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 # too painful to propagate this cleanly
65 def header (message,banner=True):
66 if not message: return
67 if banner: print "===============",
71 def timestamp_sort(o1,o2): return o1.timestamp-o2.timestamp
73 def short_hostname (hostname):
74 return hostname.split('.')[0]
77 # the place were other test instances tell about their not-yet-started
78 # instances, that go undetected through sensing
81 location='/root/starting'
86 try: self.tuples=[line.strip().split('@')
87 for line in file(Starting.location).readlines()]
88 except: self.tuples=[]
92 return [ x for (x,_) in self.tuples ]
94 def add (self, vname, bname):
95 if not vname in self.vnames():
96 file(Starting.location,'a').write("%s@%s\n"%(vname,bname))
98 def delete_vname (self, vname):
100 if vname in self.vnames():
101 f=file(Starting.location,'w')
102 for (v,b) in self.tuples:
103 if v != vname: f.write("%s@%s\n"%(v,b))
108 # allows to pick an available IP among a pool
109 # input is expressed as a list of tuples (hostname,ip,user_data)
110 # that can be searched iteratively for a free slot
112 # pool = [ (hostname1,user_data1),
113 # (hostname2,user_data2),
114 # (hostname3,user_data2),
115 # (hostname4,user_data4) ]
116 # assuming that ip1 and ip3 are taken (pingable), then we'd get
118 # pool.next_free() -> entry2
119 # pool.next_free() -> entry4
120 # pool.next_free() -> None
121 # that is, even if ip2 is not busy/pingable when the second next_free() is issued
124 def __init__ (self,hostname,userdata):
125 self.hostname=hostname
126 self.userdata=userdata
127 # slot holds 'busy' or 'free' or 'mine' or 'starting' or None
128 # 'mine' is for our own stuff, 'starting' from the concurrent tests
133 return "Pooled %s (%s) -> %s"%(self.hostname,self.userdata, self.status)
136 if self.status==None: return '?'
137 elif self.status=='busy': return '+'
138 elif self.status=='free': return '-'
139 elif self.status=='mine': return 'M'
140 elif self.status=='starting': return 'S'
143 if self.ip: return self.ip
144 ip=socket.gethostbyname(self.hostname)
150 def __init__ (self, tuples,message, substrate):
151 self.pool_items= [ PoolItem (hostname,userdata) for (hostname,userdata) in tuples ]
153 # where to send notifications upon load_starting
154 self.substrate=substrate
156 def list (self, verbose=False):
157 for i in self.pool_items: print i.line()
161 for i in self.pool_items: line += ' ' + i.char()
164 def _item (self, hostname):
165 for i in self.pool_items:
166 if i.hostname==hostname: return i
167 raise Exception ("Could not locate hostname %s in pool %s"%(hostname,self.message))
169 def retrieve_userdata (self, hostname):
170 return self._item(hostname).userdata
172 def get_ip (self, hostname):
173 try: return self._item(hostname).get_ip()
174 except: return socket.gethostbyname(hostname)
176 def set_mine (self, hostname):
178 self._item(hostname).status='mine'
180 print 'WARNING: host %s not found in IP pool %s'%(hostname,self.message)
182 def next_free (self):
183 for i in self.pool_items:
184 if i.status == 'free':
186 return (i.hostname,i.userdata)
190 # we have a starting instance of our own
191 def add_starting (self, vname, bname):
192 Starting().add(vname,bname)
193 for i in self.pool_items:
194 if i.hostname==vname: i.status='mine'
196 # load the starting instances from the common file
197 # remember that might be ours
198 # return the list of (vname,bname) that are not ours
199 def load_starting (self):
203 for (v,b) in starting.tuples:
204 for i in self.pool_items:
205 if i.hostname==v and i.status=='free':
207 new_tuples.append( (v,b,) )
210 def release_my_starting (self):
211 for i in self.pool_items:
213 Starting().delete_vname (i.hostname)
219 for item in self.pool_items:
220 if item.status is not None:
223 if self.check_ping (item.hostname):
231 print 'Sensing IP pool',self.message,
234 for (vname,bname) in self.load_starting():
235 self.substrate.add_starting_dummy (bname, vname)
236 print 'After starting: IP pool'
238 # OS-dependent ping option (support for macos, for convenience)
239 ping_timeout_option = None
240 # returns True when a given hostname/ip responds to ping
241 def check_ping (self,hostname):
242 if not Pool.ping_timeout_option:
243 (status,osname) = commands.getstatusoutput("uname -s")
245 raise Exception, "TestPool: Cannot figure your OS name"
246 if osname == "Linux":
247 Pool.ping_timeout_option="-w"
248 elif osname == "Darwin":
249 Pool.ping_timeout_option="-t"
251 command="ping -c 1 %s 1 %s"%(Pool.ping_timeout_option,hostname)
252 (status,output) = commands.getstatusoutput(command)
257 def __init__ (self,hostname):
258 self.hostname=hostname
260 def shortname (self):
261 return short_hostname(self.hostname)
262 def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
263 def reboot (self, options):
264 self.test_ssh().run("shutdown -r now",message="Rebooting %s"%self.hostname,
265 dry_run=options.dry_run)
267 def hostname_fedora (self,virt=None):
268 result = "%s {"%self.hostname
269 if virt: result += "%s-"%virt
270 result += "%s"%self.fedora()
271 # too painful to propagate this cleanly
274 result += "-%s" % self.uname()
278 separator = "===composite==="
281 # take this chance to gather useful stuff
284 if self._probed is not None: return self._probed
285 composite_command = [ ]
286 composite_command += [ "hostname" ]
287 composite_command += [ ";" , "echo", Box.separator , ";" ]
288 composite_command += [ "uptime" ]
289 composite_command += [ ";" , "echo", Box.separator , ";" ]
290 composite_command += [ "uname", "-r"]
291 composite_command += [ ";" , "echo", Box.separator , ";" ]
292 composite_command += [ "cat" , "/etc/fedora-release" ]
294 # due to colons and all, this is going wrong on the local box (typically testmaster)
295 # I am reluctant to change TestSsh as it might break all over the place, so
296 if self.test_ssh().is_local():
297 probe_argv = [ "bash", "-c", " ".join (composite_command) ]
299 probe_argv=self.test_ssh().actual_argv(composite_command)
300 composite=self.backquote ( probe_argv, trash_err=True )
301 self._hostname = self._uptime = self._uname = self._fedora = "** Unknown **"
303 print "root@%s unreachable"%self.hostname
307 pieces = composite.split(Box.separator)
308 pieces = [ x.strip() for x in pieces ]
309 [self._hostname, self._uptime, self._uname, self._fedora] = pieces
311 self._uptime = ', '.join([ x.strip() for x in self._uptime.split(',')[2:]])
312 self._fedora = self._fedora.replace("Fedora release ","f").split(" ")[0]
315 print 'BEG issue with pieces',pieces
316 traceback.print_exc()
317 print 'END issue with pieces',pieces
318 self._probed=self._hostname
321 # use argv=['bash','-c',"the command line"]
324 if hasattr(self,'_uptime') and self._uptime: return self._uptime
325 return '*unprobed* uptime'
328 if hasattr(self,'_uname') and self._uname: return self._uname
329 return '*unprobed* uname'
332 if hasattr(self,'_fedora') and self._fedora: return self._fedora
333 return '*unprobed* fedora'
335 def run(self,argv,message=None,trash_err=False,dry_run=False):
343 return subprocess.call(argv)
345 return subprocess.call(argv,stderr=file('/dev/null','w'))
347 def run_ssh (self, argv, message, trash_err=False, dry_run=False):
348 ssh_argv = self.test_ssh().actual_argv(argv)
349 result=self.run (ssh_argv, message, trash_err, dry_run=dry_run)
351 print "WARNING: failed to run %s on %s"%(" ".join(argv),self.hostname)
354 def backquote (self, argv, trash_err=False):
355 # print 'running backquote',argv
357 result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
359 result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
362 # if you have any shell-expanded arguments like *
363 # and if there's any chance the command is adressed to the local host
364 def backquote_ssh (self, argv, trash_err=False):
365 if not self.probe(): return ''
366 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
368 ############################################################
370 def __init__ (self, buildname, pid, buildbox):
371 self.buildname=buildname
372 self.buildbox=buildbox
375 def add_pid(self,pid):
376 self.pids.append(pid)
379 return "== %s == (pids=%r)"%(self.buildname,self.pids)
381 class BuildBox (Box):
382 def __init__ (self,hostname):
383 Box.__init__(self,hostname)
384 self.build_instances=[]
386 def add_build (self,buildname,pid):
387 for build in self.build_instances:
388 if build.buildname==buildname:
391 self.build_instances.append(BuildInstance(buildname, pid, self))
393 def list(self, verbose=False):
394 if not self.build_instances:
395 header ('No build process on %s (%s)'%(self.hostname_fedora(),self.uptime()))
397 header ("Builds on %s (%s)"%(self.hostname_fedora(),self.uptime()))
398 for b in self.build_instances:
399 header (b.line(),banner=False)
401 def reboot (self, options):
403 Box.reboot(self,options)
405 self.soft_reboot (options)
407 build_matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
408 build_matcher_initvm=re.compile("\s*(?P<pid>[0-9]+).*initvm.*\s+(?P<buildname>[^\s]+)\s*\Z")
410 class BuildVsBox (BuildBox):
411 def soft_reboot (self, options):
412 command=['pkill','vbuild']
413 self.run_ssh(command,"Terminating vbuild processes",dry_run=options.dry_run)
415 # inspect box and find currently running builds
416 def sense(self, options):
418 pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
420 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
421 ps_lines=self.backquote_ssh (command).split('\n')
422 for line in ps_lines:
423 if not line.strip() or line.find('PID')>=0: continue
424 m=build_matcher.match(line)
426 date=time.strftime('%Y-%m-%d',time.localtime(time.time()))
427 buildname=m.group('buildname').replace('@DATE@',date)
428 self.add_build (buildname,m.group('pid'))
430 m=build_matcher_initvm.match(line)
432 # buildname is expansed here
433 self.add_build (buildname,m.group('pid'))
435 header('BuildVsBox.sense: command %r returned line that failed to match'%command)
436 header(">>%s<<"%line)
438 class BuildLxcBox (BuildBox):
439 def soft_reboot (self, options):
440 command=['pkill','lbuild']
441 self.run_ssh(command,"Terminating vbuild processes",dry_run=options.dry_run)
443 # inspect box and find currently running builds
444 def sense(self, options):
446 pids=self.backquote_ssh(['pgrep','lbuild'],trash_err=True)
448 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
449 ps_lines=self.backquote_ssh (command).split('\n')
450 for line in ps_lines:
451 if not line.strip() or line.find('PID')>=0: continue
452 m=build_matcher.match(line)
454 date=time.strftime('%Y-%m-%d',time.localtime(time.time()))
455 buildname=m.group('buildname').replace('@DATE@',date)
456 self.add_build (buildname,m.group('pid'))
458 m=build_matcher_initvm.match(line)
460 # buildname is expansed here
461 self.add_build (buildname,m.group('pid'))
463 header('BuildLxcBox.sense: command %r returned line that failed to match'%command)
464 header(">>%s<<"%line)
466 ############################################################
468 def __init__ (self, plcbox):
473 def set_timestamp (self,timestamp): self.timestamp=timestamp
474 def set_now (self): self.timestamp=int(time.time())
475 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
477 class PlcVsInstance (PlcInstance):
478 def __init__ (self, plcbox, vservername, ctxid):
479 PlcInstance.__init__(self,plcbox)
480 self.vservername=vservername
484 return self.vservername.split('-')[-1]
485 def buildname (self):
486 return self.vservername.rsplit('-',2)[0]
489 msg="== %s =="%(self.vplcname())
490 msg += " [=%s]"%self.vservername
491 if self.ctxid==0: msg+=" not (yet?) running"
492 else: msg+=" (ctx=%s)"%self.ctxid
493 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
494 else: msg += " *unknown timestamp*"
498 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
499 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
500 self.plc_box.forget(self)
502 class PlcLxcInstance (PlcInstance):
503 # does lxc have a context id of any kind ?
504 def __init__ (self, plcbox, lxcname, pid):
505 PlcInstance.__init__(self, plcbox)
506 self.lxcname = lxcname
510 return self.lxcname.split('-')[-1]
511 def buildname (self):
512 return self.lxcname.rsplit('-',2)[0]
515 msg="== %s =="%(self.vplcname())
516 msg += " [=%s]"%self.lxcname
517 if self.pid==-1: msg+=" not (yet?) running"
518 else: msg+=" (pid=%s)"%self.pid
519 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
520 else: msg += " *unknown timestamp*"
524 command="rsync lxc-driver.sh %s:/root"%self.plc_box.hostname
525 commands.getstatusoutput(command)
526 msg="lxc container stopping %s on %s"%(self.lxcname,self.plc_box.hostname)
527 self.plc_box.run_ssh(['/root/lxc-driver.sh','-c','stop_lxc','-n',self.lxcname],msg)
528 self.plc_box.forget(self)
532 def __init__ (self, hostname, max_plcs):
533 Box.__init__(self,hostname)
534 self.plc_instances=[]
535 self.max_plcs=max_plcs
537 def free_slots (self):
538 return self.max_plcs - len(self.plc_instances)
540 # fill one slot even though this one is not started yet
541 def add_dummy (self, plcname):
542 dummy=PlcVsInstance(self,'dummy_'+plcname,0)
544 self.plc_instances.append(dummy)
546 def forget (self, plc_instance):
547 self.plc_instances.remove(plc_instance)
549 def reboot (self, options):
551 Box.reboot(self,options)
553 self.soft_reboot (options)
555 def list(self, verbose=False):
556 if not self.plc_instances:
557 header ('No plc running on %s'%(self.line()))
559 header ("Active plc VMs on %s"%self.line())
560 self.plc_instances.sort(timestamp_sort)
561 for p in self.plc_instances:
562 header (p.line(),banner=False)
564 # we do not this at INRIA any more
565 class PlcVsBox (PlcBox):
567 def add_vserver (self,vservername,ctxid):
568 for plc in self.plc_instances:
569 if plc.vservername==vservername:
570 header("WARNING, duplicate myplc %s running on %s"%\
571 (vservername,self.hostname),banner=False)
573 self.plc_instances.append(PlcVsInstance(self,vservername,ctxid))
576 msg="%s [max=%d,free=%d] (%s)"%(self.hostname_fedora(virt="vs"), self.max_plcs,self.free_slots(),self.uptime())
579 def plc_instance_by_vservername (self, vservername):
580 for p in self.plc_instances:
581 if p.vservername==vservername: return p
584 def soft_reboot (self, options):
585 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers on %s"%(self.hostname,),
586 dry_run=options.dry_run)
588 def sense (self, options):
590 # try to find fullname (vserver_stat truncates to a ridiculously short name)
591 # fetch the contexts for all vservers on that box
592 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
593 context_map=self.backquote_ssh (map_command)
594 # at this point we have a set of lines like
595 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
597 for map_line in context_map.split("\n"):
598 if not map_line: continue
599 [path,xid] = map_line.split(':')
600 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
601 # at this point ctx_id maps context id to vservername
603 command=['vserver-stat']
604 vserver_stat = self.backquote_ssh (command)
605 for vserver_line in vserver_stat.split("\n"):
606 if not vserver_line: continue
607 context=vserver_line.split()[0]
608 if context=="CTX": continue
610 longname=ctx_dict[context]
611 self.add_vserver(longname,context)
613 print 'WARNING: found ctx %s in vserver_stat but was unable to figure a corresp. vserver'%context
616 running_vsnames = [ i.vservername for i in self.plc_instances ]
617 command= ['grep','.']
618 command += ['/vservers/%s.timestamp'%vs for vs in running_vsnames]
619 command += ['/dev/null']
620 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
621 for ts_line in ts_lines:
622 if not ts_line.strip(): continue
623 # expect /vservers/<vservername>.timestamp:<timestamp>
625 (ts_file,timestamp)=ts_line.split(':')
626 ts_file=os.path.basename(ts_file)
627 (vservername,_)=os.path.splitext(ts_file)
628 timestamp=int(timestamp)
629 p=self.plc_instance_by_vservername(vservername)
631 print 'WARNING zombie plc',self.hostname,ts_line
632 print '... was expecting',vservername,'in',[i.vservername for i in self.plc_instances]
634 p.set_timestamp(timestamp)
635 except: print 'WARNING, could not parse ts line',ts_line
638 class PlcLxcBox (PlcBox):
640 def add_lxc (self,lxcname,pid):
641 for plc in self.plc_instances:
642 if plc.lxcname==lxcname:
643 header("WARNING, duplicate myplc %s running on %s"%\
644 (lxcname,self.hostname),banner=False)
646 self.plc_instances.append(PlcLxcInstance(self,lxcname,pid))
649 # a line describing the box
651 return "%s [max=%d,free=%d] (%s)"%(self.hostname_fedora(virt="lxc"),
652 self.max_plcs,self.free_slots(),
656 def plc_instance_by_lxcname (self, lxcname):
657 for p in self.plc_instances:
658 if p.lxcname==lxcname: return p
661 # essentially shutdown all running containers
662 def soft_reboot (self, options):
663 command="rsync lxc-driver.sh %s:/root"%self.hostname
664 commands.getstatusoutput(command)
665 self.run_ssh(['/root/lxc-driver.sh','-c','stop_all'],"Stopping all running lxc containers on %s"%(self.hostname,),
666 dry_run=options.dry_run)
669 # sense is expected to fill self.plc_instances with PlcLxcInstance's
670 # to describe the currently running VM's
671 def sense (self, options):
673 command="rsync lxc-driver.sh %s:/root"%self.hostname
674 commands.getstatusoutput(command)
675 command=['/root/lxc-driver.sh','-c','sense_all']
676 lxc_stat = self.backquote_ssh (command)
677 for lxc_line in lxc_stat.split("\n"):
678 if not lxc_line: continue
679 lxcname=lxc_line.split(";")[0]
680 pid=lxc_line.split(";")[1]
681 timestamp=lxc_line.split(";")[2]
682 self.add_lxc(lxcname,pid)
683 try: timestamp=int(timestamp)
685 p=self.plc_instance_by_lxcname(lxcname)
687 print 'WARNING zombie plc',self.hostname,lxcname
688 print '... was expecting',lxcname,'in',[i.lxcname for i in self.plc_instances]
690 p.set_timestamp(timestamp)
692 ############################################################
694 def __init__ (self, nodename, pid, qemubox):
695 self.nodename=nodename
697 self.qemu_box=qemubox
702 def set_buildname (self,buildname): self.buildname=buildname
703 def set_timestamp (self,timestamp): self.timestamp=timestamp
704 def set_now (self): self.timestamp=int(time.time())
705 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
708 msg = "== %s =="%(short_hostname(self.nodename))
709 msg += " [=%s]"%self.buildname
710 if self.pid: msg += " (pid=%s)"%self.pid
711 else: msg += " not (yet?) running"
712 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
713 else: msg += " *unknown timestamp*"
718 print "cannot kill qemu %s with pid==0"%self.nodename
720 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
721 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
722 self.qemu_box.forget(self)
726 def __init__ (self, hostname, max_qemus):
727 Box.__init__(self,hostname)
728 self.qemu_instances=[]
729 self.max_qemus=max_qemus
731 def add_node (self,nodename,pid):
732 for qemu in self.qemu_instances:
733 if qemu.nodename==nodename:
734 header("WARNING, duplicate qemu %s running on %s"%\
735 (nodename,self.hostname), banner=False)
737 self.qemu_instances.append(QemuInstance(nodename,pid,self))
739 def forget (self, qemu_instance):
740 self.qemu_instances.remove(qemu_instance)
742 # fill one slot even though this one is not started yet
743 def add_dummy (self, nodename):
744 dummy=QemuInstance('dummy_'+nodename,0,self)
746 self.qemu_instances.append(dummy)
749 return "%s [max=%d,free=%d] (%s) %s"%(
750 self.hostname_fedora(virt="qemu"), self.max_qemus,self.free_slots(),
751 self.uptime(),self.driver())
753 def list(self, verbose=False):
754 if not self.qemu_instances:
755 header ('No qemu on %s'%(self.line()))
757 header ("Qemus on %s"%(self.line()))
758 self.qemu_instances.sort(timestamp_sort)
759 for q in self.qemu_instances:
760 header (q.line(),banner=False)
762 def free_slots (self):
763 return self.max_qemus - len(self.qemu_instances)
766 if hasattr(self,'_driver') and self._driver: return self._driver
767 return '*undef* driver'
769 def qemu_instance_by_pid (self,pid):
770 for q in self.qemu_instances:
771 if q.pid==pid: return q
774 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
775 for q in self.qemu_instances:
776 if q.nodename==nodename and q.buildname==buildname:
780 def reboot (self, options):
782 Box.reboot(self,options)
784 self.run_ssh(['pkill','qemu'],"Killing qemu instances",
785 dry_run=options.dry_run)
787 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
788 def sense(self, options):
790 modules=self.backquote_ssh(['lsmod']).split('\n')
791 self._driver='*NO kqemu/kvm_intel MODULE LOADED*'
792 for module in modules:
793 if module.find('kqemu')==0:
794 self._driver='kqemu module loaded'
795 # kvm might be loaded without kvm_intel (we dont have AMD)
796 elif module.find('kvm_intel')==0:
797 self._driver='kvm_intel OK'
798 ########## find out running pids
799 pids=self.backquote_ssh(['pgrep','qemu'])
801 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
802 ps_lines = self.backquote_ssh (command).split("\n")
803 for line in ps_lines:
804 if not line.strip() or line.find('PID') >=0 : continue
805 m=QemuBox.matcher.match(line)
807 self.add_node (m.group('nodename'),m.group('pid'))
809 header('QemuBox.sense: command %r returned line that failed to match'%command)
810 header(">>%s<<"%line)
811 ########## retrieve alive instances and map to build
813 command=['grep','.','/vservers/*/*/qemu.pid','/dev/null']
814 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
815 for pid_line in pid_lines:
816 if not pid_line.strip(): continue
817 # expect <build>/<nodename>/qemu.pid:<pid>pid
819 (_,__,buildname,nodename,tail)=pid_line.split('/')
820 (_,pid)=tail.split(':')
821 q=self.qemu_instance_by_pid (pid)
823 q.set_buildname(buildname)
824 live_builds.append(buildname)
825 except: print 'WARNING, could not parse pid line',pid_line
826 # retrieve timestamps
827 if not live_builds: return
828 command= ['grep','.']
829 command += ['/vservers/%s/*/timestamp'%b for b in live_builds]
830 command += ['/dev/null']
831 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
832 for ts_line in ts_lines:
833 if not ts_line.strip(): continue
834 # expect <build>/<nodename>/timestamp:<timestamp>
836 (_,__,buildname,nodename,tail)=ts_line.split('/')
837 nodename=nodename.replace('qemu-','')
838 (_,timestamp)=tail.split(':')
839 timestamp=int(timestamp)
840 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
842 print 'WARNING zombie qemu',self.hostname,ts_line
843 print '... was expecting (',short_hostname(nodename),buildname,') in',\
844 [ (short_hostname(i.nodename),i.buildname) for i in self.qemu_instances ]
846 q.set_timestamp(timestamp)
847 except: print 'WARNING, could not parse ts line',ts_line
851 def __init__ (self, buildname, pid=0):
853 if pid!=0: self.pid.append(pid)
854 self.buildname=buildname
861 def set_timestamp (self,timestamp): self.timestamp=timestamp
862 def set_now (self): self.timestamp=int(time.time())
863 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
865 def is_running (self): return len(self.pids) != 0
867 def add_pid (self,pid):
868 self.pids.append(pid)
869 def set_broken (self, plcindex, step):
870 self.broken_steps.append ( (plcindex, step,) )
872 def second_letter (self):
873 if not self.broken_steps: return '='
875 really_broken = [ step for (i,step) in self.broken_steps if '_ignore' not in step ]
876 # W is for warning like what's in the build mail
877 if len(really_broken)==0: return 'W'
881 # make up a 2-letter sign
882 # first letter : '=', unless build is running : '*'
883 double = '*' if self.pids else '='
884 # second letter : '=' if fine, 'W' for warnings (only ignored steps) 'B' for broken
885 double += self.second_letter()
886 msg = " %s %s =="%(double,self.buildname)
887 if not self.pids: pass
888 elif len(self.pids)==1: msg += " (pid=%s)"%self.pids[0]
889 else: msg += " !!!pids=%s!!!"%self.pids
890 msg += " @%s"%self.pretty_timestamp()
891 if self.broken_steps:
892 # sometimes we have an empty plcindex
893 msg += " [BROKEN=" + " ".join( [ "%s@%s"%(s,i) if i else s for (i,s) in self.broken_steps ] ) + "]"
897 def __init__ (self,hostname):
898 Box.__init__(self,hostname)
900 self.test_instances=[]
902 def reboot (self, options):
903 # can't reboot a vserver VM
904 self.run_ssh (['pkill','run_log'],"Terminating current runs",
905 dry_run=options.dry_run)
906 self.run_ssh (['rm','-f',Starting.location],"Cleaning %s"%Starting.location,
907 dry_run=options.dry_run)
909 def get_test (self, buildname):
910 for i in self.test_instances:
911 if i.buildname==buildname: return i
913 # we scan ALL remaining test results, even the ones not running
914 def add_timestamp (self, buildname, timestamp):
915 i=self.get_test(buildname)
917 i.set_timestamp(timestamp)
919 i=TestInstance(buildname,0)
920 i.set_timestamp(timestamp)
921 self.test_instances.append(i)
923 def add_running_test (self, pid, buildname):
924 i=self.get_test(buildname)
926 self.test_instances.append (TestInstance (buildname,pid))
929 print "WARNING: 2 concurrent tests run on same build %s"%buildname
932 def add_broken (self, buildname, plcindex, step):
933 i=self.get_test(buildname)
935 i=TestInstance(buildname)
936 self.test_instances.append(i)
937 i.set_broken(plcindex, step)
939 matcher_proc=re.compile (".*/proc/(?P<pid>[0-9]+)/cwd.*/root/(?P<buildname>[^/]+)$")
940 matcher_grep=re.compile ("/root/(?P<buildname>[^/]+)/logs/trace.*:TRACE:\s*(?P<plcindex>[0-9]+).*step=(?P<step>\S+).*")
941 matcher_grep_missing=re.compile ("grep: /root/(?P<buildname>[^/]+)/logs/trace: No such file or directory")
942 def sense (self, options):
944 self.starting_ips=[x for x in self.backquote_ssh(['cat',Starting.location], trash_err=True).strip().split('\n') if x]
946 # scan timestamps on all tests
947 # this is likely to not invoke ssh so we need to be a bit smarter to get * expanded
948 # xxx would make sense above too
949 command=['bash','-c',"grep . /root/*/timestamp /dev/null"]
950 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
951 for ts_line in ts_lines:
952 if not ts_line.strip(): continue
953 # expect /root/<buildname>/timestamp:<timestamp>
955 (ts_file,timestamp)=ts_line.split(':')
956 ts_file=os.path.dirname(ts_file)
957 buildname=os.path.basename(ts_file)
958 timestamp=int(timestamp)
959 t=self.add_timestamp(buildname,timestamp)
960 except: print 'WARNING, could not parse ts line',ts_line
962 # let's try to be robust here -- tests that fail very early like e.g.
963 # "Cannot make space for a PLC instance: vplc IP pool exhausted", that occurs as part of provision
964 # will result in a 'trace' symlink to an inexisting 'trace-<>.txt' because no step has gone through
965 # simple 'trace' should exist though as it is created by run_log
966 command=['bash','-c',"grep KO /root/*/logs/trace /dev/null 2>&1" ]
967 trace_lines=self.backquote_ssh (command).split('\n')
968 for line in trace_lines:
969 if not line.strip(): continue
970 m=TestBox.matcher_grep_missing.match(line)
972 buildname=m.group('buildname')
973 self.add_broken(buildname,'','NO STEP DONE')
975 m=TestBox.matcher_grep.match(line)
977 buildname=m.group('buildname')
978 plcindex=m.group('plcindex')
980 self.add_broken(buildname,plcindex, step)
982 header("TestBox.sense: command %r returned line that failed to match\n%s"%(command,line))
983 header(">>%s<<"%line)
985 pids = self.backquote_ssh (['pgrep','run_log'],trash_err=True)
987 command=['ls','-ld'] + ["/proc/%s/cwd"%pid for pid in pids.split("\n") if pid]
988 ps_lines=self.backquote_ssh (command).split('\n')
989 for line in ps_lines:
990 if not line.strip(): continue
991 m=TestBox.matcher_proc.match(line)
994 buildname=m.group('buildname')
995 self.add_running_test(pid, buildname)
997 header("TestBox.sense: command %r returned line that failed to match\n%s"%(command,line))
998 header(">>%s<<"%line)
1002 return self.hostname_fedora()
1004 def list (self, verbose=False):
1005 # verbose shows all tests
1007 instances = self.test_instances
1010 instances = [ i for i in self.test_instances if i.is_running() ]
1014 header ("No %s on %s"%(msg,self.line()))
1016 header ("%s on %s"%(msg,self.line()))
1017 instances.sort(timestamp_sort)
1018 for i in instances: print i.line()
1019 # show 'starting' regardless of verbose
1020 if self.starting_ips:
1021 header ("Starting IP addresses on %s"%self.line())
1022 self.starting_ips.sort()
1023 for starting in self.starting_ips: print starting
1025 header ("Empty 'starting' on %s"%self.line())
1027 ############################################################
1032 def __init__ (self, plcs_on_vs=True, plcs_on_lxc=False):
1033 self.options=Options()
1034 self.options.dry_run=False
1035 self.options.verbose=False
1036 self.options.reboot=False
1037 self.options.soft=False
1038 self.test_box = TestBox (self.test_box_spec())
1039 self.build_vs_boxes = [ BuildVsBox(h) for h in self.build_vs_boxes_spec() ]
1040 self.build_lxc_boxes = [ BuildLxcBox(h) for h in self.build_lxc_boxes_spec() ]
1041 self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_vs_boxes_spec ()]
1042 self.plc_lxc_boxes = [ PlcLxcBox (h,m) for (h,m) in self.plc_lxc_boxes_spec ()]
1043 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
1046 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs",self)
1047 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes",self)
1049 self.rescope (plcs_on_vs=plcs_on_vs, plcs_on_lxc=plcs_on_lxc)
1051 # which plc boxes are we interested in ?
1052 def rescope (self, plcs_on_vs, plcs_on_lxc):
1053 self.build_boxes = self.build_vs_boxes + self.build_lxc_boxes
1055 if plcs_on_vs: self.plc_boxes += self.plc_vs_boxes
1056 if plcs_on_lxc: self.plc_boxes += self.plc_lxc_boxes
1057 self.default_boxes = self.plc_boxes + self.qemu_boxes
1058 self.all_boxes = self.build_boxes + [ self.test_box ] + self.plc_boxes + self.qemu_boxes
1060 def summary_line (self):
1062 msg += " %d vp"%len(self.plc_vs_boxes)
1063 msg += " %d xp"%len(self.plc_lxc_boxes)
1064 msg += " %d tried plc boxes"%len(self.plc_boxes)
1068 def fqdn (self, hostname):
1069 if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
1072 # return True if actual sensing takes place
1073 def sense (self,force=False):
1074 if self._sensed and not force: return False
1075 print 'Sensing local substrate...',
1076 for b in self.default_boxes: b.sense(self.options)
1081 def list (self, verbose=False):
1082 for b in self.default_boxes:
1085 def add_dummy_plc (self, plc_boxname, plcname):
1086 for pb in self.plc_boxes:
1087 if pb.hostname==plc_boxname:
1088 pb.add_dummy(plcname)
1090 def add_dummy_qemu (self, qemu_boxname, qemuname):
1091 for qb in self.qemu_boxes:
1092 if qb.hostname==qemu_boxname:
1093 qb.add_dummy(qemuname)
1096 def add_starting_dummy (self, bname, vname):
1097 return self.add_dummy_plc (bname, vname) or self.add_dummy_qemu (bname, vname)
1100 def provision (self,plcs,options):
1102 # attach each plc to a plc box and an IP address
1103 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
1104 # attach each node/qemu to a qemu box with an IP address
1105 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
1106 # update the SFA spec accordingly
1107 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
1110 except Exception, e:
1111 print '* Could not provision this test on current substrate','--',e,'--','exiting'
1112 traceback.print_exc()
1115 # it is expected that a couple of options like ips_bplc and ips_vplc
1116 # are set or unset together
1118 def check_options (x,y):
1119 if not x and not y: return True
1120 return len(x)==len(y)
1122 # find an available plc box (or make space)
1123 # and a free IP address (using options if present)
1124 def provision_plc (self, plc, options):
1126 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
1128 #### let's find an IP address for that plc
1130 if options.ips_vplc:
1132 # we don't check anything here,
1133 # it is the caller's responsability to cleanup and make sure this makes sense
1134 plc_boxname = options.ips_bplc.pop()
1135 vplc_hostname=options.ips_vplc.pop()
1137 if self.sense(): self.list()
1140 # try to find an available IP
1141 self.vplc_pool.sense()
1142 couple=self.vplc_pool.next_free()
1144 (vplc_hostname,unused)=couple
1145 #### we need to find one plc box that still has a slot
1147 # use the box that has max free spots for load balancing
1148 for pb in self.plc_boxes:
1149 free=pb.free_slots()
1151 plc_boxname=pb.hostname
1153 # if there's no available slot in the plc_boxes, or we need a free IP address
1154 # make space by killing the oldest running instance
1155 if not plc_boxname or not vplc_hostname:
1156 # find the oldest of all our instances
1157 all_plc_instances=reduce(lambda x, y: x+y,
1158 [ pb.plc_instances for pb in self.plc_boxes ],
1160 all_plc_instances.sort(timestamp_sort)
1162 plc_instance_to_kill=all_plc_instances[0]
1165 if not plc_boxname: msg += " PLC boxes are full"
1166 if not vplc_hostname: msg += " vplc IP pool exhausted"
1167 msg += " %s"%self.summary_line()
1168 raise Exception,"Cannot make space for a PLC instance:"+msg
1169 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
1170 freed_vplc_hostname=plc_instance_to_kill.vplcname()
1171 message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
1173 plc_instance_to_kill.kill()
1174 # use this new plcbox if that was the problem
1176 plc_boxname=freed_plc_boxname
1177 # ditto for the IP address
1178 if not vplc_hostname:
1179 vplc_hostname=freed_vplc_hostname
1180 # record in pool as mine
1181 self.vplc_pool.set_mine(vplc_hostname)
1184 self.add_dummy_plc(plc_boxname,plc['name'])
1185 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
1186 self.vplc_pool.add_starting(vplc_hostname, plc_boxname)
1188 #### compute a helpful vserver name
1189 # remove domain in hostname
1190 vplc_short = short_hostname(vplc_hostname)
1191 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_short)
1192 plc_name = "%s_%s"%(plc['name'],vplc_short)
1194 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
1195 (plc['name'],plc_boxname,vplc_hostname,vservername))
1197 #### apply in the plc_spec
1199 # label=options.personality.replace("linux","")
1200 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
1201 # 'name':'%s-'+label,
1203 'vservername':vservername,
1204 'vserverip':vplc_ip,
1205 'PLC_DB_HOST':vplc_hostname,
1206 'PLC_API_HOST':vplc_hostname,
1207 'PLC_BOOT_HOST':vplc_hostname,
1208 'PLC_WWW_HOST':vplc_hostname,
1209 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
1210 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
1215 # mappers only work on a list of plcs
1216 return TestMapper([plc],options).map(mapper)[0]
1219 def provision_qemus (self, plc, options):
1221 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
1223 test_mapper = TestMapper ([plc], options)
1224 nodenames = test_mapper.node_names()
1226 for nodename in nodenames:
1228 if options.ips_vnode:
1229 # as above, it's a rerun, take it for granted
1230 qemu_boxname=options.ips_bnode.pop()
1231 vnode_hostname=options.ips_vnode.pop()
1233 if self.sense(): self.list()
1236 # try to find an available IP
1237 self.vnode_pool.sense()
1238 couple=self.vnode_pool.next_free()
1240 (vnode_hostname,unused)=couple
1241 # find a physical box
1243 # use the box that has max free spots for load balancing
1244 for qb in self.qemu_boxes:
1245 free=qb.free_slots()
1247 qemu_boxname=qb.hostname
1249 # if we miss the box or the IP, kill the oldest instance
1250 if not qemu_boxname or not vnode_hostname:
1251 # find the oldest of all our instances
1252 all_qemu_instances=reduce(lambda x, y: x+y,
1253 [ qb.qemu_instances for qb in self.qemu_boxes ],
1255 all_qemu_instances.sort(timestamp_sort)
1257 qemu_instance_to_kill=all_qemu_instances[0]
1260 if not qemu_boxname: msg += " QEMU boxes are full"
1261 if not vnode_hostname: msg += " vnode IP pool exhausted"
1262 msg += " %s"%self.summary_line()
1263 raise Exception,"Cannot make space for a QEMU instance:"+msg
1264 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
1265 freed_vnode_hostname=short_hostname(qemu_instance_to_kill.nodename)
1267 message='killing oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
1269 qemu_instance_to_kill.kill()
1270 # use these freed resources where needed
1271 if not qemu_boxname:
1272 qemu_boxname=freed_qemu_boxname
1273 if not vnode_hostname:
1274 vnode_hostname=freed_vnode_hostname
1275 self.vnode_pool.set_mine(vnode_hostname)
1277 self.add_dummy_qemu (qemu_boxname,vnode_hostname)
1278 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
1279 ip=self.vnode_pool.get_ip (vnode_hostname)
1280 self.vnode_pool.add_starting(vnode_hostname,qemu_boxname)
1282 vnode_fqdn = self.fqdn(vnode_hostname)
1283 nodemap={'host_box':qemu_boxname,
1284 'node_fields:hostname':vnode_fqdn,
1285 'interface_fields:ip':ip,
1286 'ipaddress_fields:ip_addr':ip,
1287 'interface_fields:mac':mac,
1289 nodemap.update(self.network_settings())
1290 maps.append ( (nodename, nodemap) )
1292 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
1293 (nodename,qemu_boxname,vnode_hostname,mac))
1295 return test_mapper.map({'node':maps})[0]
1297 def localize_sfa_rspec (self,plc,options):
1299 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
1300 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
1301 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
1302 plc['sfa']['SFA_DB_HOST'] = plc['PLC_DB_HOST']
1303 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
1306 #################### release:
1307 def release (self,options):
1308 self.vplc_pool.release_my_starting()
1309 self.vnode_pool.release_my_starting()
1312 #################### show results for interactive mode
1313 def get_box (self,boxname):
1314 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes + [self.test_box] :
1315 if b.shortname()==boxname: return b
1317 if b.shortname()==boxname.split('.')[0]: return b
1319 print "Could not find box %s"%boxname
1322 def list_boxes(self,box_or_names):
1324 for box in box_or_names:
1325 if not isinstance(box,Box): box=self.get_box(box)
1326 if not box: continue
1327 box.sense(self.options)
1329 for box in box_or_names:
1330 if not isinstance(box,Box): box=self.get_box(box)
1331 if not box: continue
1332 box.list(self.options.verbose)
1334 def reboot_boxes(self,box_or_names):
1335 for box in box_or_names:
1336 if not isinstance(box,Box): box=self.get_box(box)
1337 if not box: continue
1338 box.reboot(self.options)
1340 ####################
1341 # can be run as a utility to probe/display/manage the local infrastructure
1343 parser=OptionParser()
1344 parser.add_option ('-r',"--reboot",action='store_true',dest='reboot',default=False,
1345 help='reboot mode (use shutdown -r)')
1346 parser.add_option ('-s',"--soft",action='store_true',dest='soft',default=False,
1347 help='soft mode for reboot (vserver stop or kill qemus)')
1348 parser.add_option ('-t',"--testbox",action='store_true',dest='testbox',default=False,
1349 help='add test box')
1350 parser.add_option ('-b',"--build",action='store_true',dest='builds',default=False,
1351 help='add build boxes')
1352 parser.add_option ('-p',"--plc",action='store_true',dest='plcs',default=False,
1353 help='add plc boxes')
1354 parser.add_option ('-q',"--qemu",action='store_true',dest='qemus',default=False,
1355 help='add qemu boxes')
1356 parser.add_option ('-a',"--all",action='store_true',dest='all',default=False,
1357 help='address all known boxes, like -b -t -p -q')
1358 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
1359 help='verbose mode')
1360 parser.add_option ('-n',"--dry_run",action='store_true',dest='dry_run',default=False,
1361 help='dry run mode')
1362 (self.options,args)=parser.parse_args()
1364 self.rescope (plcs_on_vs=True, plcs_on_lxc=True)
1367 if self.options.testbox: boxes += [self.test_box]
1368 if self.options.builds: boxes += self.build_boxes
1369 if self.options.plcs: boxes += self.plc_boxes
1370 if self.options.qemus: boxes += self.qemu_boxes
1371 if self.options.all: boxes += self.all_boxes
1374 verbose=self.options.verbose
1375 # default scope is -b -p -q -t
1377 boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes + [self.test_box]
1379 if self.options.reboot: self.reboot_boxes (boxes)
1380 else: self.list_boxes (boxes)