git.barrelfish.org Git - barrelfish/blob - tools/harness/machines/uw.py

   1 ##########################################################################
   2 # Copyright (c) 2013, 2014, University of Washington.
   3 # All rights reserved.
   4 #
   5 # This file is distributed under the terms in the attached LICENSE file.
   6 # If you do not find this file, copies can be found by writing to:
   7 # ETH Zurich D-INFK, Haldeneggsteig 4, CH-8092 Zurich. Attn: Systems Group.
   8 ##########################################################################
   9
  10 import sys, os, signal, time, getpass, subprocess, socket, pty
  11 import debug, machines, uw_machinedata
  12 from machines import Machine, MachineLockedError, MachineFactory
  13
  14 TFTP_PATH='/var/lib/tftpboot'
  15 TOOLS_PATH='/usr/local/bin'
  16 RACKBOOT=os.path.join(TOOLS_PATH, 'rackboot.sh')
  17 RACKPOWER=os.path.join(TOOLS_PATH, 'rackpower')
  18
  19 class UWMachine(Machine):
  20     _uw_machines = uw_machinedata.machines
  21
  22     host2mgmt = {
  23         'bigfish.cs.washington.edu': 'bigfish-e1k1.cs.washington.edu',
  24         'swingout1.cs.washington.edu': 'swingout1-brcm1.cs.washington.edu',
  25         'swingout5.cs.washington.edu': 'swingout5-brcm1.cs.washington.edu'
  26         }
  27
  28     def __init__(self, options):
  29         super(UWMachine, self).__init__(options)
  30         self.lockprocess = None
  31         self.masterfd = None
  32
  33     def get_bootarch(self):
  34         b = self._uw_machines[self.name]['bootarch']
  35         assert(b in self.get_buildarchs())
  36         return b
  37
  38     def get_machine_name(self):
  39         return self._uw_machines[self.name]['machine_name']
  40
  41     def get_buildarchs(self):
  42         return self._uw_machines[self.name]['buildarchs']
  43
  44     def get_ncores(self):
  45         return self._uw_machines[self.name]['ncores']
  46
  47     def get_cores_per_socket(self):
  48         return self._uw_machines[self.name]['cores_per_socket']
  49
  50     def get_tickrate(self):
  51         return self._uw_machines[self.name]['tickrate']
  52
  53     def get_perfcount_type(self):
  54         return self._uw_machines[self.name]['perfcount_type']
  55
  56     def get_kernel_args(self):
  57         return self._uw_machines[self.name].get('kernel_args')
  58
  59     def get_pci_args(self):
  60         return self._uw_machines[self.name].get('pci_args')
  61
  62     def get_boot_timeout(self):
  63         return self._uw_machines[self.name].get('boot_timeout')
  64
  65     def get_hostname(self):
  66         return self.get_machine_name() + '.cs.washington.edu'
  67
  68     def get_ip(self):
  69         return socket.gethostbyname(self.host2mgmt[self.get_hostname()])
  70
  71     def get_tftp_dir(self):
  72         user = getpass.getuser()
  73         return os.path.join(TFTP_PATH, user, self.name + "_harness")
  74
  75     def _write_menu_lst(self, data, path):
  76         debug.verbose('writing %s' % path)
  77         debug.debug(data)
  78         f = open(path, 'w')
  79         f.write(data)
  80         f.close()
  81
  82     def _set_menu_lst(self, relpath):
  83         ip_menu_name = os.path.join(TFTP_PATH, "menu.lst." + self.get_ip())
  84         debug.verbose('relinking %s to %s' % (ip_menu_name, relpath))
  85         os.remove(ip_menu_name)
  86         os.symlink(relpath, ip_menu_name)
  87
  88     def set_bootmodules(self, modules):
  89         fullpath = os.path.join(self.get_tftp_dir(), 'menu.lst')
  90         relpath = os.path.relpath(fullpath, TFTP_PATH)
  91         tftppath = '/' + os.path.relpath(self.get_tftp_dir(), TFTP_PATH)
  92         self._write_menu_lst(modules.get_menu_data(tftppath), fullpath)
  93         self._set_menu_lst(relpath)
  94
  95     def lock(self):
  96         """Use conserver to lock the machine."""
  97
  98         # find out current status of console
  99         debug.verbose('executing "console -i %s" to check state' %
 100                       self.get_machine_name())
 101         proc = subprocess.Popen(["console", "-i", self.get_machine_name()],
 102                                 stdout=subprocess.PIPE)
 103         line = proc.communicate()[0]
 104         assert(proc.returncode == 0)
 105
 106         # check that nobody else has it open for writing
 107         myuser = getpass.getuser()
 108         parts = line.strip().split(':')
 109         conname, child, contype, details, users, state = parts[:6]
 110         if users:
 111             for userinfo in users.split(','):
 112                 mode, username, host, port = userinfo.split('@')[:4]
 113                 if 'w' in mode and username != myuser:
 114                     raise MachineLockedError # Machine is not free
 115
 116         # run a console in the background to 'hold' the lock and read output
 117         debug.verbose('starting "console %s"' % self.get_machine_name())
 118         # run on a PTY to work around terminal mangling code in console
 119         (self.masterfd, slavefd) = pty.openpty()
 120         self.lockprocess = subprocess.Popen(["console", self.get_machine_name()],
 121                                             close_fds=True,
 122                                             stdout=slavefd, stdin=slavefd)
 123         os.close(slavefd)
 124         # XXX: open in binary mode with no buffering
 125         # otherwise select.select() may block when there is data in the buffer
 126         self.console_out = os.fdopen(self.masterfd, 'rb', 0)
 127
 128     def unlock(self):
 129         if self.lockprocess is None:
 130             return # noop
 131         debug.verbose('quitting console process (%d)' % self.lockprocess.pid)
 132         # os.kill(self.lockprocess.pid, signal.SIGTERM)
 133         os.write(self.masterfd, "\x05c.")
 134         self.lockprocess.wait()
 135         self.lockprocess = None
 136         self.masterfd = None
 137
 138     def __rackboot(self, args):
 139         debug.checkcmd([RACKBOOT] + args + [self.get_machine_name()])
 140
 141     def setup(self):
 142         self.__rackboot(["-b", "-n"])
 143
 144     def __rackpower(self, arg):
 145         retries = 3
 146         failed = False
 147         while retries > 0:
 148             try:
 149                 debug.checkcmd([RACKPOWER, arg, self.get_machine_name()])
 150             except subprocess.CalledProcessError:
 151                 debug.warning("rackpower %s %s failed" %
 152                         (arg, self.get_machine_name()))
 153                 failed = True
 154                 if retries > 0:
 155                     debug.verbose("retrying...")
 156                     retries -= 1
 157             if not failed:
 158                 break
 159
 160     def reboot(self):
 161         self.__rackpower('-r')
 162
 163     def shutdown(self):
 164         self.__rackpower('-d')
 165
 166     def get_output(self):
 167         return self.console_out
 168
 169
 170 for n in sorted(UWMachine._uw_machines.keys()):
 171     class TmpMachine(UWMachine):
 172         name = n
 173     MachineFactory.addMachine(n, TmpMachine, **UWMachine._uw_machines[n])