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
70 def short_hostname (hostname):
71 return hostname.split('.')[0]
74 # the place were other test instances tell about their not-yet-started
75 # instances, that go undetected through sensing
78 location='/root/starting'
83 try: self.tuples=[line.strip().split('@')
84 for line in file(Starting.location).readlines()]
85 except: self.tuples=[]
89 return [ x for (x,_) in self.tuples ]
91 def add (self, vname, bname):
92 if not vname in self.vnames():
93 file(Starting.location,'a').write("%s@%s\n"%(vname,bname))
95 def delete_vname (self, vname):
97 if vname in self.vnames():
98 f=file(Starting.location,'w')
99 for (v,b) in self.tuples:
100 if v != vname: f.write("%s@%s\n"%(v,b))
105 # allows to pick an available IP among a pool
106 # input is expressed as a list of tuples (hostname,ip,user_data)
107 # that can be searched iteratively for a free slot
109 # pool = [ (hostname1,user_data1),
110 # (hostname2,user_data2),
111 # (hostname3,user_data2),
112 # (hostname4,user_data4) ]
113 # assuming that ip1 and ip3 are taken (pingable), then we'd get
115 # pool.next_free() -> entry2
116 # pool.next_free() -> entry4
117 # pool.next_free() -> None
118 # that is, even if ip2 is not busy/pingable when the second next_free() is issued
121 def __init__ (self,hostname,userdata):
122 self.hostname=hostname
123 self.userdata=userdata
124 # slot holds 'busy' or 'free' or 'mine' or 'starting' or None
125 # 'mine' is for our own stuff, 'starting' from the concurrent tests
130 return "Pooled %s (%s) -> %s"%(self.hostname,self.userdata, self.status)
133 if self.status==None: return '?'
134 elif self.status=='busy': return '+'
135 elif self.status=='free': return '-'
136 elif self.status=='mine': return 'M'
137 elif self.status=='starting': return 'S'
140 if self.ip: return self.ip
141 ip=socket.gethostbyname(self.hostname)
147 def __init__ (self, tuples,message, substrate):
148 self.pool_items= [ PoolItem (hostname,userdata) for (hostname,userdata) in tuples ]
150 # where to send notifications upon load_starting
151 self.substrate=substrate
153 def list (self, verbose=False):
154 for i in self.pool_items: print i.line()
158 for i in self.pool_items: line += ' ' + i.char()
161 def _item (self, hostname):
162 for i in self.pool_items:
163 if i.hostname==hostname: return i
164 raise Exception ("Could not locate hostname %s in pool %s"%(hostname,self.message))
166 def retrieve_userdata (self, hostname):
167 return self._item(hostname).userdata
169 def get_ip (self, hostname):
170 try: return self._item(hostname).get_ip()
171 except: return socket.gethostbyname(hostname)
173 def set_mine (self, hostname):
175 self._item(hostname).status='mine'
177 print 'WARNING: host %s not found in IP pool %s'%(hostname,self.message)
179 def next_free (self):
180 for i in self.pool_items:
181 if i.status == 'free':
183 return (i.hostname,i.userdata)
187 # we have a starting instance of our own
188 def add_starting (self, vname, bname):
189 Starting().add(vname,bname)
190 for i in self.pool_items:
191 if i.hostname==vname: i.status='mine'
193 # load the starting instances from the common file
194 # remember that might be ours
195 # return the list of (vname,bname) that are not ours
196 def load_starting (self):
200 for (v,b) in starting.tuples:
201 for i in self.pool_items:
202 if i.hostname==v and i.status=='free':
204 new_tuples.append( (v,b,) )
207 def release_my_starting (self):
208 for i in self.pool_items:
210 Starting().delete_vname (i.hostname)
216 for item in self.pool_items:
217 if item.status is not None:
220 if self.check_ping (item.hostname):
228 print 'Sensing IP pool',self.message,
231 for (vname,bname) in self.load_starting():
232 self.substrate.add_starting_dummy (bname, vname)
233 print 'After starting: IP pool'
235 # OS-dependent ping option (support for macos, for convenience)
236 ping_timeout_option = None
237 # returns True when a given hostname/ip responds to ping
238 def check_ping (self,hostname):
239 if not Pool.ping_timeout_option:
240 (status,osname) = commands.getstatusoutput("uname -s")
242 raise Exception, "TestPool: Cannot figure your OS name"
243 if osname == "Linux":
244 Pool.ping_timeout_option="-w"
245 elif osname == "Darwin":
246 Pool.ping_timeout_option="-t"
248 command="ping -c 1 %s 1 %s"%(Pool.ping_timeout_option,hostname)
249 (status,output) = commands.getstatusoutput(command)
254 def __init__ (self,hostname):
255 self.hostname=hostname
257 def shortname (self):
258 return short_hostname(self.hostname)
259 def test_ssh (self): return TestSsh(self.hostname,username='root',unknown_host=False)
260 def reboot (self, options):
261 self.test_ssh().run("shutdown -r now",message="Rebooting %s"%self.hostname,
262 dry_run=options.dry_run)
264 def hostname_fedora (self): return "%s [%s]"%(self.hostname,self.fedora())
266 separator = "===composite==="
269 # take this chance to gather useful stuff
272 if self._probed is not None: return self._probed
273 composite_command = [ ]
274 composite_command += [ "hostname" ]
275 composite_command += [ ";" , "echo", Box.separator , ";" ]
276 composite_command += [ "uptime" ]
277 composite_command += [ ";" , "echo", Box.separator , ";" ]
278 composite_command += [ "uname", "-r"]
279 composite_command += [ ";" , "echo", Box.separator , ";" ]
280 composite_command += [ "cat" , "/etc/fedora-release" ]
282 # due to colons and all, this is going wrong on the local box (typically testmaster)
283 # I am reluctant to change TestSsh as it might break all over the place, so
284 if self.test_ssh().is_local():
285 probe_argv = [ "bash", "-c", " ".join (composite_command) ]
287 probe_argv=self.test_ssh().actual_argv(composite_command)
288 composite=self.backquote ( probe_argv, trash_err=True )
289 if not composite: print "root@%s unreachable"%self.hostname
290 self._hostname = self._uptime = self._uname = self._fedora = "** Unknown **"
292 pieces = composite.split(Box.separator)
293 pieces = [ x.strip() for x in pieces ]
294 [self._hostname, self._uptime, self._uname, self._fedora] = pieces
296 self._uptime = ', '.join([ x.strip() for x in self._uptime.split(',')[2:]])
297 self._fedora = self._fedora.replace("Fedora release ","f").split(" ")[0]
300 print 'BEG issue with pieces',pieces
301 traceback.print_exc()
302 print 'END issue with pieces',pieces
303 self._probed=self._hostname
306 # use argv=['bash','-c',"the command line"]
309 if hasattr(self,'_uptime') and self._uptime: return self._uptime
310 return '*unprobed* uptime'
313 if hasattr(self,'_uname') and self._uname: return self._uname
314 return '*unprobed* uname'
317 if hasattr(self,'_fedora') and self._fedora: return self._fedora
318 return '*unprobed* fedora'
320 def run(self,argv,message=None,trash_err=False,dry_run=False):
328 return subprocess.call(argv)
330 return subprocess.call(argv,stderr=file('/dev/null','w'))
332 def run_ssh (self, argv, message, trash_err=False, dry_run=False):
333 ssh_argv = self.test_ssh().actual_argv(argv)
334 result=self.run (ssh_argv, message, trash_err, dry_run=dry_run)
336 print "WARNING: failed to run %s on %s"%(" ".join(argv),self.hostname)
339 def backquote (self, argv, trash_err=False):
340 # print 'running backquote',argv
342 result= subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
344 result= subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
347 # if you have any shell-expanded arguments like *
348 # and if there's any chance the command is adressed to the local host
349 def backquote_ssh (self, argv, trash_err=False):
350 if not self.probe(): return ''
351 return self.backquote( self.test_ssh().actual_argv(argv), trash_err)
353 ############################################################
355 def __init__ (self, buildname, pid, buildbox):
356 self.buildname=buildname
357 self.buildbox=buildbox
360 def add_pid(self,pid):
361 self.pids.append(pid)
364 return "== %s == (pids=%r)"%(self.buildname,self.pids)
366 class BuildBox (Box):
367 def __init__ (self,hostname):
368 Box.__init__(self,hostname)
369 self.build_instances=[]
371 def add_build (self,buildname,pid):
372 for build in self.build_instances:
373 if build.buildname==buildname:
376 self.build_instances.append(BuildInstance(buildname, pid, self))
378 def list(self, verbose=False):
379 if not self.build_instances:
380 header ('No build process on %s (%s)'%(self.hostname_fedora(),self.uptime()))
382 header ("Builds on %s (%s)"%(self.hostname_fedora(),self.uptime()))
383 for b in self.build_instances:
384 header (b.line(),banner=False)
386 def reboot (self, options):
388 Box.reboot(self,options)
390 command=['pkill','vbuild']
391 self.run_ssh(command,"Terminating vbuild processes",dry_run=options.dry_run)
393 # inspect box and find currently running builds
394 matcher=re.compile("\s*(?P<pid>[0-9]+).*-[bo]\s+(?P<buildname>[^\s]+)(\s|\Z)")
395 matcher_building_vm=re.compile("\s*(?P<pid>[0-9]+).*init-vserver.*\s+(?P<buildname>[^\s]+)\s*\Z")
396 def sense(self, options):
398 pids=self.backquote_ssh(['pgrep','vbuild'],trash_err=True)
400 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
401 ps_lines=self.backquote_ssh (command).split('\n')
402 for line in ps_lines:
403 if not line.strip() or line.find('PID')>=0: continue
404 m=BuildBox.matcher.match(line)
406 date=time.strftime('%Y-%m-%d',time.localtime(time.time()))
407 buildname=m.group('buildname').replace('@DATE@',date)
408 self.add_build (buildname,m.group('pid'))
410 m=BuildBox.matcher_building_vm.match(line)
412 # buildname is expansed here
413 self.add_build (buildname,m.group('pid'))
415 header('BuildBox.sense: command %r returned line that failed to match'%command)
416 header(">>%s<<"%line)
418 ############################################################
420 def __init__ (self, plcbox):
425 def set_timestamp (self,timestamp): self.timestamp=timestamp
426 def set_now (self): self.timestamp=int(time.time())
427 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
429 class PlcVsInstance (PlcInstance):
430 def __init__ (self, plcbox, vservername, ctxid):
431 PlcInstance.__init__(self,plcbox)
432 self.vservername=vservername
436 return self.vservername.split('-')[-1]
437 def buildname (self):
438 return self.vservername.rsplit('-',2)[0]
441 msg="== %s =="%(self.vplcname())
442 msg += " [=%s]"%self.vservername
443 if self.ctxid==0: msg+=" not (yet?) running"
444 else: msg+=" (ctx=%s)"%self.ctxid
445 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
446 else: msg += " *unknown timestamp*"
450 msg="vserver stopping %s on %s"%(self.vservername,self.plc_box.hostname)
451 self.plc_box.run_ssh(['vserver',self.vservername,'stop'],msg)
452 self.plc_box.forget(self)
454 class PlcLxcInstance (PlcInstance):
455 # does lxc have a context id of any kind ?
456 def __init__ (self, plcbox, lxcname, pid):
457 PlcInstance.__init__(self, plcbox)
458 self.lxcname = lxcname
462 return self.lxcname.split('-')[-1]
463 def buildname (self):
464 return self.lxcname.rsplit('-',2)[0]
467 msg="== %s =="%(self.vplcname())
468 msg += " [=%s]"%self.lxcname
469 if self.pid==-1: msg+=" not (yet?) running"
470 else: msg+=" (pid=%s)"%self.pid
471 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
472 else: msg += " *unknown timestamp*"
476 command="rsync lxc-driver.sh %s:/root"%self.plc_box.hostname
477 commands.getstatusoutput(command)
478 msg="lxc container stopping %s on %s"%(self.lxcname,self.plc_box.hostname)
479 self.plc_box.run_ssh(['/root/lxc-driver.sh','-c','stop_lxc','-n',self.lxcname],msg)
480 self.plc_box.forget(self)
484 def __init__ (self, hostname, max_plcs):
485 Box.__init__(self,hostname)
486 self.plc_instances=[]
487 self.max_plcs=max_plcs
489 def free_slots (self):
490 return self.max_plcs - len(self.plc_instances)
492 # fill one slot even though this one is not started yet
493 def add_dummy (self, plcname):
494 dummy=PlcVsInstance(self,'dummy_'+plcname,0)
496 self.plc_instances.append(dummy)
498 def forget (self, plc_instance):
499 self.plc_instances.remove(plc_instance)
501 def reboot (self, options):
503 Box.reboot(self,options)
505 self.soft_reboot (options)
507 def list(self, verbose=False):
508 if not self.plc_instances:
509 header ('No plc running on %s'%(self.line()))
511 header ("Active plc VMs on %s"%self.line())
512 self.plc_instances.sort(timestamp_sort)
513 for p in self.plc_instances:
514 header (p.line(),banner=False)
517 class PlcVsBox (PlcBox):
519 def add_vserver (self,vservername,ctxid):
520 for plc in self.plc_instances:
521 if plc.vservername==vservername:
522 header("WARNING, duplicate myplc %s running on %s"%\
523 (vservername,self.hostname),banner=False)
525 self.plc_instances.append(PlcVsInstance(self,vservername,ctxid))
528 msg="%s [max=%d,free=%d, VS-based] (%s)"%(self.hostname_fedora(), self.max_plcs,self.free_slots(),self.uname())
531 def plc_instance_by_vservername (self, vservername):
532 for p in self.plc_instances:
533 if p.vservername==vservername: return p
536 def soft_reboot (self, options):
537 self.run_ssh(['service','util-vserver','stop'],"Stopping all running vservers on %s"%(self.hostname,),
538 dry_run=options.dry_run)
540 def sense (self, options):
542 # try to find fullname (vserver_stat truncates to a ridiculously short name)
543 # fetch the contexts for all vservers on that box
544 map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
545 context_map=self.backquote_ssh (map_command)
546 # at this point we have a set of lines like
547 # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
549 for map_line in context_map.split("\n"):
550 if not map_line: continue
551 [path,xid] = map_line.split(':')
552 ctx_dict[xid]=os.path.basename(os.path.dirname(path))
553 # at this point ctx_id maps context id to vservername
555 command=['vserver-stat']
556 vserver_stat = self.backquote_ssh (command)
557 for vserver_line in vserver_stat.split("\n"):
558 if not vserver_line: continue
559 context=vserver_line.split()[0]
560 if context=="CTX": continue
562 longname=ctx_dict[context]
563 self.add_vserver(longname,context)
565 print 'WARNING: found ctx %s in vserver_stat but was unable to figure a corresp. vserver'%context
568 running_vsnames = [ i.vservername for i in self.plc_instances ]
569 command= ['grep','.']
570 command += ['/vservers/%s.timestamp'%vs for vs in running_vsnames]
571 command += ['/dev/null']
572 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
573 for ts_line in ts_lines:
574 if not ts_line.strip(): continue
575 # expect /vservers/<vservername>.timestamp:<timestamp>
577 (ts_file,timestamp)=ts_line.split(':')
578 ts_file=os.path.basename(ts_file)
579 (vservername,_)=os.path.splitext(ts_file)
580 timestamp=int(timestamp)
581 p=self.plc_instance_by_vservername(vservername)
583 print 'WARNING zombie plc',self.hostname,ts_line
584 print '... was expecting',vservername,'in',[i.vservername for i in self.plc_instances]
586 p.set_timestamp(timestamp)
587 except: print 'WARNING, could not parse ts line',ts_line
590 class PlcLxcBox (PlcBox):
592 def add_lxc (self,lxcname,pid):
593 for plc in self.plc_instances:
594 if plc.lxcname==lxcname:
595 header("WARNING, duplicate myplc %s running on %s"%\
596 (lxcname,self.hostname),banner=False)
598 self.plc_instances.append(PlcLxcInstance(self,lxcname,pid))
601 # a line describing the box
603 return "%s [max=%d,free=%d, LXC-based] (%s)"%(self.hostname_fedora(), self.max_plcs,self.free_slots(),
606 def plc_instance_by_lxcname (self, lxcname):
607 for p in self.plc_instances:
608 if p.lxcname==lxcname: return p
611 # essentially shutdown all running containers
612 def soft_reboot (self, options):
613 command="rsync lxc-driver.sh %s:/root"%self.hostname
614 commands.getstatusoutput(command)
615 self.run_ssh(['/root/lxc-driver.sh','-c','stop_all'],"Stopping all running lxc containers on %s"%(self.hostname,),
616 dry_run=options.dry_run)
619 # sense is expected to fill self.plc_instances with PlcLxcInstance's
620 # to describe the currently running VM's
621 # as well as to call self.get_uname() once
622 def sense (self, options):
624 command="rsync lxc-driver.sh %s:/root"%self.hostname
625 commands.getstatusoutput(command)
626 command=['/root/lxc-driver.sh','-c','sense_all']
627 lxc_stat = self.backquote_ssh (command)
628 for lxc_line in lxc_stat.split("\n"):
629 if not lxc_line: continue
630 lxcname=lxc_line.split(";")[0]
631 pid=lxc_line.split(";")[1]
632 timestamp=lxc_line.split(";")[2]
633 self.add_lxc(lxcname,pid)
634 timestamp=int(timestamp)
635 p=self.plc_instance_by_lxcname(lxcname)
637 print 'WARNING zombie plc',self.hostname,lxcname
638 print '... was expecting',lxcname,'in',[i.lxcname for i in self.plc_instances]
640 p.set_timestamp(timestamp)
642 ############################################################
644 def __init__ (self, nodename, pid, qemubox):
645 self.nodename=nodename
647 self.qemu_box=qemubox
652 def set_buildname (self,buildname): self.buildname=buildname
653 def set_timestamp (self,timestamp): self.timestamp=timestamp
654 def set_now (self): self.timestamp=int(time.time())
655 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
658 msg = "== %s =="%(short_hostname(self.nodename))
659 msg += " [=%s]"%self.buildname
660 if self.pid: msg += " (pid=%s)"%self.pid
661 else: msg += " not (yet?) running"
662 if self.timestamp: msg += " @ %s"%self.pretty_timestamp()
663 else: msg += " *unknown timestamp*"
668 print "cannot kill qemu %s with pid==0"%self.nodename
670 msg="Killing qemu %s with pid=%s on box %s"%(self.nodename,self.pid,self.qemu_box.hostname)
671 self.qemu_box.run_ssh(['kill',"%s"%self.pid],msg)
672 self.qemu_box.forget(self)
676 def __init__ (self, hostname, max_qemus):
677 Box.__init__(self,hostname)
678 self.qemu_instances=[]
679 self.max_qemus=max_qemus
681 def add_node (self,nodename,pid):
682 for qemu in self.qemu_instances:
683 if qemu.nodename==nodename:
684 header("WARNING, duplicate qemu %s running on %s"%\
685 (nodename,self.hostname), banner=False)
687 self.qemu_instances.append(QemuInstance(nodename,pid,self))
689 def forget (self, qemu_instance):
690 self.qemu_instances.remove(qemu_instance)
692 # fill one slot even though this one is not started yet
693 def add_dummy (self, nodename):
694 dummy=QemuInstance('dummy_'+nodename,0,self)
696 self.qemu_instances.append(dummy)
699 msg="%s [max=%d,free=%d] (%s)"%(self.hostname_fedora(), self.max_qemus,self.free_slots(),self.driver())
702 def list(self, verbose=False):
703 if not self.qemu_instances:
704 header ('No qemu on %s'%(self.line()))
706 header ("Active qemus on %s"%(self.line()))
707 self.qemu_instances.sort(timestamp_sort)
708 for q in self.qemu_instances:
709 header (q.line(),banner=False)
711 def free_slots (self):
712 return self.max_qemus - len(self.qemu_instances)
715 if hasattr(self,'_driver') and self._driver: return self._driver
716 return '*undef* driver'
718 def qemu_instance_by_pid (self,pid):
719 for q in self.qemu_instances:
720 if q.pid==pid: return q
723 def qemu_instance_by_nodename_buildname (self,nodename,buildname):
724 for q in self.qemu_instances:
725 if q.nodename==nodename and q.buildname==buildname:
729 def reboot (self, options):
731 Box.reboot(self,options)
733 self.run_ssh(['pkill','qemu'],"Killing qemu instances",
734 dry_run=options.dry_run)
736 matcher=re.compile("\s*(?P<pid>[0-9]+).*-cdrom\s+(?P<nodename>[^\s]+)\.iso")
737 def sense(self, options):
739 modules=self.backquote_ssh(['lsmod']).split('\n')
740 self._driver='*NO kqemu/kvm_intel MODULE LOADED*'
741 for module in modules:
742 if module.find('kqemu')==0:
743 self._driver='kqemu module loaded'
744 # kvm might be loaded without kvm_intel (we dont have AMD)
745 elif module.find('kvm_intel')==0:
746 self._driver='kvm_intel OK'
747 ########## find out running pids
748 pids=self.backquote_ssh(['pgrep','qemu'])
750 command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
751 ps_lines = self.backquote_ssh (command).split("\n")
752 for line in ps_lines:
753 if not line.strip() or line.find('PID') >=0 : continue
754 m=QemuBox.matcher.match(line)
756 self.add_node (m.group('nodename'),m.group('pid'))
758 header('QemuBox.sense: command %r returned line that failed to match'%command)
759 header(">>%s<<"%line)
760 ########## retrieve alive instances and map to build
762 command=['grep','.','/vservers/*/*/qemu.pid','/dev/null']
763 pid_lines=self.backquote_ssh(command,trash_err=True).split('\n')
764 for pid_line in pid_lines:
765 if not pid_line.strip(): continue
766 # expect <build>/<nodename>/qemu.pid:<pid>pid
768 (_,__,buildname,nodename,tail)=pid_line.split('/')
769 (_,pid)=tail.split(':')
770 q=self.qemu_instance_by_pid (pid)
772 q.set_buildname(buildname)
773 live_builds.append(buildname)
774 except: print 'WARNING, could not parse pid line',pid_line
775 # retrieve timestamps
776 if not live_builds: return
777 command= ['grep','.']
778 command += ['/vservers/%s/*/timestamp'%b for b in live_builds]
779 command += ['/dev/null']
780 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
781 for ts_line in ts_lines:
782 if not ts_line.strip(): continue
783 # expect <build>/<nodename>/timestamp:<timestamp>
785 (_,__,buildname,nodename,tail)=ts_line.split('/')
786 nodename=nodename.replace('qemu-','')
787 (_,timestamp)=tail.split(':')
788 timestamp=int(timestamp)
789 q=self.qemu_instance_by_nodename_buildname(nodename,buildname)
791 print 'WARNING zombie qemu',self.hostname,ts_line
792 print '... was expecting (',short_hostname(nodename),buildname,') in',\
793 [ (short_hostname(i.nodename),i.buildname) for i in self.qemu_instances ]
795 q.set_timestamp(timestamp)
796 except: print 'WARNING, could not parse ts line',ts_line
800 def __init__ (self, buildname, pid=0):
802 if pid!=0: self.pid.append(pid)
803 self.buildname=buildname
810 def set_timestamp (self,timestamp): self.timestamp=timestamp
811 def set_now (self): self.timestamp=int(time.time())
812 def pretty_timestamp (self): return time.strftime("%Y-%m-%d:%H-%M",time.localtime(self.timestamp))
814 def is_running (self): return len(self.pids) != 0
816 def add_pid (self,pid):
817 self.pids.append(pid)
818 def set_broken (self, plcindex, step):
819 self.broken_steps.append ( (plcindex, step,) )
823 if self.pids: double='*'+double[1]
824 if self.broken_steps: double=double[0]+'B'
825 msg = " %s %s =="%(double,self.buildname)
826 if not self.pids: pass
827 elif len(self.pids)==1: msg += " (pid=%s)"%self.pids[0]
828 else: msg += " !!!pids=%s!!!"%self.pids
829 msg += " @%s"%self.pretty_timestamp()
830 if self.broken_steps:
831 # sometimes we have an empty plcindex
832 msg += " [BROKEN=" + " ".join( [ "%s@%s"%(s,i) if i else s for (i,s) in self.broken_steps ] ) + "]"
836 def __init__ (self,hostname):
837 Box.__init__(self,hostname)
839 self.test_instances=[]
841 def reboot (self, options):
842 # can't reboot a vserver VM
843 self.run_ssh (['pkill','run_log'],"Terminating current runs",
844 dry_run=options.dry_run)
845 self.run_ssh (['rm','-f',Starting.location],"Cleaning %s"%Starting.location,
846 dry_run=options.dry_run)
848 def get_test (self, buildname):
849 for i in self.test_instances:
850 if i.buildname==buildname: return i
852 # we scan ALL remaining test results, even the ones not running
853 def add_timestamp (self, buildname, timestamp):
854 i=self.get_test(buildname)
856 i.set_timestamp(timestamp)
858 i=TestInstance(buildname,0)
859 i.set_timestamp(timestamp)
860 self.test_instances.append(i)
862 def add_running_test (self, pid, buildname):
863 i=self.get_test(buildname)
865 self.test_instances.append (TestInstance (buildname,pid))
868 print "WARNING: 2 concurrent tests run on same build %s"%buildname
871 def add_broken (self, buildname, plcindex, step):
872 i=self.get_test(buildname)
874 i=TestInstance(buildname)
875 self.test_instances.append(i)
876 i.set_broken(plcindex, step)
878 matcher_proc=re.compile (".*/proc/(?P<pid>[0-9]+)/cwd.*/root/(?P<buildname>[^/]+)$")
879 matcher_grep=re.compile ("/root/(?P<buildname>[^/]+)/logs/trace.*:TRACE:\s*(?P<plcindex>[0-9]+).*step=(?P<step>\S+).*")
880 matcher_grep_missing=re.compile ("grep: /root/(?P<buildname>[^/]+)/logs/trace: No such file or directory")
881 def sense (self, options):
883 self.starting_ips=[x for x in self.backquote_ssh(['cat',Starting.location], trash_err=True).strip().split('\n') if x]
885 # scan timestamps on all tests
886 # this is likely to not invoke ssh so we need to be a bit smarter to get * expanded
887 # xxx would make sense above too
888 command=['bash','-c',"grep . /root/*/timestamp /dev/null"]
889 ts_lines=self.backquote_ssh(command,trash_err=True).split('\n')
890 for ts_line in ts_lines:
891 if not ts_line.strip(): continue
892 # expect /root/<buildname>/timestamp:<timestamp>
894 (ts_file,timestamp)=ts_line.split(':')
895 ts_file=os.path.dirname(ts_file)
896 buildname=os.path.basename(ts_file)
897 timestamp=int(timestamp)
898 t=self.add_timestamp(buildname,timestamp)
899 except: print 'WARNING, could not parse ts line',ts_line
901 # let's try to be robust here -- tests that fail very early like e.g.
902 # "Cannot make space for a PLC instance: vplc IP pool exhausted", that occurs as part of provision
903 # will result in a 'trace' symlink to an inexisting 'trace-<>.txt' because no step has gone through
904 # simple 'trace' sohuld exist though as it is created by run_log
905 command=['bash','-c',"grep KO /root/*/logs/trace /dev/null 2>&1" ]
906 trace_lines=self.backquote_ssh (command).split('\n')
907 for line in trace_lines:
908 if not line.strip(): continue
909 m=TestBox.matcher_grep_missing.match(line)
911 buildname=m.group('buildname')
912 self.add_broken(buildname,'','NO STEP DONE')
914 m=TestBox.matcher_grep.match(line)
916 buildname=m.group('buildname')
917 plcindex=m.group('plcindex')
919 self.add_broken(buildname,plcindex, step)
921 header("TestBox.sense: command %r returned line that failed to match\n%s"%(command,line))
922 header(">>%s<<"%line)
924 pids = self.backquote_ssh (['pgrep','run_log'],trash_err=True)
926 command=['ls','-ld'] + ["/proc/%s/cwd"%pid for pid in pids.split("\n") if pid]
927 ps_lines=self.backquote_ssh (command).split('\n')
928 for line in ps_lines:
929 if not line.strip(): continue
930 m=TestBox.matcher_proc.match(line)
933 buildname=m.group('buildname')
934 self.add_running_test(pid, buildname)
936 header("TestBox.sense: command %r returned line that failed to match\n%s"%(command,line))
937 header(">>%s<<"%line)
941 return self.hostname_fedora()
943 def list (self, verbose=False):
944 # verbose shows all tests
946 instances = self.test_instances
949 instances = [ i for i in self.test_instances if i.is_running() ]
953 header ("No %s on %s"%(msg,self.line()))
955 header ("%s on %s"%(msg,self.line()))
956 instances.sort(timestamp_sort)
957 for i in instances: print i.line()
958 # show 'starting' regardless of verbose
959 if self.starting_ips:
960 header ("Starting IP addresses on %s"%self.line())
961 self.starting_ips.sort()
962 for starting in self.starting_ips: print starting
964 header ("Empty 'starting' on %s"%self.line())
966 ############################################################
971 def __init__ (self, plcs_on_vs=True, plcs_on_lxc=False):
972 self.options=Options()
973 self.options.dry_run=False
974 self.options.verbose=False
975 self.options.reboot=False
976 self.options.soft=False
977 self.test_box = TestBox (self.test_box_spec())
978 self.build_boxes = [ BuildBox(h) for h in self.build_boxes_spec() ]
979 # for compat with older LocalSubstrate
981 self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_vs_boxes_spec ()]
982 self.plc_lxc_boxes = [ PlcLxcBox (h,m) for (h,m) in self.plc_lxc_boxes_spec ()]
984 self.plc_vs_boxes = [ PlcVsBox (h,m) for (h,m) in self.plc_boxes_spec ()]
985 self.plc_lxc_boxes = [ ]
986 self.qemu_boxes = [ QemuBox (h,m) for (h,m) in self.qemu_boxes_spec ()]
989 self.vplc_pool = Pool (self.vplc_ips(),"for vplcs",self)
990 self.vnode_pool = Pool (self.vnode_ips(),"for vnodes",self)
992 self.rescope (plcs_on_vs=plcs_on_vs, plcs_on_lxc=plcs_on_lxc)
994 # which plc boxes are we interested in ?
995 def rescope (self, plcs_on_vs, plcs_on_lxc):
997 if plcs_on_vs: self.plc_boxes += self.plc_vs_boxes
998 if plcs_on_lxc: self.plc_boxes += self.plc_lxc_boxes
999 self.default_boxes = self.plc_boxes + self.qemu_boxes
1000 self.all_boxes = self.build_boxes + [ self.test_box ] + self.plc_boxes + self.qemu_boxes
1002 def summary_line (self):
1004 msg += " %d vp"%len(self.plc_vs_boxes)
1005 msg += " %d xp"%len(self.plc_lxc_boxes)
1006 msg += " %d tried plc boxes"%len(self.plc_boxes)
1010 def fqdn (self, hostname):
1011 if hostname.find('.')<0: return "%s.%s"%(hostname,self.domain())
1014 # return True if actual sensing takes place
1015 def sense (self,force=False):
1016 if self._sensed and not force: return False
1017 print 'Sensing local substrate...',
1018 for b in self.default_boxes: b.sense(self.options)
1023 def list (self, verbose=False):
1024 for b in self.default_boxes:
1027 def add_dummy_plc (self, plc_boxname, plcname):
1028 for pb in self.plc_boxes:
1029 if pb.hostname==plc_boxname:
1030 pb.add_dummy(plcname)
1032 def add_dummy_qemu (self, qemu_boxname, qemuname):
1033 for qb in self.qemu_boxes:
1034 if qb.hostname==qemu_boxname:
1035 qb.add_dummy(qemuname)
1038 def add_starting_dummy (self, bname, vname):
1039 return self.add_dummy_plc (bname, vname) or self.add_dummy_qemu (bname, vname)
1042 def provision (self,plcs,options):
1044 # attach each plc to a plc box and an IP address
1045 plcs = [ self.provision_plc (plc,options) for plc in plcs ]
1046 # attach each node/qemu to a qemu box with an IP address
1047 plcs = [ self.provision_qemus (plc,options) for plc in plcs ]
1048 # update the SFA spec accordingly
1049 plcs = [ self.localize_sfa_rspec(plc,options) for plc in plcs ]
1052 except Exception, e:
1053 print '* Could not provision this test on current substrate','--',e,'--','exiting'
1054 traceback.print_exc()
1057 # it is expected that a couple of options like ips_bplc and ips_vplc
1058 # are set or unset together
1060 def check_options (x,y):
1061 if not x and not y: return True
1062 return len(x)==len(y)
1064 # find an available plc box (or make space)
1065 # and a free IP address (using options if present)
1066 def provision_plc (self, plc, options):
1068 assert Substrate.check_options (options.ips_bplc, options.ips_vplc)
1070 #### let's find an IP address for that plc
1072 if options.ips_vplc:
1074 # we don't check anything here,
1075 # it is the caller's responsability to cleanup and make sure this makes sense
1076 plc_boxname = options.ips_bplc.pop()
1077 vplc_hostname=options.ips_vplc.pop()
1079 if self.sense(): self.list()
1082 # try to find an available IP
1083 self.vplc_pool.sense()
1084 couple=self.vplc_pool.next_free()
1086 (vplc_hostname,unused)=couple
1087 #### we need to find one plc box that still has a slot
1089 # use the box that has max free spots for load balancing
1090 for pb in self.plc_boxes:
1091 free=pb.free_slots()
1093 plc_boxname=pb.hostname
1095 # if there's no available slot in the plc_boxes, or we need a free IP address
1096 # make space by killing the oldest running instance
1097 if not plc_boxname or not vplc_hostname:
1098 # find the oldest of all our instances
1099 all_plc_instances=reduce(lambda x, y: x+y,
1100 [ pb.plc_instances for pb in self.plc_boxes ],
1102 all_plc_instances.sort(timestamp_sort)
1104 plc_instance_to_kill=all_plc_instances[0]
1107 if not plc_boxname: msg += " PLC boxes are full"
1108 if not vplc_hostname: msg += " vplc IP pool exhausted"
1109 msg += " %s"%self.summary_line()
1110 raise Exception,"Cannot make space for a PLC instance:"+msg
1111 freed_plc_boxname=plc_instance_to_kill.plc_box.hostname
1112 freed_vplc_hostname=plc_instance_to_kill.vplcname()
1113 message='killing oldest plc instance = %s on %s'%(plc_instance_to_kill.line(),
1115 plc_instance_to_kill.kill()
1116 # use this new plcbox if that was the problem
1118 plc_boxname=freed_plc_boxname
1119 # ditto for the IP address
1120 if not vplc_hostname:
1121 vplc_hostname=freed_vplc_hostname
1122 # record in pool as mine
1123 self.vplc_pool.set_mine(vplc_hostname)
1126 self.add_dummy_plc(plc_boxname,plc['name'])
1127 vplc_ip = self.vplc_pool.get_ip(vplc_hostname)
1128 self.vplc_pool.add_starting(vplc_hostname, plc_boxname)
1130 #### compute a helpful vserver name
1131 # remove domain in hostname
1132 vplc_short = short_hostname(vplc_hostname)
1133 vservername = "%s-%d-%s" % (options.buildname,plc['index'],vplc_short)
1134 plc_name = "%s_%s"%(plc['name'],vplc_short)
1136 utils.header( 'PROVISION plc %s in box %s at IP %s as %s'%\
1137 (plc['name'],plc_boxname,vplc_hostname,vservername))
1139 #### apply in the plc_spec
1141 # label=options.personality.replace("linux","")
1142 mapper = {'plc': [ ('*' , {'host_box':plc_boxname,
1143 # 'name':'%s-'+label,
1145 'vservername':vservername,
1146 'vserverip':vplc_ip,
1147 'PLC_DB_HOST':vplc_hostname,
1148 'PLC_API_HOST':vplc_hostname,
1149 'PLC_BOOT_HOST':vplc_hostname,
1150 'PLC_WWW_HOST':vplc_hostname,
1151 'PLC_NET_DNS1' : self.network_settings() [ 'interface_fields:dns1' ],
1152 'PLC_NET_DNS2' : self.network_settings() [ 'interface_fields:dns2' ],
1157 # mappers only work on a list of plcs
1158 return TestMapper([plc],options).map(mapper)[0]
1161 def provision_qemus (self, plc, options):
1163 assert Substrate.check_options (options.ips_bnode, options.ips_vnode)
1165 test_mapper = TestMapper ([plc], options)
1166 nodenames = test_mapper.node_names()
1168 for nodename in nodenames:
1170 if options.ips_vnode:
1171 # as above, it's a rerun, take it for granted
1172 qemu_boxname=options.ips_bnode.pop()
1173 vnode_hostname=options.ips_vnode.pop()
1175 if self.sense(): self.list()
1178 # try to find an available IP
1179 self.vnode_pool.sense()
1180 couple=self.vnode_pool.next_free()
1182 (vnode_hostname,unused)=couple
1183 # find a physical box
1185 # use the box that has max free spots for load balancing
1186 for qb in self.qemu_boxes:
1187 free=qb.free_slots()
1189 qemu_boxname=qb.hostname
1191 # if we miss the box or the IP, kill the oldest instance
1192 if not qemu_boxname or not vnode_hostname:
1193 # find the oldest of all our instances
1194 all_qemu_instances=reduce(lambda x, y: x+y,
1195 [ qb.qemu_instances for qb in self.qemu_boxes ],
1197 all_qemu_instances.sort(timestamp_sort)
1199 qemu_instance_to_kill=all_qemu_instances[0]
1202 if not qemu_boxname: msg += " QEMU boxes are full"
1203 if not vnode_hostname: msg += " vnode IP pool exhausted"
1204 msg += " %s"%self.summary_line()
1205 raise Exception,"Cannot make space for a QEMU instance:"+msg
1206 freed_qemu_boxname=qemu_instance_to_kill.qemu_box.hostname
1207 freed_vnode_hostname=short_hostname(qemu_instance_to_kill.nodename)
1209 message='killing oldest qemu node = %s on %s'%(qemu_instance_to_kill.line(),
1211 qemu_instance_to_kill.kill()
1212 # use these freed resources where needed
1213 if not qemu_boxname:
1214 qemu_boxname=freed_qemu_boxname
1215 if not vnode_hostname:
1216 vnode_hostname=freed_vnode_hostname
1217 self.vnode_pool.set_mine(vnode_hostname)
1219 self.add_dummy_qemu (qemu_boxname,vnode_hostname)
1220 mac=self.vnode_pool.retrieve_userdata(vnode_hostname)
1221 ip=self.vnode_pool.get_ip (vnode_hostname)
1222 self.vnode_pool.add_starting(vnode_hostname,qemu_boxname)
1224 vnode_fqdn = self.fqdn(vnode_hostname)
1225 nodemap={'host_box':qemu_boxname,
1226 'node_fields:hostname':vnode_fqdn,
1227 'interface_fields:ip':ip,
1228 'ipaddress_fields:ip_addr':ip,
1229 'interface_fields:mac':mac,
1231 nodemap.update(self.network_settings())
1232 maps.append ( (nodename, nodemap) )
1234 utils.header("PROVISION node %s in box %s at IP %s with MAC %s"%\
1235 (nodename,qemu_boxname,vnode_hostname,mac))
1237 return test_mapper.map({'node':maps})[0]
1239 def localize_sfa_rspec (self,plc,options):
1241 plc['sfa']['SFA_REGISTRY_HOST'] = plc['PLC_DB_HOST']
1242 plc['sfa']['SFA_AGGREGATE_HOST'] = plc['PLC_DB_HOST']
1243 plc['sfa']['SFA_SM_HOST'] = plc['PLC_DB_HOST']
1244 plc['sfa']['SFA_DB_HOST'] = plc['PLC_DB_HOST']
1245 plc['sfa']['SFA_PLC_URL'] = 'https://' + plc['PLC_API_HOST'] + ':443/PLCAPI/'
1248 #################### release:
1249 def release (self,options):
1250 self.vplc_pool.release_my_starting()
1251 self.vnode_pool.release_my_starting()
1254 #################### show results for interactive mode
1255 def get_box (self,boxname):
1256 for b in self.build_boxes + self.plc_boxes + self.qemu_boxes + [self.test_box] :
1257 if b.shortname()==boxname: return b
1259 if b.shortname()==boxname.split('.')[0]: return b
1261 print "Could not find box %s"%boxname
1264 def list_boxes(self,box_or_names):
1266 for box in box_or_names:
1267 if not isinstance(box,Box): box=self.get_box(box)
1268 if not box: continue
1269 box.sense(self.options)
1271 for box in box_or_names:
1272 if not isinstance(box,Box): box=self.get_box(box)
1273 if not box: continue
1274 box.list(self.options.verbose)
1276 def reboot_boxes(self,box_or_names):
1277 for box in box_or_names:
1278 if not isinstance(box,Box): box=self.get_box(box)
1279 if not box: continue
1280 box.reboot(self.options)
1282 ####################
1283 # can be run as a utility to probe/display/manage the local infrastructure
1285 parser=OptionParser()
1286 parser.add_option ('-r',"--reboot",action='store_true',dest='reboot',default=False,
1287 help='reboot mode (use shutdown -r)')
1288 parser.add_option ('-s',"--soft",action='store_true',dest='soft',default=False,
1289 help='soft mode for reboot (vserver stop or kill qemus)')
1290 parser.add_option ('-t',"--testbox",action='store_true',dest='testbox',default=False,
1291 help='add test box')
1292 parser.add_option ('-b',"--build",action='store_true',dest='builds',default=False,
1293 help='add build boxes')
1294 parser.add_option ('-p',"--plc",action='store_true',dest='plcs',default=False,
1295 help='add plc boxes')
1296 parser.add_option ('-q',"--qemu",action='store_true',dest='qemus',default=False,
1297 help='add qemu boxes')
1298 parser.add_option ('-a',"--all",action='store_true',dest='all',default=False,
1299 help='address all known boxes, like -b -t -p -q')
1300 parser.add_option ('-v',"--verbose",action='store_true',dest='verbose',default=False,
1301 help='verbose mode')
1302 parser.add_option ('-n',"--dry_run",action='store_true',dest='dry_run',default=False,
1303 help='dry run mode')
1304 (self.options,args)=parser.parse_args()
1306 self.rescope (plcs_on_vs=True, plcs_on_lxc=True)
1309 if self.options.testbox: boxes += [self.test_box]
1310 if self.options.builds: boxes += self.build_boxes
1311 if self.options.plcs: boxes += self.plc_boxes
1312 if self.options.qemus: boxes += self.qemu_boxes
1313 if self.options.all: boxes += self.all_boxes
1315 # default scope is -b -p -q -t
1317 boxes = self.build_boxes + self.plc_boxes + self.qemu_boxes + [self.test_box]
1319 if self.options.reboot: self.reboot_boxes (boxes)
1320 else: self.list_boxes (boxes)