scripts/manage-infrastructure.py

   1 #!/usr/bin/python
   2
   3 import os.path
   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 = 'onelab.eu'
  12     testmaster = 'testmaster'
  13     build_boxes = [ "mirror", "liquid", "reed", "velvet", ]
  14     plc_boxes = [ "testplc" ]
  15     qemu_boxes = \
  16         [ "testqemu%d"%i for i in range (1,4) ] + \
  17         [ "testqemu32-%d"%i for i in range (1,6) ]
  18     test_boxes = plc_boxes + qemu_boxes
  19
  20     def __init__ (self):
  21         # dummy defaults
  22         self.boxes = []
  23         self.do_tracker = False
  24
  25     def fqdn (self, box):
  26         return "%s.%s"%(box,self.domain)
  27     @staticmethod
  28     def root (box): return "root@%s"%box
  29
  30     def header (self,message):
  31         print "===============",message
  32
  33     def run (self,argv,message):
  34         if self.options.dry_run:
  35             print 'DRY_RUN:',
  36             print " ".join(argv)
  37         else:
  38             if message: self.header(message)
  39             subprocess.call(argv)
  40
  41     def backquote (self, argv):
  42         return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
  43
  44     def reboot (self,box):
  45         command=['ssh',self.root(box),'shutdown','-r','now']
  46         self.run (command,"Rebooting %s"%box)
  47
  48     def handle_trackers (self):
  49         box = self.fqdn (self.testmaster)
  50         if self.options.probe:
  51             command=['ssh',self.root(box),"head","-100","tracker*"]
  52             self.run(command,"Inspecting trackers on %s"%box)
  53         else:
  54             command=['ssh',self.root(box),"rm","-rf","tracker*"]
  55             self.run(command,"Cleaning up trackers on %s"%box)
  56
  57     def handle_build_box (self,box):
  58         if not self.options.probe:
  59             self.reboot(box)
  60         else:
  61             command=['ssh',self.root(box),'uptime']
  62             uptime=self.backquote(command).strip()
  63
  64             command=['ssh',self.root(box),'pgrep','build']
  65             if self.options.dry_run:
  66                 self.run(command,None)
  67             else:
  68                 pids=self.backquote(command)
  69                 if not pids:
  70                     self.header ('No build process on %s (%s)'%(box,uptime))
  71                 else:
  72                     command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
  73                     self.run(command,"Active build processes on %s (%s)"%(box,uptime))
  74
  75     vplc_matcher = re.compile(".*(vplc[0-9]+$)")
  76     def vplcname (self, vservername):
  77         match = self.vplc_matcher.match(vservername)
  78         if match: return match.groups(0)
  79         else: return ""
  80
  81     def handle_plc_box (self,box):
  82         if not self.options.probe:
  83             self.reboot(box)
  84         else:
  85             command=['ssh',self.root(box),'vserver-stat']
  86             if self.options.dry_run:
  87                 self.run(command,"Active vservers on %s"%box)
  88             else:
  89                 # try to find fullname (vserver_stat truncates to a ridiculously short name)
  90                 try:
  91                     self.header ("vserver map on %s"%box)
  92                     # fetch the contexts for all vservers on that box
  93                     map_command=['ssh',self.root(box),'grep','.','/etc/vservers/*/context','/dev/null',]
  94                     context_map=self.backquote (map_command)
  95                     # at this point we have a set of lines like
  96                     # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
  97                     ctx_dict={}
  98                     for map_line in context_map.split("\n"):
  99                         if not map_line: continue
 100                         [path,xid] = map_line.split(':')
 101                         ctx_dict[xid]=os.path.basename(os.path.dirname(path))
 102                     # at this point ctx_id maps context id to vservername
 103
 104                     vserver_stat = self.backquote (command)
 105                     for vserver_line in vserver_stat.split("\n"):
 106                         if not vserver_line: continue
 107                         context=vserver_line.split()[0]
 108                         if context=="CTX":
 109                             print vserver_line
 110                             continue
 111                         longname=ctx_dict[context]
 112                         plcname=self.vplcname(longname)
 113                         if plcname: print "== %s =="%plcname
 114                         print "%(vserver_line)s [=%(longname)s]"%locals()
 115                 except:
 116                     self.run(command,"Fine-grained method failed - fallback to plain vserver-stat")
 117
 118     vnode_matcher = re.compile(".*(vnode[0-9]+)")
 119     def vnodename (self, ps_line):
 120         match = self.vnode_matcher.match(ps_line)
 121         if match: return match.groups(0)
 122         else: return ""
 123
 124
 125     def handle_qemu_box (self,box):
 126         if not self.options.probe:
 127             self.reboot(box)
 128         else:
 129             command=['ssh',self.root(box),'pgrep','qemu']
 130             if self.options.dry_run:
 131                 self.run(command,None)
 132             else:
 133                 pids=self.backquote(command)
 134                 if not pids:
 135                     self.header ('No qemu process on %s'%box)
 136                 else:
 137                     self.header ("Active qemu processes on %s"%box)
 138                     command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 139                     ps_lines = self.backquote (command).split("\n")
 140                     for ps_line in ps_lines:
 141                         if not ps_line or ps_line.find('PID') >=0 : continue
 142                         node=self.vnodename(ps_line)
 143                         if node: print "== %s =="%node
 144                         print ps_line
 145
 146     def handle_box(self,box):
 147         if box in self.qemu_boxes:
 148             self.handle_qemu_box(self.fqdn(box))
 149         elif box in self.plc_boxes:
 150             self.handle_plc_box(self.fqdn(box))
 151         else:
 152             self.handle_build_box(self.fqdn(box))
 153
 154     def main (self):
 155         usage="""%prog [options] [hostname..(s)]
 156 Default is to act on test boxes only (with trackers clean)"""
 157         parser = OptionParser (usage=usage)
 158         parser.add_option ("-n","--dry-run",action="store_true",dest="dry_run",default=False,
 159                            help="Dry run")
 160         parser.add_option ("-r","--reboot", action="store_false",dest="probe",default=True,
 161                            help="Actually reset/reboot stuff instead of just probing it")
 162         # no need for -p = probe, as this is the default
 163         parser.add_option ("-p","--plc", action="store_true",dest="plc_only",default=False,
 164                            help="Acts on the plc box only")
 165
 166         parser.add_option ("-a","--all",action="store_true",dest="all_boxes",default=False,
 167                            help="Acts on build and test boxes")
 168         parser.add_option ("-b","--build",action="store_true",dest="build_only",default=False,
 169                            help="Acts on build boxes only")
 170         parser.add_option ("-q","--qemu",action="store_true",dest="qemu_only",default=False,
 171                            help="Only acts on the qemu boxes")
 172         parser.add_option ("-t","--trackers",action="store_true",dest="trackers_only",default=False,
 173                            help="Only wipes trackers")
 174
 175         (self.options,args) = parser.parse_args()
 176
 177         # use given hostnames if provided
 178         if args:
 179             self.boxes=args
 180             # if hostnames are specified, let's stay on the safe side and don't reset trackers
 181             self.do_tracker = False
 182         elif self.options.all_boxes:
 183             self.boxes=self.test_boxes + self.build_boxes
 184             self.do_tracker = True
 185         elif self.options.build_only:
 186             self.boxes=self.build_boxes
 187             self.do_tracker = False
 188         elif self.options.qemu_only:
 189             self.boxes=self.qemu_boxes
 190             self.do_tracker = False
 191         elif self.options.plc_only:
 192             self.boxes=self.plc_boxes
 193             self.do_tracker = False
 194         elif self.options.trackers_only:
 195             self.boxes = []
 196             self.do_tracker = True
 197         # default
 198         else:
 199             self.boxes = self.test_boxes
 200             self.do_tracker = True
 201
 202         if self.do_tracker:
 203             self.handle_trackers ()
 204         for box in self.boxes:
 205             self.handle_box (box)
 206
 207
 208 if __name__ == "__main__":
 209     BuildBoxes().main()