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_qemus = False
  24         self.do_tracker_plcs = False
  25
  26     def fqdn (self, box):
  27         return "%s.%s"%(box,self.domain)
  28     @staticmethod
  29     def root (box): return "root@%s"%box
  30
  31     def header (self,message):
  32         print "===============",message
  33
  34     def run (self,argv,message, trash_err=False):
  35         if self.options.dry_run:
  36             print 'DRY_RUN:',
  37             print " ".join(argv)
  38         else:
  39             if message: self.header(message)
  40             if not trash_err:
  41                 subprocess.call(argv)
  42             else:
  43                 subprocess.call(argv,stderr=file('/dev/null','w'))
  44
  45     def backquote (self, argv, trash_err=False):
  46         if not trash_err:
  47             return subprocess.Popen(argv,stdout=subprocess.PIPE).communicate()[0]
  48         else:
  49             null = open('/dev/null','w')
  50             result = subprocess.Popen(argv,stdout=subprocess.PIPE,stderr=null).communicate()[0]
  51             null.close()
  52             return result
  53
  54     def reboot (self,box):
  55         command=['ssh',self.root(box),'shutdown','-r','now']
  56         self.run (command,"Rebooting %s"%box)
  57
  58     def handle_tracker_file (self,filename):
  59         box = self.fqdn (self.testmaster)
  60         if self.options.probe:
  61             command=['ssh',self.root(box),"head","-v","--lines=100",filename]
  62             self.run(command,"++++++++++ Inspecting %s on %s"%(filename,box))
  63         else:
  64             command=['ssh',self.root(box),"rm","-rf",filename]
  65             self.run(command,"Cleaning up %s on %s"%(filaneme,box))
  66
  67     def handle_tracker_plcs (self):
  68         self.handle_tracker_file("tracker-plcs")
  69     def handle_tracker_qemus (self):
  70         self.handle_tracker_file("tracker-qemus")
  71     def handle_trackers (self):
  72         self.handle_tracker_plcs()
  73         self.handle_tracker_qemus()
  74
  75     def handle_build_box (self,box):
  76         if not self.options.probe:
  77             self.reboot(box)
  78         else:
  79             command=['ssh',self.root(box),'uptime']
  80             uptime=self.backquote(command,True).strip()
  81
  82             command=['ssh',self.root(box),'pgrep','build']
  83             if self.options.dry_run:
  84                 self.run(command,None)
  85             else:
  86                 pids=self.backquote(command,True)
  87                 if not pids:
  88                     self.header ('No build process on %s (%s)'%(box,uptime))
  89                 else:
  90                     command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
  91                     self.run(command,"Active build processes on %s (%s)"%(box,uptime),True)
  92
  93     vplc_matcher = re.compile(".*(vplc[0-9]+$)")
  94     def vplcname (self, vservername):
  95         match = self.vplc_matcher.match(vservername)
  96         if match: return match.groups(0)
  97         else: return ""
  98
  99     def handle_plc_box (self,box):
 100         if not self.options.probe:
 101             self.reboot(box)
 102         else:
 103             command=['ssh',self.root(box),'vserver-stat']
 104             if self.options.dry_run:
 105                 self.run(command,"Active vservers on %s"%box)
 106             else:
 107                 # try to find fullname (vserver_stat truncates to a ridiculously short name)
 108                 try:
 109                     self.header ("vserver map on %s"%box)
 110                     # fetch the contexts for all vservers on that box
 111                     map_command=['ssh',self.root(box),'grep','.','/etc/vservers/*/context','/dev/null',]
 112                     context_map=self.backquote (map_command)
 113                     # at this point we have a set of lines like
 114                     # /etc/vservers/2010.01.20--k27-f12-32-vplc03/context:40144
 115                     ctx_dict={}
 116                     for map_line in context_map.split("\n"):
 117                         if not map_line: continue
 118                         [path,xid] = map_line.split(':')
 119                         ctx_dict[xid]=os.path.basename(os.path.dirname(path))
 120                     # at this point ctx_id maps context id to vservername
 121
 122                     vserver_stat = self.backquote (command)
 123                     format="%-14s"
 124                     for vserver_line in vserver_stat.split("\n"):
 125                         if not vserver_line: continue
 126                         context=vserver_line.split()[0]
 127                         if context=="CTX":
 128                             print format%"",vserver_line
 129                             continue
 130                         longname=ctx_dict[context]
 131                         plcname=self.vplcname(longname)
 132                         header="== %s =="%plcname
 133                         print format%header,"%(vserver_line)s [=%(longname)s]"%locals()
 134                 except:
 135                     self.run(command,"Fine-grained method failed - fallback to plain vserver-stat")
 136
 137     vnode_matcher = re.compile(".*(vnode[0-9]+)")
 138     def vnodename (self, ps_line):
 139         match = self.vnode_matcher.match(ps_line)
 140         if match: return match.groups(0)
 141         else: return ""
 142
 143
 144     def handle_qemu_box (self,box):
 145         if not self.options.probe:
 146             self.reboot(box)
 147         else:
 148             command=['ssh',self.root(box),'pgrep','qemu']
 149             if self.options.dry_run:
 150                 self.run(command,None)
 151             else:
 152                 pids=self.backquote(command)
 153                 if not pids:
 154                     self.header ('No qemu process on %s'%box)
 155                 else:
 156                     format="%-14s"
 157                     self.header ("Active qemu processes on %s"%box)
 158                     command=['ssh',self.root(box),'ps','-o','pid,command'] + [ pid for pid in pids.split("\n") if pid]
 159                     ps_lines = self.backquote (command).split("\n")
 160                     for ps_line in ps_lines:
 161                         if not ps_line or ps_line.find('PID') >=0 : continue
 162                         header=self.vnodename(ps_line)
 163                         print format%header, ps_line
 164
 165     def handle_box(self,box,type):
 166         if box in self.qemu_boxes:
 167             if type=="qemu": self.handle_qemu_box(self.fqdn(box))
 168         elif box in self.plc_boxes:
 169             if type=="plc":  self.handle_plc_box(self.fqdn(box))
 170         elif type=="build":
 171             self.handle_build_box(self.fqdn(box))
 172
 173     def main (self):
 174         usage="""%prog [options] [hostname..(s)]
 175 Default is to act on test boxes only (with trackers clean)"""
 176         parser = OptionParser (usage=usage)
 177         parser.add_option ("-n","--dry-run",action="store_true",dest="dry_run",default=False,
 178                            help="Dry run")
 179         parser.add_option ("-r","--reboot", action="store_false",dest="probe",default=True,
 180                            help="Actually reset/reboot stuff instead of just probing it")
 181         # no need for -p = probe, as this is the default
 182         parser.add_option ("-p","--plc", action="store_true",dest="plc_only",default=False,
 183                            help="Acts on the plc box only")
 184
 185         parser.add_option ("-a","--all",action="store_true",dest="all_boxes",default=False,
 186                            help="Acts on build and test boxes")
 187         parser.add_option ("-b","--build",action="store_true",dest="build_only",default=False,
 188                            help="Acts on build boxes only")
 189         parser.add_option ("-q","--qemu",action="store_true",dest="qemu_only",default=False,
 190                            help="Only acts on the qemu boxes")
 191         parser.add_option ("-t","--trackers",action="store_true",dest="trackers_only",default=False,
 192                            help="Only wipes trackers")
 193
 194         (self.options,args) = parser.parse_args()
 195
 196         # use given hostnames if provided
 197         if args:
 198             self.boxes=args
 199             # if hostnames are specified, let's stay on the safe side and don't reset trackers
 200             self.do_tracker_plcs = False
 201             self.do_tracker_qemus = False
 202         elif self.options.all_boxes:
 203             self.boxes=self.test_boxes + self.build_boxes
 204             self.do_tracker_plcs = True
 205             self.do_tracker_qemus = True
 206         elif self.options.build_only:
 207             self.boxes=self.build_boxes
 208             self.do_tracker_plcs = False
 209             self.do_tracker_qemus = False
 210         elif self.options.qemu_only:
 211             self.boxes=self.qemu_boxes
 212             self.do_tracker_plcs = False
 213             self.do_tracker_qemus = True
 214         elif self.options.plc_only:
 215             self.boxes=self.plc_boxes
 216             self.do_tracker_plcs = True
 217             self.do_tracker_qemus = False
 218         elif self.options.trackers_only:
 219             self.boxes = []
 220             self.do_tracker_plcs = True
 221             self.do_tracker_qemus = True
 222         # default
 223         else:
 224             self.boxes = self.test_boxes
 225             self.do_tracker_plcs = True
 226             self.do_tracker_qemus = True
 227
 228         # ALL OTHERS
 229         for box in self.boxes:  self.handle_box (box,"build")
 230         # PLCS
 231         if self.do_tracker_plcs:self.handle_tracker_plcs ()
 232         for box in self.boxes:  self.handle_box (box,"plc")
 233         # QEMU
 234         if self.do_tracker_qemus:self.handle_tracker_qemus ()
 235         for box in self.boxes:  self.handle_box (box,"qemu")
 236
 237 if __name__ == "__main__":
 238     BuildBoxes().main()