,plcs,options):
670 # attach each plc to a plc box and an IP address
671 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
672 # attach each node/qemu to a qemu box with an IP address
673 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
674 # update the SFA spec accordingly
675 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
678 print '* Could not provision this test on current substrate','--',e,'--','exiting'
679 traceback.print_exc()
682 # it is expected that a couple of options like ips_bplc and ips_vplc
683 # are set or unset together
685 def check_options (x,y):
686 if not x and not y: return True
687 return len(x)==len(y)
689 # find an available plc box (or make space)
690 # and a free IP address (using options if present)
691 def provision_plc (self, plc, options):
693 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
695 #### let's find an IP address for that plc
699 # we don't check anything here,
700 # it is the caller's responsability to cleanup and make sure this makes sense
701 plc_boxname = options.ips_bplc.pop()
702 vplc_hostname=options.ips_vplc.pop()
704 if self.sense(): self.list_all()
707 # try to find an available IP
708 self.vplc_pool.sense()
709 couple=self.vplc_pool.next_free()
711 (vplc_hostname,unused)=couple
712 #### we need to find one plc box that still has a slot
714 # use the box that has max free spots for load balancing
715 for pb in self.plc_boxes:
718 plc_boxname=pb.hostname
720 # if there's no available slot in the plc_boxes, or we need a free IP address
721 # make space by killing the oldest running instance
722 if not plc_boxname or not vplc_hostname:
723 # find the oldest of all our instances
724 all_plc_instances=reduce(lambda x, y: x+y,
725 [ pb.plc_instances for pb in self.plc_boxes ],
727 all_plc_instances.sort(timestamp_sort)
729 plc_instance_to_kill=all_plc_instances[0]
732 if not plc_boxname: msg += " PLC boxes are full"
733 if not vplc_hostname: msg += " vplc IP pool exhausted"
734 raise Exception,"Could not make space for a PLC instance:"+msg
735 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
736 freed_vplc_hostname=plc_instance_to_kill.vplcname()
737 message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
739 plc_instance_to_kill.kill()
740 # use this new plcbox if that was the problem
742 plc_boxname=freed_plc_boxname
743 # ditto for the IP address
744 if not vplc_hostname:
745 vplc_hostname=freed_vplc_hostname
746 # record in pool as mine
747 self.vplc_pool.set_mine(vplc_hostname)
750 self.add_dummy_plc(plc_boxname,plc['name'])
751 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
752 self.vplc_pool.add_starting(vplc_hostname)
754 #### compute a helpful vserver name
755 # remove domain in hostname
756 vplc_short = self.short_hostname(vplc_hostname)
757 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_short)
758 plc_name = "%s_%s"%(plc['name'],vplc_short)
760 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
761 (plc['name'],plc_boxname,vplc_hostname,vservername))
763 #### apply in the plc_spec
765 # label=options.personality.replace("linux","")
766 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
767 # 'name':'%s-'+label,
769 'vservername':vservername,
771 'PLC_DB_HOST':vplc_hostname,
772 'PLC_API_HOST':vplc_hostname,
773 'PLC_BOOT_HOST':vplc_hostname,
774 'PLC_WWW_HOST':vplc_hostname,
775 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
776 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
781 # mappers only work on a list of plcs
782 return TestMapper([plc],options).map(mapper)[0]
785 def provision_qemus (self, plc, options):
787 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
789 test_mapper = TestMapper ([plc], options)
790 nodenames = test_mapper.node_names()
792 for nodename in nodenames:
794 if options.ips_vnode:
795 # as above, it's a rerun, take it for granted
796 qemu_boxname=options.ips_bnode.pop()
797 vnode_hostname=options.ips_vnode.pop()
799 if self.sense(): self.list_all()
802 # try to find an available IP
803 self.vnode_pool.sense()
804 couple=self.vnode_pool.next_free()
806 (vnode_hostname,unused)=couple
807 # find a physical box
809 # use the box that has max free spots for load balancing
810 for qb in self.qemu_boxes:
813 qemu_boxname=qb.hostname
815 # if we miss the box or the IP, kill the oldest instance
816 if not qemu_boxname or not vnode_hostname:
817 # find the oldest of all our instances
818 all_qemu_instances=reduce(lambda x, y: x+y,
819 [ qb.qemu_instances for qb in self.qemu_boxes ],
821 all_qemu_instances.sort(timestamp_sort)
823 qemu_instance_to_kill=all_qemu_instances[0]
826 if not qemu_boxname: msg += " QEMU boxes are full"
827 if not vnode_hostname: msg += " vnode IP pool exhausted"
828 raise Exception,"Could not make space for a QEMU instance:"+msg
829 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
830 freed_vnode_hostname=self.short_hostname(qemu_instance_to_kill.nodename)
832 message='killing oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
834 qemu_instance_to_kill.kill()
835 # use these freed resources where needed
837 qemu_boxname=freed_qemu_boxname
838 if not vnode_hostname:
839 vnode_hostname=freed_vnode_hostname
840 self.vnode_pool.set_mine(vnode_hostname)
842 self.add_dummy_qemu (qemu_boxname,nodename)
843 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
844 ip=self.vnode_pool.get_ip (vnode_hostname)
845 self.vnode_pool.add_starting(vnode_hostname)
847 vnode_fqdn = self.fqdn(vnode_hostname)
848 nodemap={'host_box':qemu_boxname,
849 'node_fields:hostname':vnode_fqdn,
850 'interface_fields:ip':ip,
851 'interface_fields:mac':mac,
853 nodemap.update(self.network_settings())
854 maps.append ( (nodename, nodemap) )
856 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
857 (nodename,qemu_boxname,vnode_hostname,mac))
859 return test_mapper.map({'node':maps})[0]
861 def localize_sfa_rspec (self,plc,options):
863 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
864 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
865 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
866 plc['sfa']['SFA_PLC_DB_HOST'] = plc['PLC_DB_HOST']
867 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
868 for site in plc['sites']:
869 for node in site['nodes']:
870 plc['sfa']['sfa_slice_rspec']['part4'] = node['node_fields']['hostname']
873 #################### release:
874 def release (self,options):
875 self.vplc_pool.release_my_starting()
876 self.vnode_pool.release_my_starting()
879 #################### show results for interactive mode
880 def get_box (self,box):
881 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes:
882 if b.short_hostname()==box:
884 print "Could not find box %s"%box
889 for b in self.all_boxes: b.list()
891 def list_boxes(self,boxes):
892 print 'Partial Sensing',
896 b.sense(self.options)
904 # can be run as a utility to manage the local infrastructure
906 parser=OptionParser()
907 parser.add_option ('-r',"--reboot",action='store_true',dest='reboot',default=False,
908 help='reboot mode (use shutdown -r)')
909 parser.add_option ('-s',"--soft",action='store_true',dest='soft',default=False,
910 help='soft mode for reboot (vserver stop or kill qemus)')
911 parser.add_option ('-b',"--build",action='store_true',dest='builds',default=False,
912 help='add build boxes')
913 parser.add_option ('-p',"--plc",action='store_true',dest='plcs',default=False,
914 help='add plc boxes')
915 parser.add_option ('-q',"--qemu",action='store_true',dest='qemus',default=False,
916 help='add qemu boxes')
917 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
919 parser.add_option ('-n',"--dry_run",action='store_true',dest='dry_run',default=False,
921 (self.options,args)=parser.parse_args()
924 if self.options.builds: boxes += [b.hostname for b in self.build_boxes]
925 if self.options.plcs: boxes += [b.hostname for b in self.plc_boxes]
926 if self.options.qemus: boxes += [b.hostname for b in self.qemu_boxes]
927 boxes=list(set(boxes))
932 self.list_boxes (boxes)