qaapi/qa/tests/boot_node.py

   1 #!/usr/bin/python
   2
   3 import os
   4 import sys
   5 import signal
   6 import time
   7 import tempfile
   8 import select
   9 import base64
  10 import traceback
  11 from Test import Test
  12 from qa import utils
  13 from qa.Nodes import Node, Nodes
  14
  15 image_types = ['node-iso', 'node-usb', 'generic-iso', 'generic-usb']
  16
  17 def killqemu(pid):
  18     try:
  19         os.kill(pid, signal.SIGKILL)
  20         os.waitpid(pid,os.WNOHANG)
  21     except:
  22         pass
  23
  24 class boot_node(Test):
  25     """
  26     Attempts to boot the specified node using qemu.
  27     """
  28
  29     def state_nanny(self):
  30         hostname = self.hostname
  31         nodes = self.config.api.GetNodes(self.config.auth, [hostname], ['boot_state'])
  32         node = nodes[0]
  33         boot_state = node['boot_state']
  34         if True or self.config.verbose:
  35             utils.header("%(hostname)s boot_state is %(boot_state)s" % locals(), False)
  36
  37         if boot_state in ['boot']:
  38             self.exit = True
  39
  40         if self.config.verbose:
  41             if boot_state in ['boot']:
  42                 utils.header("%(hostname)s correctly installed and booted" % locals(), False)
  43             else:
  44                 utils.header("%(hostname)s not fully booted" % locals(), False)
  45         self.boot_state = boot_state
  46         return self.exit
  47
  48     def console_nanny(self,console_states):
  49         #ready = select.select([],[self.stdout],[],1)[1]
  50         output = 0
  51         if True or len(ready)>0:
  52             output = 1
  53             retry = True
  54             while retry:
  55                 try:
  56                     lines = self.stdout.readlines()
  57                     retry = False
  58                 except IOError, e:
  59                     pass
  60
  61             #for searchstring in console_states.keys():
  62             #    result = line.find(searchstring)
  63                 # if result... ret = console_states[searchstring]
  64             #    break
  65                 # check result for whether we found it
  66
  67             # for now just print it out
  68             for line in lines:
  69                 print line
  70                 utils.header(line)
  71             # should be parsing for special strings that indicate whether
  72             # we've reached a particular state within the boot sequence
  73
  74         return output
  75
  76     def call(self, plc_name, hostname, image_type = 'node-iso', disk_size="17G", wait = 30):
  77
  78
  79         # Get this nodes configuration
  80         node = self.config.get_node(hostname)
  81         # Which plc does this node talk to
  82         plc = self.config.get_plc(plc_name)
  83         api = plc.config.api
  84         auth = plc.config.auth
  85         path = node.get_path()
  86         host = node['host']
  87         homedir = node['homedir']
  88         tmpdir = '/usr/tmp/'
  89         bootimage_filename = "%(hostname)s-bootcd.iso" % locals()
  90         diskimage_filename = "%(hostname)s-hda.img" % locals()
  91         bootimage = "%(homedir)s/%(bootimage_filename)s" % locals()
  92         diskimage = "%(homedir)s/%(diskimage_filename)s" % locals()
  93         diskimage_path = "/%(path)s/%(diskimage_filename)s" % locals()
  94         bootimage_tmppath = "%(tmpdir)s/%(bootimage_filename)s" % locals()
  95         bootimage_path = "%(path)s/%(bootimage_filename)s" % locals()
  96         remote_bootimage_path = "%(homedir)s/%(bootimage_filename)s" % locals()
  97
  98         # wait up to 30 minutes for a node to boot and install itself correctly
  99         self.hostname = hostname
 100         self.totaltime = 60*60*wait
 101
 102         # validate hostname
 103         nodes = api.GetNodes(auth, [hostname], ['hostname', 'boot_state'])
 104         if not nodes:
 105             raise Exception, "%s not found at plc  %s" % (hostname, plc['name'])
 106         node.update(nodes[0])
 107
 108         # try reinstalling the node if it is in debug state
 109         if node['boot_state'] in ['dbg']:
 110             if self.config.verbose:
 111                 utils.header("%(hostname)s is in debug state. Attempting a re-install" % locals())
 112             api.UpdateNode(auth, node['node_id'], {'boot_state': 'rins'})
 113
 114         # Create boot image
 115         if self.config.verbose:
 116             utils.header("Creating bootcd for %(hostname)s at %(bootimage_path)s" % locals())
 117         nodeimage = api.GetBootMedium(auth, hostname, image_type, '', ['serial'])
 118         fp = open(bootimage_tmppath, 'w')
 119         fp.write(base64.b64decode(nodeimage))
 120         fp.close()
 121
 122         # Move the boot image to the nodes home directory
 123         node.host_commands("mkdir -p %(homedir)s" % locals())
 124         node.scp_to_host(bootimage_tmppath, "%(remote_bootimage_path)s" % locals())
 125
 126         # Create a temporary disk image if it doesnt already exist or we are reinstalling
 127         img_check_cmd =  "ls -ld %(diskimage)s" % locals()
 128         (status, output) = node.host_commands(img_check_cmd, False)
 129         if status != 0 or node['boot_state'] in ['rins', 'inst']:
 130             qemu_img_cmd = "qemu-img create -f qcow2 %(diskimage)s %(disk_size)s" % locals()
 131             node.host_commands(qemu_img_cmd)
 132
 133         if self.config.verbose:
 134             utils.header("Booting %(hostname)s" % locals())
 135
 136         # Attempt to boot this node image
 137
 138         # generate a temp filename to which qemu should store its pid (crappy approach)
 139         tmp = tempfile.mkstemp(".pid","qemu_")
 140         pidfile=tmp[1]
 141         os.unlink(pidfile)
 142
 143         os.close(tmp[0])
 144         # boot node with ramsize memory
 145         ramsize=1024
 146
 147         # always use the 64 bit version of qemu, as this will work on both 32 & 64 bit host kernels
 148         bootcmd = "qemu-system-x86_64"
 149         # tell qemu to store its pid ina  file
 150         bootcmd = bootcmd + " -pidfile %(pidfile)s " % locals()
 151         # boot with ramsize memory
 152         bootcmd = bootcmd + " -m %(ramsize)s" % locals()
 153         # uniprocessor only
 154         bootcmd = bootcmd + " -smp 1"
 155         # redirect incomming tcp connections on specified port to guest node
 156         if 'redir_ssh_port' in node and node['redir_ssh_port']:
 157             port = node['redir_ssh_port']
 158             ip = node['nodenetworks'][0]['ip']
 159             bootcmd = bootcmd + " -redir tcp:%(port)s:%(ip)s:22" % locals()
 160         # no graphics support -> assume we are booting via serial console
 161         bootcmd = bootcmd + " -nographic"
 162         # boot from the supplied cdrom iso file
 163         bootcmd = bootcmd + " -boot d"
 164         bootcmd = bootcmd + " -cdrom %(bootimage)s" % locals()
 165         # hard disk image to use for the node
 166         bootcmd = bootcmd + " %(diskimage)s" % locals()
 167         # redirect stdout, stderr to logfile
 168         bootcmd = bootcmd + " 2>&1 >> %s " % (node.logfile.filename)
 169
 170         # kill any old qemu processes for this node
 171         pid_cmd = "ps -elfy | grep qemu | grep %(hostname)s | awk '{print$3}'" % locals()
 172         (status, output) = node.host_commands(pid_cmd)
 173         pids = " ".join(output.split("\n")).strip()
 174         if pids:
 175             kill_cmd = "kill %(pids)s" % locals()
 176             (status, output) = node.host_commands(kill_cmd)
 177
 178         time.sleep(2)
 179         # launch qemu
 180         (self.stdin, self.stdout, self.stderr) = node.host_popen3(bootcmd)
 181
 182         # wait for qemu to start up
 183         time.sleep(5)
 184         # get qemu's pid from its pidfile (crappy approach)
 185         pid_cmd = "cat %(pidfile)s" % locals()
 186         (staus, output)  = node.host_commands(pid_cmd)
 187         self.pid = output.strip()
 188         #fp = file(pidfile)
 189         #buf=fp.read()
 190         #self.pid=int(buf)
 191         #fp.close()
 192         #os.unlink(pidfile)
 193
 194         # loop until the node is either fully booted, some error
 195         # occured, or we've reached our totaltime out
 196         def catch(sig, frame):
 197             self.totaltime = self.totaltime -1
 198             utils.header("beep %d\n" %self.totaltime, False)
 199             total = self.totaltime
 200             if (total == 0) or \
 201                    (((total % 60)==0) and self.state_nanny()):
 202                 killqemu(self.pid)
 203                 self.exit = True
 204             else:
 205                 signal.alarm(1)
 206
 207         try:
 208             signal.signal(signal.SIGALRM, catch)
 209             signal.alarm(1)
 210             self.exit = False
 211
 212             #console_states = {"login:":1}
 213             while not self.exit:
 214                 pass
 215                 #try:
 216                 #    self.console_nanny(console_states)
 217                 #except: # need a better way to catch exceptions
 218                 #    traceback.print_exc()
 219                 #    pass
 220
 221             signal.alarm(0)
 222         except:
 223             signal.alarm(0)
 224
 225         killqemu(self.pid)
 226         return 1
 227
 228 if __name__ == '__main__':
 229     args = tuple(sys.argv[1:])
 230     boot_node()(*args)