scripts/manage-infrastructure.py

   1 #!/usr/bin/python
   2
   3 import os.path, sys
   4 import re
   5 import subprocess
   6 from optparse import OptionParser
   7
   8 class BuildBoxes:
   9
  10     # everything in the onelab.eu domain
  11     domain = 'pl.sophia.inria.fr'
  12     build_boxes = [ "mirror", "liquid", "reed", "velvet", ]
  13     plc_boxes = [ "testplc" ]
  14     testmaster = 'testmaster'
  15     testmaster_boxes = [ testmaster ]
  16     # cache the list of qemu boxes in ~/.qemu-boxes
  17     # this can be refreshed by running -c
  18     qemu_boxes=[]
  19
  20     def cache_file (self): return os.path.expanduser("~/.qemu-boxes")
  21
  22     def load_cache (self):
  23         cache=self.cache_file()
  24         if os.path.isfile(cache):
  25             self.qemu_boxes=file(cache).read().split()
  26         self.test_boxes = self.plc_boxes + self.qemu_boxes
  27
  28     # run LocalTestResources on testmaster
  29     def refresh_cache (self):
  30         retrieved= \
  31             self.backquote_ssh(self.fqdn(self.testmaster),['LocalTestResources.py'],trash_err=True)
  32         remove="."+BuildBoxes.domain
  33         retrieved = [ x.replace(remove,"").strip() for x in retrieved.split()]
  34         self.qemu_boxes = retrieved
  35         cache=self.cache_file()
  36         file(cache,'w').write(' '.join(self.qemu_boxes)+'\n')
  37         print "New contents of %s:"%cache
  38         print file(cache).read(),
  39
  40     def __init__ (self):
  41         # dummy defaults
  42         self.boxes = []
  43         self.do_tracker_qemus = False
  44         self.do_tracker_plcs = False
  45         self.load_cache()
  46
  47     def fqdn (self, box):
  48         return "%s.%s"%(box,self.domain)
  49
  50     ssh_command=['ssh','-o','ConnectTimeout=3']
  51     @staticmethod
  52     def root (box): return "root@%s"%box
  53
  54     @staticmethod
  55     def ssh(box):
  56         return BuildBoxes.ssh_command + [ BuildBoxes.root(box) ]
  57
  58     def header (self,message):
  59         print "===============",message
  60         sys.stdout.flush()
  61
  62     def run (self,argv,message, trash_err=False):
  63         if self.options.dry_run:
  64             print 'DRY_RUN:',
  65             print " ".join(argv)
  66             return 0
  67         else:
  68             if message: self.header(message)
  69             if not trash_err:
  70                 return subprocess.call(argv)
  71             else:
  72                 return subprocess.call(argv,stderr=file('/dev/null','w'))
  73
  74     def run_ssh (self, box, argv, message, trash_err=False):
  75         result=self.run (self.ssh(box) + argv, message, trash_err)
  76         if result!=0:
  77             print "WARNING: failed to run %s on %s"%(" ".join(argv),box)
  78         return result
  79
  80     def backquote (self, argv, trash_err=False):
  81         if not trash_err:
  82             return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
  83         else:
  84             return subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
  85
  86     def backquote_ssh (self, box, argv, trash_err=False):
  87         # first probe the ssh link
  88         hostname=self.backquote ( self.ssh(box) + [ "hostname"], trash_err=True )
  89         if not hostname:
  90             print "%s unreachable"%self.root(box)
  91             return ''
  92         else:
  93             return self.backquote( ['ssh',self.root(box)] + argv, trash_err)
  94
  95     def reboot (self,box):
  96         command=['ssh',self.root(box),'shutdown','-r','now']
  97         self.run (command,"Rebooting %s"%box)
  98
  99     def handle_tracker_plcs (self):
 100         box = self.fqdn (self.testmaster)
 101         filename="tracker-plcs"
 102         if not self.options.probe:
 103             command=["rm","-rf",filename]
 104             self.run_ssh(box,command,"Cleaning up %s on %s"%(filename,box))
 105         else:
 106             self.header ("++++++++++ Inspecting %s on %s"%(filename,box))
 107             read_command = ["cat",filename]
 108             trackers=self.backquote_ssh(box,read_command)
 109             for tracker in trackers.split('\n'):
 110                 if not tracker: continue
 111                 try:
 112                     tracker=tracker.strip()
 113                     [hostname,buildname]=tracker.split('@')
 114                     [left,plcname]=buildname.rsplit('-',1)
 115                     print self.margin_outline(plcname),tracker
 116                 except:
 117                     print self.margin(""),tracker
 118
 119     def handle_tracker_qemus (self):
 120         box = self.fqdn (self.testmaster)
 121         filename="tracker-qemus"
 122         if not self.options.probe:
 123             command=["rm","-rf",filename]
 124             self.run_ssh(box,command,"Cleaning up %s on %s"%(filename,box))
 125         else:
 126             self.header ("++++++++++ Inspecting %s on %s"%(filename,box))
 127             read_command = ["cat",filename]
 128             trackers=self.backquote_ssh(box,read_command)
 129             for tracker in trackers.split('\n'):
 130                 if not tracker: continue
 131                 try:
 132                     tracker=tracker.strip()
 133                     [hostname,buildname,nodename]=tracker.split('@')
 134                     nodename=nodename.split('.')[0]
 135                     print self.margin_outline(nodename),tracker
 136                 except:
 137                     print self.margin(""),tracker
 138
 139     def handle_build_box (self,box):
 140         if not self.options.probe:
 141             self.reboot(box)
 142         else:
 143             command=['uptime']
 144             uptime=self.backquote_ssh(box,command,True).strip()
 145
 146             command=['pgrep','build']
 147             if self.options.dry_run:
 148                 self.run_ssh(box,command,None)
 149             else:
 150                 pids=self.backquote_ssh(box,command,True)
 151                 if not pids:
 152                     self.header ('No build process on %s (%s)'%(box,uptime))
 153                 else:
 154                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 155                     self.run_ssh(box,command,"Active build processes on %s (%s)"%(box,uptime),True)
 156
 157     # this one is more accurate as it locates processes in the vservers as well
 158     # but it's so sloooowww
 159     def handle_build_box_deep (self,box):
 160         if not self.options.probe:
 161             self.reboot(box)
 162         else:
 163             command=['uptime']
 164             uptime=self.backquote_ssh(box,command,True).strip()
 165
 166             command=['vps','-e']
 167             if self.options.dry_run:
 168                 self.run_ssh(box,command,None)
 169             else:
 170                 # simulate grep vbuild
 171                 vps_lines=[ line for line in self.backquote_ssh(box,command,True).split("\n")
 172                             if line.find('vbuild') >= 0]
 173                 pids=[ line.split()[0] for line in vps_lines ]
 174                 if not pids:
 175                     self.header ('No build process on %s (%s)'%(box,uptime))
 176                 else:
 177                     command=['vps','-o','pid,command'] + pids
 178                     self.run_ssh(box,command,"Active build processes on %s (%s)"%(box,uptime),True)
 179
 180
 181     vplc_matcher = re.compile(".*(vplc[0-9]+$)")
 182     def vplcname (self, vservername):
 183         match = self.vplc_matcher.match(vservername)
 184         if match: return match.groups(0)
 185         else: return ""
 186
 187     margin_format="%-14s"
 188     def margin(self,string): return self.margin_format%string
 189     def outline (self, string): return '== %s =='%string
 190     def margin_outline (self, string): return self.margin(self.outline(string))
 191
 192     def handle_plc_box (self,box):
 193 # initial approach was to first scan vserver-stat, but it's not needed
 194         if not self.options.probe:
 195 #            # remove mark for all running servers to avoid resurrection
 196 #            if vserver_names:
 197 #                bash="; ".join( [ "rm -f /etc/vservers/%s/apps/init/mark"%vs for vs in vserver_names ] )
 198 #                stop_command=['bash','-c',"'" + bash + "'"]
 199 #                self.run_ssh(box,stop_command,"Removing mark on running vservers on %s"%box)
 200             # just trash all marks
 201             stop_command=['rm','-rf','/etc/vservers/*/apps/init/mark']
 202             self.run_ssh(box,stop_command,"Removing all vserver marks on %s"%box)
 203             if not self.options.soft:
 204                 self.reboot(box)
 205             else:
 206                 self.run_ssh(box,['service','util-vserver','stop'],"Stopping all running vservers")
 207             return
 208         # even for rebooting we need to scan vserver-stat to stop the vservers properly
 209         vserver_names=[]
 210         command=['vserver-stat']
 211         if self.options.dry_run:
 212             self.run_ssh(box,command,"Active vservers on %s"%box)
 213         # try to find fullname (vserver_stat truncates to a ridiculously short name)
 214         self.header ("vserver map on %s"%box)
 215         # fetch the contexts for all vservers on that box
 216         map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
 217         context_map=self.backquote_ssh (box,map_command)
 218         # at this point we have a set of lines like
 219         # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
 220         ctx_dict={}
 221         for map_line in context_map.split("\n"):
 222             if not map_line: continue
 223             [path,xid] = map_line.split(':')
 224             ctx_dict[xid]=os.path.basename(os.path.dirname(path))
 225         # at this point ctx_id maps context id to vservername
 226
 227         vserver_stat = self.backquote_ssh (box,command)
 228         for vserver_line in vserver_stat.split("\n"):
 229             if not vserver_line: continue
 230             context=vserver_line.split()[0]
 231             if context=="CTX":
 232                 print self.margin(""),vserver_line
 233                 continue
 234             longname=ctx_dict[context]
 235             vserver_names.append(longname)
 236             print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
 237
 238     vnode_matcher = re.compile(".*(vnode[0-9]+)")
 239     def vnodename (self, ps_line):
 240         match = self.vnode_matcher.match(ps_line)
 241         if match: return match.groups(0)
 242         else: return ""
 243
 244     def handle_qemu_box (self,box):
 245         if not self.options.probe:
 246             if not self.options.soft:
 247                 self.reboot(box)
 248             else:
 249                 self.run_ssh(box,['pkill','qemu'],"Killing qemu instances")
 250         else:
 251             command=['lsmod']
 252             modules=self.backquote_ssh(box,command).split('\n')
 253             kqemu_msg='*NO kqemu/kmv_intel MODULE LOADED*'
 254             for module in modules:
 255                 if module.find('kqemu')==0:
 256                     kqemu_msg='kqemu module loaded'
 257                 # kvm might be loaded without vkm_intel (we dont have AMD)
 258                 elif module.find('kvm_intel')==0:
 259                     kqemu_msg='kvm_intel module loaded'
 260
 261             command=['pgrep','qemu']
 262             if self.options.dry_run:
 263                 self.run_ssh(box,command,None)
 264             else:
 265                 pids=self.backquote_ssh(box,command)
 266                 if not pids:
 267                     self.header ('No qemu process on %s (%s)'%(box,kqemu_msg))
 268                 else:
 269                     self.header ("Active qemu processes on %s (%s)"%(box,kqemu_msg))
 270                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 271                     ps_lines = self.backquote_ssh (box,command).split("\n")
 272                     for ps_line in ps_lines:
 273                         if not ps_line or ps_line.find('PID') >=0 : continue
 274                         print self.margin_outline(self.vnodename(ps_line)), ps_line
 275
 276     # the ouput of ps -o pid,command gives us <pid> bash <buildname>/run_log
 277     def testmaster_buildname (self, ps_line):
 278         chunks=ps_line.split()
 279         path=chunks[2]
 280         [buildname,command]=path.split('/')
 281         return buildname
 282
 283     def handle_testmaster_box (self, box):
 284         if not self.options.probe:
 285             pass
 286         else:
 287             command=['pgrep','run_log']
 288             if self.options.dry_run:
 289                 self.run_ssh(box,command,None)
 290             else:
 291                 pids=self.backquote_ssh(box,command)
 292                 if not pids:
 293                     self.header ('No run_log process on %s'%box)
 294                 else:
 295                     self.header ("Active run_log processes on %s"%(box))
 296                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 297                     ps_lines = self.backquote_ssh (box,command).split("\n")
 298                     for ps_line in ps_lines:
 299                         if not ps_line or ps_line.find('PID') >=0 : continue
 300                         print self.margin_outline(self.testmaster_buildname(ps_line)), ps_line
 301
 302
 303     def handle_box(self,box,type):
 304         if box in self.qemu_boxes:
 305             if type=="qemu": self.handle_qemu_box(self.fqdn(box))
 306         elif box in self.plc_boxes:
 307             if type=="plc":  self.handle_plc_box(self.fqdn(box))
 308         elif box in self.testmaster_boxes:
 309             if type=='testmaster': self.handle_testmaster_box(self.fqdn(box))
 310         elif type=="build":
 311             if self.options.deep:
 312                 self.handle_build_box_deep(self.fqdn(box))
 313             else:
 314                 self.handle_build_box(self.fqdn(box))
 315
 316     def handle_disk (self,box):
 317         box=self.fqdn(box)
 318         return self.run_ssh(box,["df","-h",],"Disk space on %s"%box)
 319
 320     def main (self):
 321         usage="""%prog [options] [hostname..(s)]
 322 Default is to act on test boxes only"""
 323         parser = OptionParser (usage=usage)
 324         parser.add_option ("-n","--dry-run",action="store_true",dest="dry_run",default=False,
 325                            help="Dry run")
 326         parser.add_option ("-r","--reboot", action="store_false",dest="probe",default=True,
 327                            help="Actually reset/reboot stuff instead of just probing it")
 328         parser.add_option ("-s","--soft",action="store_true",dest="soft",default=False,
 329                            help="Soft reset instead of hard reboot of the boxes")
 330         # no need for -p = probe, as this is the default
 331         parser.add_option ("-p","--plc", action="store_true",dest="plc_only",default=False,
 332                            help="Acts on the plc box only")
 333
 334         parser.add_option ("-e","--deep",action="store_true", dest="deep", default=False,
 335                            help="on build boxes, shows vbuild processes in vservers as well; signif. slower")
 336
 337         parser.add_option ("-a","--all",action="store_true",dest="all_boxes",default=False,
 338                            help="Acts on build and test boxes")
 339         parser.add_option ("-b","--build",action="store_true",dest="build_only",default=False,
 340                            help="Acts on build boxes only")
 341         parser.add_option ("-q","--qemu",action="store_true",dest="qemu_only",default=False,
 342                            help="Only acts on the qemu boxes")
 343         parser.add_option ("-t","--trackers",action="store_true",dest="trackers_only",default=False,
 344                            help="Only wipes trackers")
 345         parser.add_option ("-m","--master",action="store_true",dest="testmaster_only",default=False,
 346                            help="Display the testmaster status")
 347         parser.add_option ("-d","--disk",action="store_true",dest="show_disk",default=False,
 348                            help="Only inspects disk status")
 349         parser.add_option ("-c","--refresh-cache",action="store_true",dest="refresh_cache", default=False,
 350                            help="Refresh cached list of qemu boxes at testmaster - implies -q")
 351
 352         (self.options,args) = parser.parse_args()
 353
 354         # -c implies -q
 355         if self.options.refresh_cache:
 356             self.options.qemu_only=True
 357             self.refresh_cache()
 358
 359         # use given hostnames if provided
 360         if args:
 361             self.boxes=args
 362             # if hostnames are specified, let's stay on the safe side and don't reset trackers
 363             self.do_tracker_plcs = False
 364             self.do_tracker_qemus = False
 365         elif self.options.all_boxes:
 366             self.boxes=self.test_boxes + self.build_boxes + self.testmaster_boxes
 367             self.do_tracker_plcs = True
 368             self.do_tracker_qemus = True
 369         elif self.options.build_only:
 370             self.boxes=self.build_boxes
 371             self.do_tracker_plcs = False
 372             self.do_tracker_qemus = False
 373         elif self.options.qemu_only:
 374             self.boxes=self.qemu_boxes
 375             self.do_tracker_plcs = False
 376             self.do_tracker_qemus = True
 377         elif self.options.plc_only:
 378             self.boxes=self.plc_boxes
 379             self.do_tracker_plcs = True
 380             self.do_tracker_qemus = False
 381         elif self.options.testmaster_only:
 382             self.boxes=self.testmaster_boxes
 383             self.do_tracker_plcs = False
 384             self.do_tracker_qemus = False
 385         elif self.options.trackers_only:
 386             self.boxes = []
 387             self.do_tracker_plcs = True
 388             self.do_tracker_qemus = True
 389         # default
 390         else:
 391             self.boxes = self.test_boxes
 392             self.do_tracker_plcs = True
 393             self.do_tracker_qemus = True
 394
 395         if self.options.show_disk:
 396             for box in self.boxes: self.handle_disk(box)
 397             return
 398
 399         # PLCS
 400         if self.do_tracker_plcs:self.handle_tracker_plcs ()
 401         for box in self.boxes:  self.handle_box (box,"plc")
 402         # QEMU
 403         if self.do_tracker_qemus:self.handle_tracker_qemus ()
 404         for box in self.boxes:  self.handle_box (box,"qemu")
 405         # ALL OTHERS
 406         for box in self.boxes:  self.handle_box (box,"build")
 407         # TESTMASTER
 408         for box in self.boxes:  self.handle_box (box,"testmaster")
 409
 410 if __name__ == "__main__":
 411     BuildBoxes().main()