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     testmaster = 'testmaster'
  13     build_boxes = [ "mirror", "liquid", "reed", "velvet", ]
  14     plc_boxes = [ "testplc" ]
  15     qemu_boxes = \
  16         [ "qemu64-%d"%i for i in range (1,4) ] + \
  17         [ "qemu32-%d"%i for i in range (1,6) ]
  18     test_boxes = plc_boxes + qemu_boxes
  19     testmaster_boxes = [ testmaster ]
  20
  21     def __init__ (self):
  22         # dummy defaults
  23         self.boxes = []
  24         self.do_tracker_qemus = False
  25         self.do_tracker_plcs = False
  26
  27     def fqdn (self, box):
  28         return "%s.%s"%(box,self.domain)
  29
  30     ssh_command=['ssh','-o','ConnectTimeout=3']
  31     @staticmethod
  32     def root (box): return "root@%s"%box
  33
  34     @staticmethod
  35     def ssh(box):
  36         return BuildBoxes.ssh_command + [ BuildBoxes.root(box) ]
  37
  38     def header (self,message):
  39         print "===============",message
  40         sys.stdout.flush()
  41
  42     def run (self,argv,message, trash_err=False):
  43         if self.options.dry_run:
  44             print 'DRY_RUN:',
  45             print " ".join(argv)
  46             return 0
  47         else:
  48             if message: self.header(message)
  49             if not trash_err:
  50                 return subprocess.call(argv)
  51             else:
  52                 return subprocess.call(argv,stderr=file('/dev/null','w'))
  53
  54     def run_ssh (self, box, argv, message, trash_err=False):
  55         result=self.run (self.ssh(box) + argv, message, trash_err)
  56         if result!=0:
  57             print "WARNING: failed to run %s on %s"%(" ".join(argv),box)
  58         return result
  59
  60     def backquote (self, argv, trash_err=False):
  61         if not trash_err:
  62             return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
  63         else:
  64             return subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=file('/dev/null','w')).communicate()[0]
  65
  66     def backquote_ssh (self, box, argv, trash_err=False):
  67         # first probe the ssh link
  68         hostname=self.backquote ( self.ssh(box) + [ "hostname"], trash_err=True )
  69         if not hostname:
  70             print "%s unreachable"%self.root(box)
  71             return ''
  72         else:
  73             return self.backquote( ['ssh',self.root(box)] + argv, trash_err)
  74
  75     def reboot (self,box):
  76         command=['ssh',self.root(box),'shutdown','-r','now']
  77         self.run (command,"Rebooting %s"%box)
  78
  79     def handle_tracker_plcs (self):
  80         box = self.fqdn (self.testmaster)
  81         filename="tracker-plcs"
  82         if not self.options.probe:
  83             command=["rm","-rf",filename]
  84             self.run_ssh(box,command,"Cleaning up %s on %s"%(filename,box))
  85         else:
  86             self.header ("++++++++++ Inspecting %s on %s"%(filename,box))
  87             read_command = ["cat",filename]
  88             trackers=self.backquote_ssh(box,read_command)
  89             for tracker in trackers.split('\n'):
  90                 if not tracker: continue
  91                 try:
  92                     tracker=tracker.strip()
  93                     [hostname,buildname]=tracker.split('@')
  94                     [left,plcname]=buildname.rsplit('-',1)
  95                     print self.margin_outline(plcname),tracker
  96                 except:
  97                     print self.margin(""),tracker
  98
  99     def handle_tracker_qemus (self):
 100         box = self.fqdn (self.testmaster)
 101         filename="tracker-qemus"
 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,nodename]=tracker.split('@')
 114                     nodename=nodename.split('.')[0]
 115                     print self.margin_outline(nodename),tracker
 116                 except:
 117                     print self.margin(""),tracker
 118
 119     def handle_build_box (self,box):
 120         if not self.options.probe:
 121             self.reboot(box)
 122         else:
 123             command=['uptime']
 124             uptime=self.backquote_ssh(box,command,True).strip()
 125
 126             command=['pgrep','build']
 127             if self.options.dry_run:
 128                 self.run_ssh(box,command,None)
 129             else:
 130                 pids=self.backquote_ssh(box,command,True)
 131                 if not pids:
 132                     self.header ('No build process on %s (%s)'%(box,uptime))
 133                 else:
 134                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 135                     self.run_ssh(box,command,"Active build processes on %s (%s)"%(box,uptime),True)
 136
 137     vplc_matcher = re.compile(".*(vplc[0-9]+$)")
 138     def vplcname (self, vservername):
 139         match = self.vplc_matcher.match(vservername)
 140         if match: return match.groups(0)
 141         else: return ""
 142
 143     margin_format="%-14s"
 144     def margin(self,string): return self.margin_format%string
 145     def outline (self, string): return '== %s =='%string
 146     def margin_outline (self, string): return self.margin(self.outline(string))
 147
 148     def handle_plc_box (self,box):
 149         if not self.options.probe:
 150             self.reboot(box)
 151         else:
 152             command=['vserver-stat']
 153             if self.options.dry_run:
 154                 self.run_ssh(box,command,"Active vservers on %s"%box)
 155             else:
 156                 # try to find fullname (vserver_stat truncates to a ridiculously short name)
 157                 try:
 158                     self.header ("vserver map on %s"%box)
 159                     # fetch the contexts for all vservers on that box
 160                     map_command=['grep','.','/etc/vservers/*/context','/dev/null',]
 161                     context_map=self.backquote_ssh (box,map_command)
 162                     # at this point we have a set of lines like
 163                     # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
 164                     ctx_dict={}
 165                     for map_line in context_map.split("\n"):
 166                         if not map_line: continue
 167                         [path,xid] = map_line.split(':')
 168                         ctx_dict[xid]=os.path.basename(os.path.dirname(path))
 169                     # at this point ctx_id maps context id to vservername
 170
 171                     vserver_stat = self.backquote_ssh (box,command)
 172                     for vserver_line in vserver_stat.split("\n"):
 173                         if not vserver_line: continue
 174                         context=vserver_line.split()[0]
 175                         if context=="CTX":
 176                             print self.margin(""),vserver_line
 177                             continue
 178                         longname=ctx_dict[context]
 179                         print self.margin_outline(self.vplcname(longname)),"%(vserver_line)s [=%(longname)s]"%locals()
 180                 except:
 181                     self.run_ssh(box,command,"Fine-grained method failed - fallback to plain vserver-stat")
 182
 183     vnode_matcher = re.compile(".*(vnode[0-9]+)")
 184     def vnodename (self, ps_line):
 185         match = self.vnode_matcher.match(ps_line)
 186         if match: return match.groups(0)
 187         else: return ""
 188
 189     def handle_qemu_box (self,box):
 190         if not self.options.probe:
 191             self.reboot(box)
 192         else:
 193             command=['lsmod']
 194             modules=self.backquote_ssh(box,command).split('\n')
 195             kqemu_msg='*NO kqemu MODULE LOADED*'
 196             for module in modules:
 197                 if module.find('kqemu')==0:
 198                     kqemu_msg='kqemu OK'
 199
 200             command=['pgrep','qemu']
 201             if self.options.dry_run:
 202                 self.run_ssh(box,command,None)
 203             else:
 204                 pids=self.backquote_ssh(box,command)
 205                 if not pids:
 206                     self.header ('No qemu process on %s'%box)
 207                 else:
 208                     self.header ("Active qemu processes on %s (%s)"%(box,kqemu_msg))
 209                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 210                     ps_lines = self.backquote_ssh (box,command).split("\n")
 211                     for ps_line in ps_lines:
 212                         if not ps_line or ps_line.find('PID') >=0 : continue
 213                         print self.margin_outline(self.vnodename(ps_line)), ps_line
 214
 215     # the ouput of ps -o pid,command gives us <pid> bash <buildname>/run_log
 216     def testmaster_buildname (self, ps_line):
 217         chunks=ps_line.split()
 218         path=chunks[2]
 219         [buildname,command]=path.split('/')
 220         return buildname
 221
 222     def handle_testmaster_box (self, box):
 223             command=['pgrep','run_log']
 224             if self.options.dry_run:
 225                 self.run_ssh(box,command,None)
 226             else:
 227                 pids=self.backquote_ssh(box,command)
 228                 if not pids:
 229                     self.header ('No run_log process on %s'%box)
 230                 else:
 231                     self.header ("Active run_log processes on %s"%(box))
 232                     command=['ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 233                     ps_lines = self.backquote_ssh (box,command).split("\n")
 234                     for ps_line in ps_lines:
 235                         if not ps_line or ps_line.find('PID') >=0 : continue
 236                         print self.margin_outline(self.testmaster_buildname(ps_line)), ps_line
 237
 238
 239     def handle_box(self,box,type):
 240         if box in self.qemu_boxes:
 241             if type=="qemu": self.handle_qemu_box(self.fqdn(box))
 242         elif box in self.plc_boxes:
 243             if type=="plc":  self.handle_plc_box(self.fqdn(box))
 244         elif box in self.testmaster_boxes:
 245             if type=='testmaster': self.handle_testmaster_box(self.fqdn(box))
 246         elif type=="build":
 247             self.handle_build_box(self.fqdn(box))
 248
 249     def handle_disk (self,box):
 250         box=self.fqdn(box)
 251         return self.run_ssh(box,["df","-h",],"Disk space on %s"%box)
 252
 253     def main (self):
 254         usage="""%prog [options] [hostname..(s)]
 255 Default is to act on test boxes only"""
 256         parser = OptionParser (usage=usage)
 257         parser.add_option ("-n","--dry-run",action="store_true",dest="dry_run",default=False,
 258                            help="Dry run")
 259         parser.add_option ("-r","--reboot", action="store_false",dest="probe",default=True,
 260                            help="Actually reset/reboot stuff instead of just probing it")
 261         # no need for -p = probe, as this is the default
 262         parser.add_option ("-p","--plc", action="store_true",dest="plc_only",default=False,
 263                            help="Acts on the plc box only")
 264
 265         parser.add_option ("-a","--all",action="store_true",dest="all_boxes",default=False,
 266                            help="Acts on build and test boxes")
 267         parser.add_option ("-b","--build",action="store_true",dest="build_only",default=False,
 268                            help="Acts on build boxes only")
 269         parser.add_option ("-q","--qemu",action="store_true",dest="qemu_only",default=False,
 270                            help="Only acts on the qemu boxes")
 271         parser.add_option ("-t","--trackers",action="store_true",dest="trackers_only",default=False,
 272                            help="Only wipes trackers")
 273         parser.add_option ("-m","--master",action="store_true",dest="testmaster_only",default=False,
 274                            help="Display the testmaster status")
 275         parser.add_option ("-d","--disk",action="store_true",dest="show_disk",default=False,
 276                            help="Only inspects disk status")
 277
 278         (self.options,args) = parser.parse_args()
 279
 280         # use given hostnames if provided
 281         if args:
 282             self.boxes=args
 283             # if hostnames are specified, let's stay on the safe side and don't reset trackers
 284             self.do_tracker_plcs = False
 285             self.do_tracker_qemus = False
 286         elif self.options.all_boxes:
 287             self.boxes=self.test_boxes + self.build_boxes
 288             self.do_tracker_plcs = True
 289             self.do_tracker_qemus = True
 290         elif self.options.build_only:
 291             self.boxes=self.build_boxes
 292             self.do_tracker_plcs = False
 293             self.do_tracker_qemus = False
 294         elif self.options.qemu_only:
 295             self.boxes=self.qemu_boxes
 296             self.do_tracker_plcs = False
 297             self.do_tracker_qemus = True
 298         elif self.options.plc_only:
 299             self.boxes=self.plc_boxes
 300             self.do_tracker_plcs = True
 301             self.do_tracker_qemus = False
 302         elif self.options.testmaster_only:
 303             self.boxes=self.testmaster_boxes
 304             self.do_tracker_plcs = False
 305             self.do_tracker_qemus = False
 306         elif self.options.trackers_only:
 307             self.boxes = []
 308             self.do_tracker_plcs = True
 309             self.do_tracker_qemus = True
 310         # default
 311         else:
 312             self.boxes = self.test_boxes
 313             self.do_tracker_plcs = True
 314             self.do_tracker_qemus = True
 315
 316         if self.options.show_disk:
 317             for box in self.boxes: self.handle_disk(box)
 318             return
 319
 320         # ALL OTHERS
 321         for box in self.boxes:  self.handle_box (box,"build")
 322         # TESTMASTER
 323         for box in self.boxes:  self.handle_box (box,"testmaster")
 324         # PLCS
 325         if self.do_tracker_plcs:self.handle_tracker_plcs ()
 326         for box in self.boxes:  self.handle_box (box,"plc")
 327         # QEMU
 328         if self.do_tracker_qemus:self.handle_tracker_qemus ()
 329         for box in self.boxes:  self.handle_box (box,"qemu")
 330
 331 if __name__ == "__main__":
 332     BuildBoxes().main()