system/TestPlc.py

   1 # Thierry Parmentelat <thierry.parmentelat@inria.fr>
   2 # Copyright (C) 2010 INRIA
   3 #
   4 import sys
   5 import time
   6 import os, os.path
   7 import traceback
   8 import socket
   9 from datetime import datetime, timedelta
  10
  11 import utils
  12 from Completer import Completer, CompleterTask
  13 from TestSite import TestSite
  14 from TestNode import TestNode, CompleterTaskNodeSsh
  15 from TestUser import TestUser
  16 from TestKey import TestKey
  17 from TestSlice import TestSlice
  18 from TestSliver import TestSliver
  19 from TestBoxQemu import TestBoxQemu
  20 from TestSsh import TestSsh
  21 from TestApiserver import TestApiserver
  22 from TestAuthSfa import TestAuthSfa
  23 from PlcapiUrlScanner import PlcapiUrlScanner
  24
  25 from TestBonding import TestBonding
  26
  27 has_sfa_cache_filename="sfa-cache"
  28
  29 # step methods must take (self) and return a boolean (options is a member of the class)
  30
  31 def standby(minutes, dry_run):
  32     utils.header('Entering StandBy for {:d} mn'.format(minutes))
  33     if dry_run:
  34         print('dry_run')
  35     else:
  36         time.sleep(60*minutes)
  37     return True
  38
  39 def standby_generic(func):
  40     def actual(self):
  41         minutes = int(func.__name__.split("_")[1])
  42         return standby(minutes, self.options.dry_run)
  43     return actual
  44
  45 def node_mapper(method):
  46     def map_on_nodes(self, *args, **kwds):
  47         overall = True
  48         node_method = TestNode.__dict__[method.__name__]
  49         for test_node in self.all_nodes():
  50             if not node_method(test_node, *args, **kwds):
  51                 overall=False
  52         return overall
  53     # maintain __name__ for ignore_result
  54     map_on_nodes.__name__ = method.__name__
  55     # restore the doc text
  56     map_on_nodes.__doc__ = TestNode.__dict__[method.__name__].__doc__
  57     return map_on_nodes
  58
  59 def slice_mapper(method):
  60     def map_on_slices(self):
  61         overall = True
  62         slice_method = TestSlice.__dict__[method.__name__]
  63         for slice_spec in self.plc_spec['slices']:
  64             site_spec = self.locate_site (slice_spec['sitename'])
  65             test_site = TestSite(self,site_spec)
  66             test_slice = TestSlice(self,test_site,slice_spec)
  67             if not slice_method(test_slice, self.options):
  68                 overall=False
  69         return overall
  70     # maintain __name__ for ignore_result
  71     map_on_slices.__name__ = method.__name__
  72     # restore the doc text
  73     map_on_slices.__doc__ = TestSlice.__dict__[method.__name__].__doc__
  74     return map_on_slices
  75
  76 def bonding_redirector(method):
  77     bonding_name = method.__name__.replace('bonding_', '')
  78     def redirect(self):
  79         bonding_method = TestBonding.__dict__[bonding_name]
  80         return bonding_method(self.test_bonding)
  81     # maintain __name__ for ignore_result
  82     redirect.__name__ = method.__name__
  83     # restore the doc text
  84     redirect.__doc__ = TestBonding.__dict__[bonding_name].__doc__
  85     return redirect
  86
  87 # run a step but return True so that we can go on
  88 def ignore_result(method):
  89     def ignoring(self):
  90         # ssh_slice_ignore->ssh_slice
  91         ref_name = method.__name__.replace('_ignore', '').replace('force_', '')
  92         ref_method = TestPlc.__dict__[ref_name]
  93         result = ref_method(self)
  94         print("Actual (but ignored) result for {ref_name} is {result}".format(**locals()))
  95         return Ignored(result)
  96     name = method.__name__.replace('_ignore', '').replace('force_', '')
  97     ignoring.__name__ = name
  98     ignoring.__doc__ = "ignored version of " + name
  99     return ignoring
 100
 101 # a variant that expects the TestSlice method to return a list of CompleterTasks that
 102 # are then merged into a single Completer run to avoid wating for all the slices
 103 # esp. useful when a test fails of course
 104 # because we need to pass arguments we use a class instead..
 105 class slice_mapper__tasks(object):
 106     # could not get this to work with named arguments
 107     def __init__(self, timeout_minutes, silent_minutes, period_seconds):
 108         self.timeout = timedelta(minutes = timeout_minutes)
 109         self.silent = timedelta(minutes = silent_minutes)
 110         self.period = timedelta(seconds = period_seconds)
 111     def __call__(self, method):
 112         decorator_self=self
 113         # compute augmented method name
 114         method_name = method.__name__ + "__tasks"
 115         # locate in TestSlice
 116         slice_method = TestSlice.__dict__[ method_name ]
 117         def wrappee(self):
 118             tasks=[]
 119             for slice_spec in self.plc_spec['slices']:
 120                 site_spec = self.locate_site (slice_spec['sitename'])
 121                 test_site = TestSite(self, site_spec)
 122                 test_slice = TestSlice(self, test_site, slice_spec)
 123                 tasks += slice_method (test_slice, self.options)
 124             return Completer (tasks, message=method.__name__).\
 125                 run(decorator_self.timeout, decorator_self.silent, decorator_self.period)
 126         # restore the doc text from the TestSlice method even if a bit odd
 127         wrappee.__name__ = method.__name__
 128         wrappee.__doc__ = slice_method.__doc__
 129         return wrappee
 130
 131 def auth_sfa_mapper(method):
 132     def actual(self):
 133         overall = True
 134         auth_method = TestAuthSfa.__dict__[method.__name__]
 135         for auth_spec in self.plc_spec['sfa']['auth_sfa_specs']:
 136             test_auth = TestAuthSfa(self, auth_spec)
 137             if not auth_method(test_auth, self.options):
 138                 overall=False
 139         return overall
 140     # restore the doc text
 141     actual.__doc__ = TestAuthSfa.__dict__[method.__name__].__doc__
 142     return actual
 143
 144 class Ignored:
 145     def __init__(self, result):
 146         self.result = result
 147
 148 SEP = '<sep>'
 149 SEPSFA = '<sep_sfa>'
 150
 151 class TestPlc:
 152
 153     default_steps = [
 154         'show', SEP,
 155         'plcvm_delete','plcvm_timestamp','plcvm_create', SEP,
 156         'plc_install', 'plc_configure', 'plc_start', SEP,
 157         'keys_fetch', 'keys_store', 'keys_clear_known_hosts', SEP,
 158         'plcapi_urls','speed_up_slices', SEP,
 159         'initscripts', 'sites', 'nodes', 'slices', 'nodegroups', 'leases', SEP,
 160 # slices created under plcsh interactively seem to be fine but these ones don't have the tags
 161 # keep this our of the way for now
 162         'check_vsys_defaults_ignore', SEP,
 163 # run this first off so it's easier to re-run on another qemu box
 164         'qemu_kill_mine', 'nodestate_reinstall', 'qemu_local_init','bootcd', 'qemu_local_config', SEP,
 165         'qemu_clean_mine', 'qemu_export', 'qemu_cleanlog', SEP,
 166         'qemu_start', 'qemu_timestamp', 'qemu_nodefamily', SEP,
 167         'sfa_install_all', 'sfa_configure', 'cross_sfa_configure', 'sfa_start', 'sfa_import', SEPSFA,
 168         'sfi_configure@1', 'sfa_register_site@1','sfa_register_pi@1', SEPSFA,
 169         'sfa_register_user@1', 'sfa_update_user@1', 'sfa_register_slice@1', 'sfa_renew_slice@1', SEPSFA,
 170         'sfa_remove_user_from_slice@1','sfi_show_slice_researchers@1',
 171         'sfa_insert_user_in_slice@1','sfi_show_slice_researchers@1', SEPSFA,
 172         'sfa_discover@1', 'sfa_rspec@1', SEPSFA,
 173         'sfa_allocate@1', 'sfa_provision@1', 'sfa_describe@1', SEPSFA,
 174         'sfa_check_slice_plc@1', 'sfa_update_slice@1', SEPSFA,
 175         'sfi_list@1', 'sfi_show_site@1', 'sfa_utest@1', SEPSFA,
 176         # we used to run plcsh_stress_test, and then ssh_node_debug and ssh_node_boot
 177         # but as the stress test might take a while, we sometimes missed the debug mode..
 178         'probe_kvm_iptables',
 179         'ping_node', 'ssh_node_debug', 'plcsh_stress_test@1', SEP,
 180         'ssh_node_boot', 'node_bmlogs', 'ssh_slice', 'ssh_slice_basics', 'check_initscripts', SEP,
 181         'ssh_slice_sfa@1', SEPSFA,
 182         'sfa_rspec_empty@1', 'sfa_allocate_empty@1', 'sfa_provision_empty@1','sfa_check_slice_plc_empty@1', SEPSFA,
 183         'sfa_delete_slice@1', 'sfa_delete_user@1', SEPSFA,
 184         'cross_check_tcp@1', 'check_system_slice', SEP,
 185         # for inspecting the slice while it runs the first time
 186         #'fail',
 187         # check slices are turned off properly
 188         'debug_nodemanager',
 189         'empty_slices', 'ssh_slice_off', 'slice_fs_deleted_ignore', SEP,
 190         # check they are properly re-created with the same name
 191         'fill_slices', 'ssh_slice_again', SEP,
 192         'gather_logs_force', SEP,
 193         ]
 194     other_steps = [
 195         'export', 'show_boxes', 'super_speed_up_slices', SEP,
 196         'check_hooks', 'plc_stop', 'plcvm_start', 'plcvm_stop', SEP,
 197         'delete_initscripts', 'delete_nodegroups','delete_all_sites', SEP,
 198         'delete_sites', 'delete_nodes', 'delete_slices', 'keys_clean', SEP,
 199         'delete_leases', 'list_leases', SEP,
 200         'populate', SEP,
 201         'nodestate_show','nodestate_safeboot','nodestate_boot', 'nodestate_upgrade', SEP,
 202         'nodedistro_show','nodedistro_f14','nodedistro_f18', SEP,
 203         'nodedistro_f20', 'nodedistro_f21','nodedistro_f22', SEP,
 204         'qemu_list_all', 'qemu_list_mine', 'qemu_kill_all', SEP,
 205         'sfa_install_core', 'sfa_install_sfatables', 'sfa_install_plc', 'sfa_install_client', SEPSFA,
 206         'sfa_plcclean', 'sfa_dbclean', 'sfa_stop','sfa_uninstall', 'sfi_clean', SEPSFA,
 207         'sfa_get_expires', SEPSFA,
 208         'plc_db_dump' , 'plc_db_restore', SEP,
 209         'check_netflow','check_drl', SEP,
 210         'slice_fs_present', SEP,
 211         'standby_1_through_20','yes','no',SEP,
 212         'install_syslinux6', 'bonding_builds', 'bonding_nodes', SEP,
 213         ]
 214     default_bonding_steps = [
 215         'bonding_init_partial',
 216         'bonding_add_yum',
 217         'bonding_install_rpms', SEP,
 218         ]
 219
 220     @staticmethod
 221     def printable_steps(list):
 222         single_line = " ".join(list) + " "
 223         return single_line.replace(" "+SEP+" ", " \\\n").replace(" "+SEPSFA+" ", " \\\n")
 224     @staticmethod
 225     def valid_step(step):
 226         return step != SEP and step != SEPSFA
 227
 228     # turn off the sfa-related steps when build has skipped SFA
 229     # this was originally for centos5 but is still valid
 230     # for up to f12 as recent SFAs with sqlalchemy won't build before f14
 231     @staticmethod
 232     def _has_sfa_cached(rpms_url):
 233         if os.path.isfile(has_sfa_cache_filename):
 234             with open(has_sfa_cache_filename) as cache:
 235                 cached = cache.read() == "yes"
 236             utils.header("build provides SFA (cached):{}".format(cached))
 237             return cached
 238         # warning, we're now building 'sface' so let's be a bit more picky
 239         # full builds are expected to return with 0 here
 240         utils.header("Checking if build provides SFA package...")
 241         retcod = utils.system("curl --silent {}/ | grep -q sfa-".format(rpms_url)) == 0
 242         encoded = 'yes' if retcod else 'no'
 243         with open(has_sfa_cache_filename,'w') as cache:
 244             cache.write(encoded)
 245         return retcod
 246
 247     @staticmethod
 248     def check_whether_build_has_sfa(rpms_url):
 249         has_sfa = TestPlc._has_sfa_cached(rpms_url)
 250         if has_sfa:
 251             utils.header("build does provide SFA")
 252         else:
 253             # move all steps containing 'sfa' from default_steps to other_steps
 254             utils.header("SFA package not found - removing steps with sfa or sfi")
 255             sfa_steps = [ step for step in TestPlc.default_steps
 256                           if step.find('sfa') >= 0 or step.find("sfi") >= 0 ]
 257             TestPlc.other_steps += sfa_steps
 258             for step in sfa_steps:
 259                 TestPlc.default_steps.remove(step)
 260
 261     def __init__(self, plc_spec, options):
 262         self.plc_spec = plc_spec
 263         self.options = options
 264         self.test_ssh = TestSsh(self.plc_spec['host_box'], self.options.buildname)
 265         self.vserverip = plc_spec['vserverip']
 266         self.vservername = plc_spec['vservername']
 267         self.vplchostname = self.vservername.split('-')[-1]
 268         self.url = "https://{}:443/PLCAPI/".format(plc_spec['vserverip'])
 269         self.apiserver = TestApiserver(self.url, options.dry_run)
 270         (self.ssh_node_boot_timeout, self.ssh_node_boot_silent) = plc_spec['ssh_node_boot_timers']
 271         (self.ssh_node_debug_timeout, self.ssh_node_debug_silent) = plc_spec['ssh_node_debug_timers']
 272
 273     def has_addresses_api(self):
 274         return self.apiserver.has_method('AddIpAddress')
 275
 276     def name(self):
 277         name = self.plc_spec['name']
 278         return "{}.{}".format(name,self.vservername)
 279
 280     def hostname(self):
 281         return self.plc_spec['host_box']
 282
 283     def is_local(self):
 284         return self.test_ssh.is_local()
 285
 286     # define the API methods on this object through xmlrpc
 287     # would help, but not strictly necessary
 288     def connect(self):
 289         pass
 290
 291     def actual_command_in_guest(self,command, backslash=False):
 292         raw1 = self.host_to_guest(command)
 293         raw2 = self.test_ssh.actual_command(raw1, dry_run=self.options.dry_run, backslash=backslash)
 294         return raw2
 295
 296     def start_guest(self):
 297       return utils.system(self.test_ssh.actual_command(self.start_guest_in_host(),
 298                                                        dry_run=self.options.dry_run))
 299
 300     def stop_guest(self):
 301       return utils.system(self.test_ssh.actual_command(self.stop_guest_in_host(),
 302                                                        dry_run=self.options.dry_run))
 303
 304     def run_in_guest(self, command, backslash=False):
 305         raw = self.actual_command_in_guest(command, backslash)
 306         return utils.system(raw)
 307
 308     def run_in_host(self,command):
 309         return self.test_ssh.run_in_buildname(command, dry_run=self.options.dry_run)
 310
 311     # backslashing turned out so awful at some point that I've turned off auto-backslashing
 312     # see e.g. plc_start esp. the version for f14
 313     #command gets run in the plc's vm
 314     def host_to_guest(self, command):
 315         ssh_leg = TestSsh(self.vplchostname)
 316         return ssh_leg.actual_command(command, keep_stdin=True)
 317
 318     # this /vservers thing is legacy...
 319     def vm_root_in_host(self):
 320         return "/vservers/{}/".format(self.vservername)
 321
 322     def vm_timestamp_path(self):
 323         return "/vservers/{}/{}.timestamp".format(self.vservername, self.vservername)
 324
 325     #start/stop the vserver
 326     def start_guest_in_host(self):
 327         return "virsh -c lxc:/// start {}".format(self.vservername)
 328
 329     def stop_guest_in_host(self):
 330         return "virsh -c lxc:/// destroy {}".format(self.vservername)
 331
 332     # xxx quick n dirty
 333     def run_in_guest_piped(self,local,remote):
 334         return utils.system(local+" | "+self.test_ssh.actual_command(self.host_to_guest(remote),
 335                                                                      keep_stdin = True))
 336
 337     def yum_check_installed(self, rpms):
 338         if isinstance(rpms, list):
 339             rpms=" ".join(rpms)
 340         return self.run_in_guest("rpm -q {}".format(rpms)) == 0
 341
 342     # does a yum install in the vs, ignore yum retcod, check with rpm
 343     def yum_install(self, rpms):
 344         if isinstance(rpms, list):
 345             rpms=" ".join(rpms)
 346         yum_mode = self.run_in_guest("yum -y install {}".format(rpms))
 347         if yum_mode != 0:
 348             self.run_in_guest("dnf -y install --allowerasing {}".format(rpms))
 349         # yum-complete-transaction comes with yum-utils, that is in vtest.pkgs
 350         self.run_in_guest("yum-complete-transaction -y")
 351         return self.yum_check_installed(rpms)
 352
 353     def auth_root(self):
 354         return {'Username'   : self.plc_spec['settings']['PLC_ROOT_USER'],
 355                 'AuthMethod' : 'password',
 356                 'AuthString' : self.plc_spec['settings']['PLC_ROOT_PASSWORD'],
 357                 'Role'       : self.plc_spec['role'],
 358                 }
 359
 360     def locate_site(self,sitename):
 361         for site in self.plc_spec['sites']:
 362             if site['site_fields']['name'] == sitename:
 363                 return site
 364             if site['site_fields']['login_base'] == sitename:
 365                 return site
 366         raise Exception("Cannot locate site {}".format(sitename))
 367
 368     def locate_node(self, nodename):
 369         for site in self.plc_spec['sites']:
 370             for node in site['nodes']:
 371                 if node['name'] == nodename:
 372                     return site, node
 373         raise Exception("Cannot locate node {}".format(nodename))
 374
 375     def locate_hostname(self, hostname):
 376         for site in self.plc_spec['sites']:
 377             for node in site['nodes']:
 378                 if node['node_fields']['hostname'] == hostname:
 379                     return(site, node)
 380         raise Exception("Cannot locate hostname {}".format(hostname))
 381
 382     def locate_key(self, key_name):
 383         for key in self.plc_spec['keys']:
 384             if key['key_name'] == key_name:
 385                 return key
 386         raise Exception("Cannot locate key {}".format(key_name))
 387
 388     def locate_private_key_from_key_names(self, key_names):
 389         # locate the first avail. key
 390         found = False
 391         for key_name in key_names:
 392             key_spec = self.locate_key(key_name)
 393             test_key = TestKey(self,key_spec)
 394             publickey = test_key.publicpath()
 395             privatekey = test_key.privatepath()
 396             if os.path.isfile(publickey) and os.path.isfile(privatekey):
 397                 found = True
 398         if found:
 399             return privatekey
 400         else:
 401             return None
 402
 403     def locate_slice(self, slicename):
 404         for slice in self.plc_spec['slices']:
 405             if slice['slice_fields']['name'] == slicename:
 406                 return slice
 407         raise Exception("Cannot locate slice {}".format(slicename))
 408
 409     def all_sliver_objs(self):
 410         result = []
 411         for slice_spec in self.plc_spec['slices']:
 412             slicename = slice_spec['slice_fields']['name']
 413             for nodename in slice_spec['nodenames']:
 414                 result.append(self.locate_sliver_obj(nodename, slicename))
 415         return result
 416
 417     def locate_sliver_obj(self, nodename, slicename):
 418         site,node = self.locate_node(nodename)
 419         slice = self.locate_slice(slicename)
 420         # build objects
 421         test_site = TestSite(self, site)
 422         test_node = TestNode(self, test_site, node)
 423         # xxx the slice site is assumed to be the node site - mhh - probably harmless
 424         test_slice = TestSlice(self, test_site, slice)
 425         return TestSliver(self, test_node, test_slice)
 426
 427     def locate_first_node(self):
 428         nodename = self.plc_spec['slices'][0]['nodenames'][0]
 429         site,node = self.locate_node(nodename)
 430         test_site = TestSite(self, site)
 431         test_node = TestNode(self, test_site, node)
 432         return test_node
 433
 434     def locate_first_sliver(self):
 435         slice_spec = self.plc_spec['slices'][0]
 436         slicename = slice_spec['slice_fields']['name']
 437         nodename = slice_spec['nodenames'][0]
 438         return self.locate_sliver_obj(nodename,slicename)
 439
 440     # all different hostboxes used in this plc
 441     def get_BoxNodes(self):
 442         # maps on sites and nodes, return [ (host_box,test_node) ]
 443         tuples = []
 444         for site_spec in self.plc_spec['sites']:
 445             test_site = TestSite(self,site_spec)
 446             for node_spec in site_spec['nodes']:
 447                 test_node = TestNode(self, test_site, node_spec)
 448                 if not test_node.is_real():
 449                     tuples.append( (test_node.host_box(),test_node) )
 450         # transform into a dict { 'host_box' -> [ test_node .. ] }
 451         result = {}
 452         for (box,node) in tuples:
 453             if box not in result:
 454                 result[box] = [node]
 455             else:
 456                 result[box].append(node)
 457         return result
 458
 459     # a step for checking this stuff
 460     def show_boxes(self):
 461         'print summary of nodes location'
 462         for box,nodes in self.get_BoxNodes().items():
 463             print(box,":"," + ".join( [ node.name() for node in nodes ] ))
 464         return True
 465
 466     # make this a valid step
 467     def qemu_kill_all(self):
 468         'kill all qemu instances on the qemu boxes involved by this setup'
 469         # this is the brute force version, kill all qemus on that host box
 470         for (box,nodes) in self.get_BoxNodes().items():
 471             # pass the first nodename, as we don't push template-qemu on testboxes
 472             nodedir = nodes[0].nodedir()
 473             TestBoxQemu(box, self.options.buildname).qemu_kill_all(nodedir)
 474         return True
 475
 476     # make this a valid step
 477     def qemu_list_all(self):
 478         'list all qemu instances on the qemu boxes involved by this setup'
 479         for box,nodes in self.get_BoxNodes().items():
 480             # this is the brute force version, kill all qemus on that host box
 481             TestBoxQemu(box, self.options.buildname).qemu_list_all()
 482         return True
 483
 484     # kill only the qemus related to this test
 485     def qemu_list_mine(self):
 486         'list qemu instances for our nodes'
 487         for (box,nodes) in self.get_BoxNodes().items():
 488             # the fine-grain version
 489             for node in nodes:
 490                 node.list_qemu()
 491         return True
 492
 493     # kill only the qemus related to this test
 494     def qemu_clean_mine(self):
 495         'cleanup (rm -rf) qemu instances for our nodes'
 496         for box,nodes in self.get_BoxNodes().items():
 497             # the fine-grain version
 498             for node in nodes:
 499                 node.qemu_clean()
 500         return True
 501
 502     # kill only the right qemus
 503     def qemu_kill_mine(self):
 504         'kill the qemu instances for our nodes'
 505         for box,nodes in self.get_BoxNodes().items():
 506             # the fine-grain version
 507             for node in nodes:
 508                 node.kill_qemu()
 509         return True
 510
 511     #################### display config
 512     def show(self):
 513         "show test configuration after localization"
 514         self.show_pass(1)
 515         self.show_pass(2)
 516         return True
 517
 518     # uggly hack to make sure 'run export' only reports about the 1st plc
 519     # to avoid confusion - also we use 'inri_slice1' in various aliases..
 520     exported_id = 1
 521     def export(self):
 522         "print cut'n paste-able stuff to export env variables to your shell"
 523         # guess local domain from hostname
 524         if TestPlc.exported_id > 1:
 525             print("export GUESTHOSTNAME{:d}={}".format(TestPlc.exported_id, self.plc_spec['vservername']))
 526             return True
 527         TestPlc.exported_id += 1
 528         domain = socket.gethostname().split('.',1)[1]
 529         fqdn   = "{}.{}".format(self.plc_spec['host_box'], domain)
 530         print("export BUILD={}".format(self.options.buildname))
 531         print("export PLCHOSTLXC={}".format(fqdn))
 532         print("export GUESTNAME={}".format(self.vservername))
 533         print("export GUESTHOSTNAME={}.{}".format(self.vplchostname, domain))
 534         # find hostname of first node
 535         hostname, qemubox = self.all_node_infos()[0]
 536         print("export KVMHOST={}.{}".format(qemubox, domain))
 537         print("export NODE={}".format(hostname))
 538         return True
 539
 540     # entry point
 541     always_display_keys=['PLC_WWW_HOST', 'nodes', 'sites']
 542     def show_pass(self, passno):
 543         for (key,val) in self.plc_spec.items():
 544             if not self.options.verbose and key not in TestPlc.always_display_keys:
 545                 continue
 546             if passno == 2:
 547                 if key == 'sites':
 548                     for site in val:
 549                         self.display_site_spec(site)
 550                         for node in site['nodes']:
 551                             self.display_node_spec(node)
 552                 elif key == 'initscripts':
 553                     for initscript in val:
 554                         self.display_initscript_spec(initscript)
 555                 elif key == 'slices':
 556                     for slice in val:
 557                         self.display_slice_spec(slice)
 558                 elif key == 'keys':
 559                     for key in val:
 560                         self.display_key_spec(key)
 561             elif passno == 1:
 562                 if key not in ['sites', 'initscripts', 'slices', 'keys']:
 563                     print('+   ', key, ':', val)
 564
 565     def display_site_spec(self, site):
 566         print('+ ======== site', site['site_fields']['name'])
 567         for k,v in site.items():
 568             if not self.options.verbose and k not in TestPlc.always_display_keys:
 569                 continue
 570             if k == 'nodes':
 571                 if v:
 572                     print('+       ','nodes : ', end=' ')
 573                     for node in v:
 574                         print(node['node_fields']['hostname'],'', end=' ')
 575                     print('')
 576             elif k == 'users':
 577                 if v:
 578                     print('+       users : ', end=' ')
 579                     for user in v:
 580                         print(user['name'],'', end=' ')
 581                     print('')
 582             elif k == 'site_fields':
 583                 print('+       login_base', ':', v['login_base'])
 584             elif k == 'address_fields':
 585                 pass
 586             else:
 587                 print('+       ', end=' ')
 588                 utils.pprint(k, v)
 589
 590     def display_initscript_spec(self, initscript):
 591         print('+ ======== initscript', initscript['initscript_fields']['name'])
 592
 593     def display_key_spec(self, key):
 594         print('+ ======== key', key['key_name'])
 595
 596     def display_slice_spec(self, slice):
 597         print('+ ======== slice', slice['slice_fields']['name'])
 598         for k,v in slice.items():
 599             if k == 'nodenames':
 600                 if v:
 601                     print('+       nodes : ', end=' ')
 602                     for nodename in v:
 603                         print(nodename,'', end=' ')
 604                     print('')
 605             elif k == 'usernames':
 606                 if v:
 607                     print('+       users : ', end=' ')
 608                     for username in v:
 609                         print(username,'', end=' ')
 610                     print('')
 611             elif k == 'slice_fields':
 612                 print('+       fields',':', end=' ')
 613                 print('max_nodes=',v['max_nodes'], end=' ')
 614                 print('')
 615             else:
 616                 print('+       ',k,v)
 617
 618     def display_node_spec(self, node):
 619         print("+           node={} host_box={}".format(node['name'], node['host_box']), end=' ')
 620         print("hostname=", node['node_fields']['hostname'], end=' ')
 621         print("ip=", node['interface_fields']['ip'])
 622         if self.options.verbose:
 623             utils.pprint("node details", node, depth=3)
 624
 625     # another entry point for just showing the boxes involved
 626     def display_mapping(self):
 627         TestPlc.display_mapping_plc(self.plc_spec)
 628         return True
 629
 630     @staticmethod
 631     def display_mapping_plc(plc_spec):
 632         print('+ MyPLC',plc_spec['name'])
 633         # WARNING this would not be right for lxc-based PLC's - should be harmless though
 634         print('+\tvserver address = root@{}:/vservers/{}'.format(plc_spec['host_box'], plc_spec['vservername']))
 635         print('+\tIP = {}/{}'.format(plc_spec['settings']['PLC_API_HOST'], plc_spec['vserverip']))
 636         for site_spec in plc_spec['sites']:
 637             for node_spec in site_spec['nodes']:
 638                 TestPlc.display_mapping_node(node_spec)
 639
 640     @staticmethod
 641     def display_mapping_node(node_spec):
 642         print('+   NODE {}'.format(node_spec['name']))
 643         print('+\tqemu box {}'.format(node_spec['host_box']))
 644         print('+\thostname={}'.format(node_spec['node_fields']['hostname']))
 645
 646     # write a timestamp in /vservers/<>.timestamp
 647     # cannot be inside the vserver, that causes vserver .. build to cough
 648     def plcvm_timestamp(self):
 649         "Create a timestamp to remember creation date for this plc"
 650         now = int(time.time())
 651         # TODO-lxc check this one
 652         # a first approx. is to store the timestamp close to the VM root like vs does
 653         stamp_path = self.vm_timestamp_path()
 654         stamp_dir = os.path.dirname(stamp_path)
 655         utils.system(self.test_ssh.actual_command("mkdir -p {}".format(stamp_dir)))
 656         return utils.system(self.test_ssh.actual_command("echo {:d} > {}".format(now, stamp_path))) == 0
 657
 658     # this is called inconditionnally at the beginning of the test sequence
 659     # just in case this is a rerun, so if the vm is not running it's fine
 660     def plcvm_delete(self):
 661         "vserver delete the test myplc"
 662         stamp_path = self.vm_timestamp_path()
 663         self.run_in_host("rm -f {}".format(stamp_path))
 664         self.run_in_host("virsh -c lxc:// destroy {}".format(self.vservername))
 665         self.run_in_host("virsh -c lxc:// undefine {}".format(self.vservername))
 666         self.run_in_host("rm -fr /vservers/{}".format(self.vservername))
 667         return True
 668
 669     ### install
 670     # historically the build was being fetched by the tests
 671     # now the build pushes itself as a subdir of the tests workdir
 672     # so that the tests do not have to worry about extracting the build (svn, git, or whatever)
 673     def plcvm_create(self):
 674         "vserver creation (no install done)"
 675         # push the local build/ dir to the testplc box
 676         if self.is_local():
 677             # a full path for the local calls
 678             build_dir = os.path.dirname(sys.argv[0])
 679             # sometimes this is empty - set to "." in such a case
 680             if not build_dir:
 681                 build_dir="."
 682             build_dir += "/build"
 683         else:
 684             # use a standard name - will be relative to remote buildname
 685             build_dir = "build"
 686             # remove for safety; do *not* mkdir first, otherwise we end up with build/build/
 687             self.test_ssh.rmdir(build_dir)
 688             self.test_ssh.copy(build_dir, recursive=True)
 689         # the repo url is taken from arch-rpms-url
 690         # with the last step (i386) removed
 691         repo_url = self.options.arch_rpms_url
 692         for level in [ 'arch' ]:
 693             repo_url = os.path.dirname(repo_url)
 694
 695         # invoke initvm (drop support for vs)
 696         script = "lbuild-initvm.sh"
 697         script_options = ""
 698         # pass the vbuild-nightly options to [lv]test-initvm
 699         script_options += " -p {}".format(self.options.personality)
 700         script_options += " -d {}".format(self.options.pldistro)
 701         script_options += " -f {}".format(self.options.fcdistro)
 702         script_options += " -r {}".format(repo_url)
 703         vserver_name = self.vservername
 704         try:
 705             vserver_hostname = socket.gethostbyaddr(self.vserverip)[0]
 706             script_options += " -n {}".format(vserver_hostname)
 707         except:
 708             print("Cannot reverse lookup {}".format(self.vserverip))
 709             print("This is considered fatal, as this might pollute the test results")
 710             return False
 711         create_vserver="{build_dir}/{script} {script_options} {vserver_name}".format(**locals())
 712         return self.run_in_host(create_vserver) == 0
 713
 714     ### install_rpm
 715     def plc_install(self):
 716         """
 717         yum install myplc, noderepo
 718         """
 719
 720         # compute nodefamily
 721         if self.options.personality == "linux32":
 722             arch = "i386"
 723         elif self.options.personality == "linux64":
 724             arch = "x86_64"
 725         else:
 726             raise Exception("Unsupported personality {}".format(self.options.personality))
 727         nodefamily = "{}-{}-{}".format(self.options.pldistro, self.options.fcdistro, arch)
 728
 729         pkgs_list=[]
 730         pkgs_list.append("slicerepo-{}".format(nodefamily))
 731         pkgs_list.append("myplc")
 732         pkgs_list.append("noderepo-{}".format(nodefamily))
 733         pkgs_string=" ".join(pkgs_list)
 734         return self.yum_install(pkgs_list)
 735
 736     def install_syslinux6(self):
 737         """
 738         install syslinux6 from the fedora21 release
 739         """
 740         key = 'http://mirror.onelab.eu/keys/RPM-GPG-KEY-fedora-21-primary'
 741
 742         rpms = [
 743             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-6.03-1.fc21.x86_64.rpm',
 744             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-nonlinux-6.03-1.fc21.noarch.rpm',
 745             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-perl-6.03-1.fc21.x86_64.rpm',
 746         ]
 747         # this can be done several times
 748         self.run_in_guest("rpm --import {key}".format(**locals()))
 749         return self.run_in_guest("yum -y localinstall {}".format(" ".join(rpms))) == 0
 750
 751     def bonding_builds(self):
 752         """
 753         list /etc/yum.repos.d on the myplc side
 754         """
 755         self.run_in_guest("ls /etc/yum.repos.d/*partial.repo")
 756         return True
 757
 758     def bonding_nodes(self):
 759         """
 760         List nodes known to the myplc together with their nodefamiliy
 761         """
 762         print("---------------------------------------- nodes")
 763         for node in self.apiserver.GetNodes(self.auth_root()):
 764             print("{} -> {}".format(node['hostname'],
 765                                     self.apiserver.GetNodeFlavour(self.auth_root(),node['hostname'])['nodefamily']))
 766         print("---------------------------------------- nodes")
 767
 768
 769     ###
 770     def mod_python(self):
 771         """yum install mod_python, useful on f18 and above so as to avoid broken wsgi"""
 772         return self.yum_install( ['mod_python'] )
 773
 774     ###
 775     def plc_configure(self):
 776         "run plc-config-tty"
 777         tmpname = '{}.plc-config-tty'.format(self.name())
 778         with open(tmpname,'w') as fileconf:
 779             for var, value in self.plc_spec['settings'].items():
 780                 fileconf.write('e {}\n{}\n'.format(var, value))
 781             fileconf.write('w\n')
 782             fileconf.write('q\n')
 783         utils.system('cat {}'.format(tmpname))
 784         self.run_in_guest_piped('cat {}'.format(tmpname), 'plc-config-tty')
 785         utils.system('rm {}'.format(tmpname))
 786         return True
 787
 788 # f14 is a bit odd in this respect, although this worked fine in guests up to f18
 789 # however using a vplc guest under f20 requires this trick
 790 # the symptom is this: service plc start
 791 # Starting plc (via systemctl):  Failed to get D-Bus connection: \
 792 #    Failed to connect to socket /org/freedesktop/systemd1/private: Connection refused
 793 # weird thing is the doc says f14 uses upstart by default and not systemd
 794 # so this sounds kind of harmless
 795     def start_service(self, service):
 796         return self.start_stop_service(service, 'start')
 797     def stop_service(self, service):
 798         return self.start_stop_service(service, 'stop')
 799
 800     def start_stop_service(self, service, start_or_stop):
 801         "utility to start/stop a service with the special trick for f14"
 802         if self.options.fcdistro != 'f14':
 803             return self.run_in_guest("service {} {}".format(service, start_or_stop)) == 0
 804         else:
 805             # patch /sbin/service so it does not reset environment
 806             self.run_in_guest('sed -i -e \\"s,env -i,env,\\" /sbin/service')
 807             # this is because our own scripts in turn call service
 808             return self.run_in_guest("SYSTEMCTL_SKIP_REDIRECT=true service {} {}"\
 809                                      .format(service, start_or_stop)) == 0
 810
 811     def plc_start(self):
 812         "service plc start"
 813         return self.start_service('plc')
 814
 815     def plc_stop(self):
 816         "service plc stop"
 817         return self.stop_service('plc')
 818
 819     def plcvm_start(self):
 820         "start the PLC vserver"
 821         self.start_guest()
 822         return True
 823
 824     def plcvm_stop(self):
 825         "stop the PLC vserver"
 826         self.stop_guest()
 827         return True
 828
 829     # stores the keys from the config for further use
 830     def keys_store(self):
 831         "stores test users ssh keys in keys/"
 832         for key_spec in self.plc_spec['keys']:
 833                 TestKey(self,key_spec).store_key()
 834         return True
 835
 836     def keys_clean(self):
 837         "removes keys cached in keys/"
 838         utils.system("rm -rf ./keys")
 839         return True
 840
 841     # fetches the ssh keys in the plc's /etc/planetlab and stores them in keys/
 842     # for later direct access to the nodes
 843     def keys_fetch(self):
 844         "gets ssh keys in /etc/planetlab/ and stores them locally in keys/"
 845         dir="./keys"
 846         if not os.path.isdir(dir):
 847             os.mkdir(dir)
 848         vservername = self.vservername
 849         vm_root = self.vm_root_in_host()
 850         overall = True
 851         prefix = 'debug_ssh_key'
 852         for ext in ['pub', 'rsa'] :
 853             src = "{vm_root}/etc/planetlab/{prefix}.{ext}".format(**locals())
 854             dst = "keys/{vservername}-debug.{ext}".format(**locals())
 855             if self.test_ssh.fetch(src, dst) != 0:
 856                 overall=False
 857         return overall
 858
 859     def sites(self):
 860         "create sites with PLCAPI"
 861         return self.do_sites()
 862
 863     def delete_sites(self):
 864         "delete sites with PLCAPI"
 865         return self.do_sites(action="delete")
 866
 867     def do_sites(self, action="add"):
 868         for site_spec in self.plc_spec['sites']:
 869             test_site = TestSite(self,site_spec)
 870             if (action != "add"):
 871                 utils.header("Deleting site {} in {}".format(test_site.name(), self.name()))
 872                 test_site.delete_site()
 873                 # deleted with the site
 874                 #test_site.delete_users()
 875                 continue
 876             else:
 877                 utils.header("Creating site {} & users in {}".format(test_site.name(), self.name()))
 878                 test_site.create_site()
 879                 test_site.create_users()
 880         return True
 881
 882     def delete_all_sites(self):
 883         "Delete all sites in PLC, and related objects"
 884         print('auth_root', self.auth_root())
 885         sites = self.apiserver.GetSites(self.auth_root(), {}, ['site_id','login_base'])
 886         for site in sites:
 887             # keep automatic site - otherwise we shoot in our own foot, root_auth is not valid anymore
 888             if site['login_base'] == self.plc_spec['settings']['PLC_SLICE_PREFIX']:
 889                 continue
 890             site_id = site['site_id']
 891             print('Deleting site_id', site_id)
 892             self.apiserver.DeleteSite(self.auth_root(), site_id)
 893         return True
 894
 895     def nodes(self):
 896         "create nodes with PLCAPI"
 897         return self.do_nodes()
 898     def delete_nodes(self):
 899         "delete nodes with PLCAPI"
 900         return self.do_nodes(action="delete")
 901
 902     def do_nodes(self, action="add"):
 903         for site_spec in self.plc_spec['sites']:
 904             test_site = TestSite(self, site_spec)
 905             if action != "add":
 906                 utils.header("Deleting nodes in site {}".format(test_site.name()))
 907                 for node_spec in site_spec['nodes']:
 908                     test_node = TestNode(self, test_site, node_spec)
 909                     utils.header("Deleting {}".format(test_node.name()))
 910                     test_node.delete_node()
 911             else:
 912                 utils.header("Creating nodes for site {} in {}".format(test_site.name(), self.name()))
 913                 for node_spec in site_spec['nodes']:
 914                     utils.pprint('Creating node {}'.format(node_spec), node_spec)
 915                     test_node = TestNode(self, test_site, node_spec)
 916                     test_node.create_node()
 917         return True
 918
 919     def nodegroups(self):
 920         "create nodegroups with PLCAPI"
 921         return self.do_nodegroups("add")
 922     def delete_nodegroups(self):
 923         "delete nodegroups with PLCAPI"
 924         return self.do_nodegroups("delete")
 925
 926     YEAR = 365*24*3600
 927     @staticmethod
 928     def translate_timestamp(start, grain, timestamp):
 929         if timestamp < TestPlc.YEAR:
 930             return start + timestamp*grain
 931         else:
 932             return timestamp
 933
 934     @staticmethod
 935     def timestamp_printable(timestamp):
 936         return time.strftime('%m-%d %H:%M:%S UTC', time.gmtime(timestamp))
 937
 938     def leases(self):
 939         "create leases (on reservable nodes only, use e.g. run -c default -c resa)"
 940         now = int(time.time())
 941         grain = self.apiserver.GetLeaseGranularity(self.auth_root())
 942         print('API answered grain=', grain)
 943         start = (now//grain)*grain
 944         start += grain
 945         # find out all nodes that are reservable
 946         nodes = self.all_reservable_nodenames()
 947         if not nodes:
 948             utils.header("No reservable node found - proceeding without leases")
 949             return True
 950         ok = True
 951         # attach them to the leases as specified in plc_specs
 952         # this is where the 'leases' field gets interpreted as relative of absolute
 953         for lease_spec in self.plc_spec['leases']:
 954             # skip the ones that come with a null slice id
 955             if not lease_spec['slice']:
 956                 continue
 957             lease_spec['t_from']  = TestPlc.translate_timestamp(start, grain, lease_spec['t_from'])
 958             lease_spec['t_until'] = TestPlc.translate_timestamp(start, grain, lease_spec['t_until'])
 959             lease_addition = self.apiserver.AddLeases(self.auth_root(), nodes, lease_spec['slice'],
 960                                                       lease_spec['t_from'], lease_spec['t_until'])
 961             if lease_addition['errors']:
 962                 utils.header("Cannot create leases, {}".format(lease_addition['errors']))
 963                 ok = False
 964             else:
 965                 utils.header('Leases on nodes {} for {} from {:d} ({}) until {:d} ({})'\
 966                              .format(nodes, lease_spec['slice'],
 967                                      lease_spec['t_from'],  TestPlc.timestamp_printable(lease_spec['t_from']),
 968                                      lease_spec['t_until'], TestPlc.timestamp_printable(lease_spec['t_until'])))
 969
 970         return ok
 971
 972     def delete_leases(self):
 973         "remove all leases in the myplc side"
 974         lease_ids = [ l['lease_id'] for l in self.apiserver.GetLeases(self.auth_root())]
 975         utils.header("Cleaning leases {}".format(lease_ids))
 976         self.apiserver.DeleteLeases(self.auth_root(), lease_ids)
 977         return True
 978
 979     def list_leases(self):
 980         "list all leases known to the myplc"
 981         leases = self.apiserver.GetLeases(self.auth_root())
 982         now = int(time.time())
 983         for l in leases:
 984             current = l['t_until'] >= now
 985             if self.options.verbose or current:
 986                 utils.header("{} {} from {} until {}"\
 987                              .format(l['hostname'], l['name'],
 988                                      TestPlc.timestamp_printable(l['t_from']),
 989                                      TestPlc.timestamp_printable(l['t_until'])))
 990         return True
 991
 992     # create nodegroups if needed, and populate
 993     def do_nodegroups(self, action="add"):
 994         # 1st pass to scan contents
 995         groups_dict = {}
 996         for site_spec in self.plc_spec['sites']:
 997             test_site = TestSite(self,site_spec)
 998             for node_spec in site_spec['nodes']:
 999                 test_node = TestNode(self, test_site, node_spec)
1000                 if 'nodegroups' in node_spec:
1001                     nodegroupnames = node_spec['nodegroups']
1002                     if isinstance(nodegroupnames, str):
1003                         nodegroupnames = [ nodegroupnames ]
1004                     for nodegroupname in nodegroupnames:
1005                         if nodegroupname not in groups_dict:
1006                             groups_dict[nodegroupname] = []
1007                         groups_dict[nodegroupname].append(test_node.name())
1008         auth = self.auth_root()
1009         overall = True
1010         for (nodegroupname,group_nodes) in groups_dict.items():
1011             if action == "add":
1012                 print('nodegroups:', 'dealing with nodegroup',\
1013                     nodegroupname, 'on nodes', group_nodes)
1014                 # first, check if the nodetagtype is here
1015                 tag_types = self.apiserver.GetTagTypes(auth, {'tagname':nodegroupname})
1016                 if tag_types:
1017                     tag_type_id = tag_types[0]['tag_type_id']
1018                 else:
1019                     tag_type_id = self.apiserver.AddTagType(auth,
1020                                                             {'tagname' : nodegroupname,
1021                                                              'description' : 'for nodegroup {}'.format(nodegroupname),
1022                                                              'category' : 'test'})
1023                 print('located tag (type)', nodegroupname, 'as', tag_type_id)
1024                 # create nodegroup
1025                 nodegroups = self.apiserver.GetNodeGroups(auth, {'groupname' : nodegroupname})
1026                 if not nodegroups:
1027                     self.apiserver.AddNodeGroup(auth, nodegroupname, tag_type_id, 'yes')
1028                     print('created nodegroup', nodegroupname, \
1029                         'from tagname', nodegroupname, 'and value', 'yes')
1030                 # set node tag on all nodes, value='yes'
1031                 for nodename in group_nodes:
1032                     try:
1033                         self.apiserver.AddNodeTag(auth, nodename, nodegroupname, "yes")
1034                     except:
1035                         traceback.print_exc()
1036                         print('node', nodename, 'seems to already have tag', nodegroupname)
1037                     # check anyway
1038                     try:
1039                         expect_yes = self.apiserver.GetNodeTags(auth,
1040                                                                 {'hostname' : nodename,
1041                                                                  'tagname'  : nodegroupname},
1042                                                                 ['value'])[0]['value']
1043                         if expect_yes != "yes":
1044                             print('Mismatch node tag on node',nodename,'got',expect_yes)
1045                             overall = False
1046                     except:
1047                         if not self.options.dry_run:
1048                             print('Cannot find tag', nodegroupname, 'on node', nodename)
1049                             overall = False
1050             else:
1051                 try:
1052                     print('cleaning nodegroup', nodegroupname)
1053                     self.apiserver.DeleteNodeGroup(auth, nodegroupname)
1054                 except:
1055                     traceback.print_exc()
1056                     overall = False
1057         return overall
1058
1059     # a list of TestNode objs
1060     def all_nodes(self):
1061         nodes=[]
1062         for site_spec in self.plc_spec['sites']:
1063             test_site = TestSite(self,site_spec)
1064             for node_spec in site_spec['nodes']:
1065                 nodes.append(TestNode(self, test_site, node_spec))
1066         return nodes
1067
1068     # return a list of tuples (nodename,qemuname)
1069     def all_node_infos(self) :
1070         node_infos = []
1071         for site_spec in self.plc_spec['sites']:
1072             node_infos += [ (node_spec['node_fields']['hostname'], node_spec['host_box']) \
1073                                 for node_spec in site_spec['nodes'] ]
1074         return node_infos
1075
1076     def all_nodenames(self):
1077         return [ x[0] for x in self.all_node_infos() ]
1078     def all_reservable_nodenames(self):
1079         res = []
1080         for site_spec in self.plc_spec['sites']:
1081             for node_spec in site_spec['nodes']:
1082                 node_fields = node_spec['node_fields']
1083                 if 'node_type' in node_fields and node_fields['node_type'] == 'reservable':
1084                     res.append(node_fields['hostname'])
1085         return res
1086
1087     # silent_minutes : during the first <silent_minutes> minutes nothing gets printed
1088     def nodes_check_boot_state(self, target_boot_state, timeout_minutes,
1089                                silent_minutes, period_seconds = 15):
1090         if self.options.dry_run:
1091             print('dry_run')
1092             return True
1093
1094         class CompleterTaskBootState(CompleterTask):
1095             def __init__(self, test_plc, hostname):
1096                 self.test_plc = test_plc
1097                 self.hostname = hostname
1098                 self.last_boot_state = 'undef'
1099             def actual_run(self):
1100                 try:
1101                     node = self.test_plc.apiserver.GetNodes(self.test_plc.auth_root(),
1102                                                             [ self.hostname ],
1103                                                             ['boot_state'])[0]
1104                     self.last_boot_state = node['boot_state']
1105                     return self.last_boot_state == target_boot_state
1106                 except:
1107                     return False
1108             def message(self):
1109                 return "CompleterTaskBootState with node {}".format(self.hostname)
1110             def failure_epilogue(self):
1111                 print("node {} in state {} - expected {}"\
1112                     .format(self.hostname, self.last_boot_state, target_boot_state))
1113
1114         timeout = timedelta(minutes=timeout_minutes)
1115         graceout = timedelta(minutes=silent_minutes)
1116         period   = timedelta(seconds=period_seconds)
1117         # the nodes that haven't checked yet - start with a full list and shrink over time
1118         utils.header("checking nodes boot state (expected {})".format(target_boot_state))
1119         tasks = [ CompleterTaskBootState(self,hostname) \
1120                       for (hostname,_) in self.all_node_infos() ]
1121         message = 'check_boot_state={}'.format(target_boot_state)
1122         return Completer(tasks, message=message).run(timeout, graceout, period)
1123
1124     def nodes_booted(self):
1125         return self.nodes_check_boot_state('boot', timeout_minutes=30, silent_minutes=28)
1126
1127     def probe_kvm_iptables(self):
1128         (_,kvmbox) = self.all_node_infos()[0]
1129         TestSsh(kvmbox).run("iptables-save")
1130         return True
1131
1132     # probing nodes
1133     def check_nodes_ping(self, timeout_seconds=60, period_seconds=10):
1134         class CompleterTaskPingNode(CompleterTask):
1135             def __init__(self, hostname):
1136                 self.hostname = hostname
1137             def run(self, silent):
1138                 command="ping -c 1 -w 1 {} >& /dev/null".format(self.hostname)
1139                 return utils.system(command, silent=silent) == 0
1140             def failure_epilogue(self):
1141                 print("Cannot ping node with name {}".format(self.hostname))
1142         timeout = timedelta(seconds = timeout_seconds)
1143         graceout = timeout
1144         period = timedelta(seconds = period_seconds)
1145         node_infos = self.all_node_infos()
1146         tasks = [ CompleterTaskPingNode(h) for (h,_) in node_infos ]
1147         return Completer(tasks, message='ping_node').run(timeout, graceout, period)
1148
1149     # ping node before we try to reach ssh, helpful for troubleshooting failing bootCDs
1150     def ping_node(self):
1151         "Ping nodes"
1152         return self.check_nodes_ping()
1153
1154     def check_nodes_ssh(self, debug, timeout_minutes, silent_minutes, period_seconds=15):
1155         # various delays
1156         timeout  = timedelta(minutes=timeout_minutes)
1157         graceout = timedelta(minutes=silent_minutes)
1158         period   = timedelta(seconds=period_seconds)
1159         vservername = self.vservername
1160         if debug:
1161             message = "debug"
1162             completer_message = 'ssh_node_debug'
1163             local_key = "keys/{vservername}-debug.rsa".format(**locals())
1164         else:
1165             message = "boot"
1166             completer_message = 'ssh_node_boot'
1167             local_key = "keys/key_admin.rsa"
1168         utils.header("checking ssh access to nodes (expected in {} mode)".format(message))
1169         node_infos = self.all_node_infos()
1170         tasks = [ CompleterTaskNodeSsh(nodename, qemuname, local_key,
1171                                         boot_state=message, dry_run=self.options.dry_run) \
1172                       for (nodename, qemuname) in node_infos ]
1173         return Completer(tasks, message=completer_message).run(timeout, graceout, period)
1174
1175     def ssh_node_debug(self):
1176         "Tries to ssh into nodes in debug mode with the debug ssh key"
1177         return self.check_nodes_ssh(debug = True,
1178                                     timeout_minutes = self.ssh_node_debug_timeout,
1179                                     silent_minutes = self.ssh_node_debug_silent)
1180
1181     def ssh_node_boot(self):
1182         "Tries to ssh into nodes in production mode with the root ssh key"
1183         return self.check_nodes_ssh(debug = False,
1184                                     timeout_minutes = self.ssh_node_boot_timeout,
1185                                     silent_minutes = self.ssh_node_boot_silent)
1186
1187     def node_bmlogs(self):
1188         "Checks that there's a non-empty dir. /var/log/bm/raw"
1189         return utils.system(self.actual_command_in_guest("ls /var/log/bm/raw")) == 0
1190
1191     @node_mapper
1192     def qemu_local_init(self): pass
1193     @node_mapper
1194     def bootcd(self): pass
1195     @node_mapper
1196     def qemu_local_config(self): pass
1197     @node_mapper
1198     def qemu_export(self): pass
1199     @node_mapper
1200     def qemu_cleanlog(self): pass
1201     @node_mapper
1202     def nodestate_reinstall(self): pass
1203     @node_mapper
1204     def nodestate_upgrade(self): pass
1205     @node_mapper
1206     def nodestate_safeboot(self): pass
1207     @node_mapper
1208     def nodestate_boot(self): pass
1209     @node_mapper
1210     def nodestate_show(self): pass
1211     @node_mapper
1212     def nodedistro_f14(self): pass
1213     @node_mapper
1214     def nodedistro_f18(self): pass
1215     @node_mapper
1216     def nodedistro_f20(self): pass
1217     @node_mapper
1218     def nodedistro_f21(self): pass
1219     @node_mapper
1220     def nodedistro_f22(self): pass
1221     @node_mapper
1222     def nodedistro_show(self): pass
1223
1224     ### check hooks : invoke scripts from hooks/{node,slice}
1225     def check_hooks_node(self):
1226         return self.locate_first_node().check_hooks()
1227     def check_hooks_sliver(self) :
1228         return self.locate_first_sliver().check_hooks()
1229
1230     def check_hooks(self):
1231         "runs unit tests in the node and slice contexts - see hooks/{node,slice}"
1232         return self.check_hooks_node() and self.check_hooks_sliver()
1233
1234     ### initscripts
1235     def do_check_initscripts(self):
1236         class CompleterTaskInitscript(CompleterTask):
1237             def __init__(self, test_sliver, stamp):
1238                 self.test_sliver = test_sliver
1239                 self.stamp = stamp
1240             def actual_run(self):
1241                 return self.test_sliver.check_initscript_stamp(self.stamp)
1242             def message(self):
1243                 return "initscript checker for {}".format(self.test_sliver.name())
1244             def failure_epilogue(self):
1245                 print("initscript stamp {} not found in sliver {}"\
1246                     .format(self.stamp, self.test_sliver.name()))
1247
1248         tasks = []
1249         for slice_spec in self.plc_spec['slices']:
1250             if 'initscriptstamp' not in slice_spec:
1251                 continue
1252             stamp = slice_spec['initscriptstamp']
1253             slicename = slice_spec['slice_fields']['name']
1254             for nodename in slice_spec['nodenames']:
1255                 print('nodename', nodename, 'slicename', slicename, 'stamp', stamp)
1256                 site,node = self.locate_node(nodename)
1257                 # xxx - passing the wrong site - probably harmless
1258                 test_site = TestSite(self, site)
1259                 test_slice = TestSlice(self, test_site, slice_spec)
1260                 test_node = TestNode(self, test_site, node)
1261                 test_sliver = TestSliver(self, test_node, test_slice)
1262                 tasks.append(CompleterTaskInitscript(test_sliver, stamp))
1263         return Completer(tasks, message='check_initscripts').\
1264             run (timedelta(minutes=5), timedelta(minutes=4), timedelta(seconds=10))
1265
1266     def check_initscripts(self):
1267         "check that the initscripts have triggered"
1268         return self.do_check_initscripts()
1269
1270     def initscripts(self):
1271         "create initscripts with PLCAPI"
1272         for initscript in self.plc_spec['initscripts']:
1273             utils.pprint('Adding Initscript in plc {}'.format(self.plc_spec['name']), initscript)
1274             self.apiserver.AddInitScript(self.auth_root(), initscript['initscript_fields'])
1275         return True
1276
1277     def delete_initscripts(self):
1278         "delete initscripts with PLCAPI"
1279         for initscript in self.plc_spec['initscripts']:
1280             initscript_name = initscript['initscript_fields']['name']
1281             print(('Attempting to delete {} in plc {}'.format(initscript_name, self.plc_spec['name'])))
1282             try:
1283                 self.apiserver.DeleteInitScript(self.auth_root(), initscript_name)
1284                 print(initscript_name, 'deleted')
1285             except:
1286                 print('deletion went wrong - probably did not exist')
1287         return True
1288
1289     ### manage slices
1290     def slices(self):
1291         "create slices with PLCAPI"
1292         return self.do_slices(action="add")
1293
1294     def delete_slices(self):
1295         "delete slices with PLCAPI"
1296         return self.do_slices(action="delete")
1297
1298     def fill_slices(self):
1299         "add nodes in slices with PLCAPI"
1300         return self.do_slices(action="fill")
1301
1302     def empty_slices(self):
1303         "remove nodes from slices with PLCAPI"
1304         return self.do_slices(action="empty")
1305
1306     def do_slices(self,  action="add"):
1307         for slice in self.plc_spec['slices']:
1308             site_spec = self.locate_site(slice['sitename'])
1309             test_site = TestSite(self,site_spec)
1310             test_slice=TestSlice(self,test_site,slice)
1311             if action == "delete":
1312                 test_slice.delete_slice()
1313             elif action == "fill":
1314                 test_slice.add_nodes()
1315             elif action == "empty":
1316                 test_slice.delete_nodes()
1317             else:
1318                 test_slice.create_slice()
1319         return True
1320
1321     @slice_mapper__tasks(20, 10, 15)
1322     def ssh_slice(self): pass
1323     @slice_mapper__tasks(20, 19, 15)
1324     def ssh_slice_off(self): pass
1325     @slice_mapper__tasks(1, 1, 15)
1326     def slice_fs_present(self): pass
1327     @slice_mapper__tasks(1, 1, 15)
1328     def slice_fs_deleted(self): pass
1329
1330     # use another name so we can exclude/ignore it from the tests on the nightly command line
1331     def ssh_slice_again(self): return self.ssh_slice()
1332     # note that simply doing ssh_slice_again=ssh_slice would kind of work too
1333     # but for some reason the ignore-wrapping thing would not
1334
1335     @slice_mapper
1336     def ssh_slice_basics(self): pass
1337     @slice_mapper
1338     def check_vsys_defaults(self): pass
1339
1340     @node_mapper
1341     def keys_clear_known_hosts(self): pass
1342
1343     def plcapi_urls(self):
1344         """
1345         attempts to reach the PLCAPI with various forms for the URL
1346         """
1347         return PlcapiUrlScanner(self.auth_root(), ip=self.vserverip).scan()
1348
1349     def speed_up_slices(self):
1350         "tweak nodemanager cycle (wait time) to 30+/-10 s"
1351         return self._speed_up_slices (30, 10)
1352     def super_speed_up_slices(self):
1353         "dev mode: tweak nodemanager cycle (wait time) to 5+/-1 s"
1354         return self._speed_up_slices(5, 1)
1355
1356     def _speed_up_slices(self, p, r):
1357         # create the template on the server-side
1358         template = "{}.nodemanager".format(self.name())
1359         with open(template,"w") as template_file:
1360             template_file.write('OPTIONS="-p {} -r {} -d"\n'.format(p, r))
1361         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1362         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1363         self.test_ssh.copy_abs(template, remote)
1364         # Add a conf file
1365         if not self.apiserver.GetConfFiles(self.auth_root(),
1366                                            {'dest' : '/etc/sysconfig/nodemanager'}):
1367             self.apiserver.AddConfFile(self.auth_root(),
1368                                         {'dest' : '/etc/sysconfig/nodemanager',
1369                                          'source' : 'PlanetLabConf/nodemanager',
1370                                          'postinstall_cmd' : 'service nm restart',})
1371         return True
1372
1373     def debug_nodemanager(self):
1374         "sets verbose mode for nodemanager, and speeds up cycle even more (needs speed_up_slices first)"
1375         template = "{}.nodemanager".format(self.name())
1376         with open(template,"w") as template_file:
1377             template_file.write('OPTIONS="-p 10 -r 6 -v -d"\n')
1378         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1379         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1380         self.test_ssh.copy_abs(template, remote)
1381         return True
1382
1383     @node_mapper
1384     def qemu_start(self) : pass
1385
1386     @node_mapper
1387     def qemu_timestamp(self) : pass
1388
1389     @node_mapper
1390     def qemu_nodefamily(self): pass
1391
1392     # when a spec refers to a node possibly on another plc
1393     def locate_sliver_obj_cross(self, nodename, slicename, other_plcs):
1394         for plc in [ self ] + other_plcs:
1395             try:
1396                 return plc.locate_sliver_obj(nodename, slicename)
1397             except:
1398                 pass
1399         raise Exception("Cannot locate sliver {}@{} among all PLCs".format(nodename, slicename))
1400
1401     # implement this one as a cross step so that we can take advantage of different nodes
1402     # in multi-plcs mode
1403     def cross_check_tcp(self, other_plcs):
1404         "check TCP connectivity between 2 slices (or in loopback if only one is defined)"
1405         if 'tcp_specs' not in self.plc_spec or not self.plc_spec['tcp_specs']:
1406             utils.header("check_tcp: no/empty config found")
1407             return True
1408         specs = self.plc_spec['tcp_specs']
1409         overall = True
1410
1411         # first wait for the network to be up and ready from the slices
1412         class CompleterTaskNetworkReadyInSliver(CompleterTask):
1413             def __init__(self, test_sliver):
1414                 self.test_sliver = test_sliver
1415             def actual_run(self):
1416                 return self.test_sliver.check_tcp_ready(port = 9999)
1417             def message(self):
1418                 return "network ready checker for {}".format(self.test_sliver.name())
1419             def failure_epilogue(self):
1420                 print("could not bind port from sliver {}".format(self.test_sliver.name()))
1421
1422         sliver_specs = {}
1423         tasks = []
1424         managed_sliver_names = set()
1425         for spec in specs:
1426             # locate the TestSliver instances involved, and cache them in the spec instance
1427             spec['s_sliver'] = self.locate_sliver_obj_cross(spec['server_node'], spec['server_slice'], other_plcs)
1428             spec['c_sliver'] = self.locate_sliver_obj_cross(spec['client_node'], spec['client_slice'], other_plcs)
1429             message = "Will check TCP between s={} and c={}"\
1430                       .format(spec['s_sliver'].name(), spec['c_sliver'].name())
1431             if 'client_connect' in spec:
1432                 message += " (using {})".format(spec['client_connect'])
1433             utils.header(message)
1434             # we need to check network presence in both slivers, but also
1435             # avoid to insert a sliver several times
1436             for sliver in [ spec['s_sliver'], spec['c_sliver'] ]:
1437                 if sliver.name() not in managed_sliver_names:
1438                     tasks.append(CompleterTaskNetworkReadyInSliver(sliver))
1439                     # add this sliver's name in the set
1440                     managed_sliver_names .update( {sliver.name()} )
1441
1442         # wait for the netork to be OK in all server sides
1443         if not Completer(tasks, message='check for network readiness in slivers').\
1444            run(timedelta(seconds=30), timedelta(seconds=24), period=timedelta(seconds=5)):
1445             return False
1446
1447         # run server and client
1448         for spec in specs:
1449             port = spec['port']
1450             # server side
1451             # the issue here is that we have the server run in background
1452             # and so we have no clue if it took off properly or not
1453             # looks like in some cases it does not
1454             if not spec['s_sliver'].run_tcp_server(port, timeout=20):
1455                 overall = False
1456                 break
1457
1458             # idem for the client side
1459             # use nodename from located sliver, unless 'client_connect' is set
1460             if 'client_connect' in spec:
1461                 destination = spec['client_connect']
1462             else:
1463                 destination = spec['s_sliver'].test_node.name()
1464             if not spec['c_sliver'].run_tcp_client(destination, port):
1465                 overall = False
1466         return overall
1467
1468     # painfully enough, we need to allow for some time as netflow might show up last
1469     def check_system_slice(self):
1470         "all nodes: check that a system slice is alive"
1471         # netflow currently not working in the lxc distro
1472         # drl not built at all in the wtx distro
1473         # if we find either of them we're happy
1474         return self.check_netflow() or self.check_drl()
1475
1476     # expose these
1477     def check_netflow(self): return self._check_system_slice('netflow')
1478     def check_drl(self): return self._check_system_slice('drl')
1479
1480     # we have the slices up already here, so it should not take too long
1481     def _check_system_slice(self, slicename, timeout_minutes=5, period_seconds=15):
1482         class CompleterTaskSystemSlice(CompleterTask):
1483             def __init__(self, test_node, dry_run):
1484                 self.test_node = test_node
1485                 self.dry_run = dry_run
1486             def actual_run(self):
1487                 return self.test_node._check_system_slice(slicename, dry_run=self.dry_run)
1488             def message(self):
1489                 return "System slice {} @ {}".format(slicename, self.test_node.name())
1490             def failure_epilogue(self):
1491                 print("COULD not find system slice {} @ {}".format(slicename, self.test_node.name()))
1492         timeout = timedelta(minutes=timeout_minutes)
1493         silent  = timedelta(0)
1494         period  = timedelta(seconds=period_seconds)
1495         tasks = [ CompleterTaskSystemSlice(test_node, self.options.dry_run) \
1496                       for test_node in self.all_nodes() ]
1497         return Completer(tasks, message='_check_system_slice').run(timeout, silent, period)
1498
1499     def plcsh_stress_test(self):
1500         "runs PLCAPI stress test, that checks Add/Update/Delete on all types - preserves contents"
1501         # install the stress-test in the plc image
1502         location = "/usr/share/plc_api/plcsh_stress_test.py"
1503         remote = "{}/{}".format(self.vm_root_in_host(), location)
1504         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1505         command = location
1506         command += " -- --check"
1507         if self.options.size == 1:
1508             command +=  " --tiny"
1509         return self.run_in_guest(command) == 0
1510
1511     # populate runs the same utility without slightly different options
1512     # in particular runs with --preserve (dont cleanup) and without --check
1513     # also it gets run twice, once with the --foreign option for creating fake foreign entries
1514
1515     def sfa_install_all(self):
1516         "yum install sfa sfa-plc sfa-sfatables sfa-client"
1517         return self.yum_install("sfa sfa-plc sfa-sfatables sfa-client")
1518
1519     def sfa_install_core(self):
1520         "yum install sfa"
1521         return self.yum_install("sfa")
1522
1523     def sfa_install_plc(self):
1524         "yum install sfa-plc"
1525         return self.yum_install("sfa-plc")
1526
1527     def sfa_install_sfatables(self):
1528         "yum install sfa-sfatables"
1529         return self.yum_install("sfa-sfatables")
1530
1531     # for some very odd reason, this sometimes fails with the following symptom
1532     # # yum install sfa-client
1533     # Setting up Install Process
1534     # ...
1535     # Downloading Packages:
1536     # Running rpm_check_debug
1537     # Running Transaction Test
1538     # Transaction Test Succeeded
1539     # Running Transaction
1540     # Transaction couldn't start:
1541     # installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem
1542     # [('installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem', (9, '/', 69632L))]
1543     # even though in the same context I have
1544     # [2012.05.23--f14-32-sfastd1-1-vplc07] / # df -h
1545     # Filesystem            Size  Used Avail Use% Mounted on
1546     # /dev/hdv1             806G  264G  501G  35% /
1547     # none                   16M   36K   16M   1% /tmp
1548     #
1549     # so as a workaround, we first try yum install, and then invoke rpm on the cached rpm...
1550     def sfa_install_client(self):
1551         "yum install sfa-client"
1552         first_try = self.yum_install("sfa-client")
1553         if first_try:
1554             return True
1555         utils.header("********** Regular yum failed - special workaround in place, 2nd chance")
1556         code, cached_rpm_path = \
1557                 utils.output_of(self.actual_command_in_guest('find /var/cache/yum -name sfa-client\*.rpm'))
1558         utils.header("rpm_path=<<{}>>".format(rpm_path))
1559         # just for checking
1560         self.run_in_guest("rpm -i {}".format(cached_rpm_path))
1561         return self.yum_check_installed("sfa-client")
1562
1563     def sfa_dbclean(self):
1564         "thoroughly wipes off the SFA database"
1565         return self.run_in_guest("sfaadmin reg nuke") == 0 or \
1566             self.run_in_guest("sfa-nuke.py") == 0 or \
1567             self.run_in_guest("sfa-nuke-plc.py") == 0 or \
1568             self.run_in_guest("sfaadmin registry nuke") == 0
1569
1570     def sfa_fsclean(self):
1571         "cleanup /etc/sfa/trusted_roots and /var/lib/sfa"
1572         self.run_in_guest("rm -rf /etc/sfa/trusted_roots /var/lib/sfa/authorities")
1573         return True
1574
1575     def sfa_plcclean(self):
1576         "cleans the PLC entries that were created as a side effect of running the script"
1577         # ignore result
1578         sfa_spec = self.plc_spec['sfa']
1579
1580         for auth_sfa_spec in sfa_spec['auth_sfa_specs']:
1581             login_base = auth_sfa_spec['login_base']
1582             try:
1583                 self.apiserver.DeleteSite(self.auth_root(),login_base)
1584             except:
1585                 print("Site {} already absent from PLC db".format(login_base))
1586
1587             for spec_name in ['pi_spec','user_spec']:
1588                 user_spec = auth_sfa_spec[spec_name]
1589                 username = user_spec['email']
1590                 try:
1591                     self.apiserver.DeletePerson(self.auth_root(),username)
1592                 except:
1593                     # this in fact is expected as sites delete their members
1594                     #print "User {} already absent from PLC db".format(username)
1595                     pass
1596
1597         print("REMEMBER TO RUN sfa_import AGAIN")
1598         return True
1599
1600     def sfa_uninstall(self):
1601         "uses rpm to uninstall sfa - ignore result"
1602         self.run_in_guest("rpm -e sfa sfa-sfatables sfa-client sfa-plc")
1603         self.run_in_guest("rm -rf /var/lib/sfa")
1604         self.run_in_guest("rm -rf /etc/sfa")
1605         self.run_in_guest("rm -rf /var/log/sfa_access.log /var/log/sfa_import_plc.log /var/log/sfa.daemon")
1606         # xxx tmp
1607         self.run_in_guest("rpm -e --noscripts sfa-plc")
1608         return True
1609
1610     ### run unit tests for SFA
1611     # NOTE: for some reason on f14/i386, yum install sfa-tests fails for no reason
1612     # Running Transaction
1613     # Transaction couldn't start:
1614     # installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem
1615     # [('installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem', (9, '/', 208896L))]
1616     # no matter how many Gbs are available on the testplc
1617     # could not figure out what's wrong, so...
1618     # if the yum install phase fails, consider the test is successful
1619     # other combinations will eventually run it hopefully
1620     def sfa_utest(self):
1621         "yum install sfa-tests and run SFA unittests"
1622         self.run_in_guest("yum -y install sfa-tests")
1623         # failed to install - forget it
1624         if self.run_in_guest("rpm -q sfa-tests") != 0:
1625             utils.header("WARNING: SFA unit tests failed to install, ignoring")
1626             return True
1627         return self.run_in_guest("/usr/share/sfa/tests/testAll.py") == 0
1628
1629     ###
1630     def confdir(self):
1631         dirname = "conf.{}".format(self.plc_spec['name'])
1632         if not os.path.isdir(dirname):
1633             utils.system("mkdir -p {}".format(dirname))
1634         if not os.path.isdir(dirname):
1635             raise Exception("Cannot create config dir for plc {}".format(self.name()))
1636         return dirname
1637
1638     def conffile(self, filename):
1639         return "{}/{}".format(self.confdir(), filename)
1640     def confsubdir(self, dirname, clean, dry_run=False):
1641         subdirname = "{}/{}".format(self.confdir(), dirname)
1642         if clean:
1643             utils.system("rm -rf {}".format(subdirname))
1644         if not os.path.isdir(subdirname):
1645             utils.system("mkdir -p {}".format(subdirname))
1646         if not dry_run and not os.path.isdir(subdirname):
1647             raise "Cannot create config subdir {} for plc {}".format(dirname, self.name())
1648         return subdirname
1649
1650     def conffile_clean(self, filename):
1651         filename=self.conffile(filename)
1652         return utils.system("rm -rf {}".format(filename))==0
1653
1654     ###
1655     def sfa_configure(self):
1656         "run sfa-config-tty"
1657         tmpname = self.conffile("sfa-config-tty")
1658         with open(tmpname,'w') as fileconf:
1659             for var, value in self.plc_spec['sfa']['settings'].items():
1660                 fileconf.write('e {}\n{}\n'.format(var, value))
1661             fileconf.write('w\n')
1662             fileconf.write('R\n')
1663             fileconf.write('q\n')
1664         utils.system('cat {}'.format(tmpname))
1665         self.run_in_guest_piped('cat {}'.format(tmpname), 'sfa-config-tty')
1666         return True
1667
1668     def aggregate_xml_line(self):
1669         port = self.plc_spec['sfa']['neighbours-port']
1670         return '<aggregate addr="{}" hrn="{}" port="{}"/>'\
1671             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'], port)
1672
1673     def registry_xml_line(self):
1674         return '<registry addr="{}" hrn="{}" port="12345"/>'\
1675             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'])
1676
1677
1678     # a cross step that takes all other plcs in argument
1679     def cross_sfa_configure(self, other_plcs):
1680         "writes aggregates.xml and registries.xml that point to all other PLCs in the test"
1681         # of course with a single plc, other_plcs is an empty list
1682         if not other_plcs:
1683             return True
1684         agg_fname = self.conffile("agg.xml")
1685         with open(agg_fname,"w") as out:
1686             out.write("<aggregates>{}</aggregates>\n"\
1687                       .format(" ".join([ plc.aggregate_xml_line() for plc in other_plcs ])))
1688         utils.header("(Over)wrote {}".format(agg_fname))
1689         reg_fname=self.conffile("reg.xml")
1690         with open(reg_fname,"w") as out:
1691             out.write("<registries>{}</registries>\n"\
1692                       .format(" ".join([ plc.registry_xml_line() for plc in other_plcs ])))
1693         utils.header("(Over)wrote {}".format(reg_fname))
1694         return self.test_ssh.copy_abs(agg_fname,
1695                                       '/{}/etc/sfa/aggregates.xml'.format(self.vm_root_in_host())) == 0 \
1696            and self.test_ssh.copy_abs(reg_fname,
1697                                       '/{}/etc/sfa/registries.xml'.format(self.vm_root_in_host())) == 0
1698
1699     def sfa_import(self):
1700         "use sfaadmin to import from plc"
1701         auth = self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH']
1702         return self.run_in_guest('sfaadmin reg import_registry') == 0
1703
1704     def sfa_start(self):
1705         "service sfa start"
1706         return self.start_service('sfa')
1707
1708
1709     def sfi_configure(self):
1710         "Create /root/sfi on the plc side for sfi client configuration"
1711         if self.options.dry_run:
1712             utils.header("DRY RUN - skipping step")
1713             return True
1714         sfa_spec = self.plc_spec['sfa']
1715         # cannot use auth_sfa_mapper to pass dir_name
1716         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1717             test_slice = TestAuthSfa(self, slice_spec)
1718             dir_basename = os.path.basename(test_slice.sfi_path())
1719             dir_name = self.confsubdir("dot-sfi/{}".format(dir_basename),
1720                                        clean=True, dry_run=self.options.dry_run)
1721             test_slice.sfi_configure(dir_name)
1722             # push into the remote /root/sfi area
1723             location = test_slice.sfi_path()
1724             remote = "{}/{}".format(self.vm_root_in_host(), location)
1725             self.test_ssh.mkdir(remote, abs=True)
1726             # need to strip last level or remote otherwise we get an extra dir level
1727             self.test_ssh.copy_abs(dir_name, os.path.dirname(remote), recursive=True)
1728
1729         return True
1730
1731     def sfi_clean(self):
1732         "clean up /root/sfi on the plc side"
1733         self.run_in_guest("rm -rf /root/sfi")
1734         return True
1735
1736     def sfa_rspec_empty(self):
1737         "expose a static empty rspec (ships with the tests module) in the sfi directory"
1738         filename = "empty-rspec.xml"
1739         overall = True
1740         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1741             test_slice = TestAuthSfa(self, slice_spec)
1742             in_vm = test_slice.sfi_path()
1743             remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1744             if self.test_ssh.copy_abs(filename, remote) !=0:
1745                 overall = False
1746         return overall
1747
1748     @auth_sfa_mapper
1749     def sfa_register_site(self): pass
1750     @auth_sfa_mapper
1751     def sfa_register_pi(self): pass
1752     @auth_sfa_mapper
1753     def sfa_register_user(self): pass
1754     @auth_sfa_mapper
1755     def sfa_update_user(self): pass
1756     @auth_sfa_mapper
1757     def sfa_register_slice(self): pass
1758     @auth_sfa_mapper
1759     def sfa_renew_slice(self): pass
1760     @auth_sfa_mapper
1761     def sfa_get_expires(self): pass
1762     @auth_sfa_mapper
1763     def sfa_discover(self): pass
1764     @auth_sfa_mapper
1765     def sfa_rspec(self): pass
1766     @auth_sfa_mapper
1767     def sfa_allocate(self): pass
1768     @auth_sfa_mapper
1769     def sfa_allocate_empty(self): pass
1770     @auth_sfa_mapper
1771     def sfa_provision(self): pass
1772     @auth_sfa_mapper
1773     def sfa_provision_empty(self): pass
1774     @auth_sfa_mapper
1775     def sfa_describe(self): pass
1776     @auth_sfa_mapper
1777     def sfa_check_slice_plc(self): pass
1778     @auth_sfa_mapper
1779     def sfa_check_slice_plc_empty(self): pass
1780     @auth_sfa_mapper
1781     def sfa_update_slice(self): pass
1782     @auth_sfa_mapper
1783     def sfa_remove_user_from_slice(self): pass
1784     @auth_sfa_mapper
1785     def sfa_insert_user_in_slice(self): pass
1786     @auth_sfa_mapper
1787     def sfi_list(self): pass
1788     @auth_sfa_mapper
1789     def sfi_show_site(self): pass
1790     @auth_sfa_mapper
1791     def sfi_show_slice(self): pass
1792     @auth_sfa_mapper
1793     def sfi_show_slice_researchers(self): pass
1794     @auth_sfa_mapper
1795     def ssh_slice_sfa(self): pass
1796     @auth_sfa_mapper
1797     def sfa_delete_user(self): pass
1798     @auth_sfa_mapper
1799     def sfa_delete_slice(self): pass
1800
1801     def sfa_stop(self):
1802         "service sfa stop"
1803         return self.stop_service('sfa')
1804
1805     def populate(self):
1806         "creates random entries in the PLCAPI"
1807         # install the stress-test in the plc image
1808         location = "/usr/share/plc_api/plcsh_stress_test.py"
1809         remote = "{}/{}".format(self.vm_root_in_host(), location)
1810         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1811         command = location
1812         command += " -- --preserve --short-names"
1813         local = (self.run_in_guest(command) == 0);
1814         # second run with --foreign
1815         command += ' --foreign'
1816         remote = (self.run_in_guest(command) == 0);
1817         return local and remote
1818
1819
1820     ####################
1821     @bonding_redirector
1822     def bonding_init_partial(self): pass
1823
1824     @bonding_redirector
1825     def bonding_add_yum(self): pass
1826
1827     @bonding_redirector
1828     def bonding_install_rpms(self): pass
1829
1830     ####################
1831
1832     def gather_logs(self):
1833         "gets all possible logs from plc's/qemu node's/slice's for future reference"
1834         # (1.a) get the plc's /var/log/ and store it locally in logs/myplc.var-log.<plcname>/*
1835         # (1.b) get the plc's  /var/lib/pgsql/data/pg_log/ -> logs/myplc.pgsql-log.<plcname>/*
1836         # (1.c) get the plc's /root/sfi -> logs/sfi.<plcname>/
1837         # (2) get all the nodes qemu log and store it as logs/node.qemu.<node>.log
1838         # (3) get the nodes /var/log and store is as logs/node.var-log.<node>/*
1839         # (4) as far as possible get the slice's /var/log as logs/sliver.var-log.<sliver>/*
1840         # (1.a)
1841         print("-------------------- TestPlc.gather_logs : PLC's /var/log")
1842         self.gather_var_logs()
1843         # (1.b)
1844         print("-------------------- TestPlc.gather_logs : PLC's /var/lib/psql/data/pg_log/")
1845         self.gather_pgsql_logs()
1846         # (1.c)
1847         print("-------------------- TestPlc.gather_logs : PLC's /root/sfi/")
1848         self.gather_root_sfi()
1849         # (2)
1850         print("-------------------- TestPlc.gather_logs : nodes's QEMU logs")
1851         for site_spec in self.plc_spec['sites']:
1852             test_site = TestSite(self,site_spec)
1853             for node_spec in site_spec['nodes']:
1854                 test_node = TestNode(self, test_site, node_spec)
1855                 test_node.gather_qemu_logs()
1856         # (3)
1857         print("-------------------- TestPlc.gather_logs : nodes's /var/log")
1858         self.gather_nodes_var_logs()
1859         # (4)
1860         print("-------------------- TestPlc.gather_logs : sample sliver's /var/log")
1861         self.gather_slivers_var_logs()
1862         return True
1863
1864     def gather_slivers_var_logs(self):
1865         for test_sliver in self.all_sliver_objs():
1866             remote = test_sliver.tar_var_logs()
1867             utils.system("mkdir -p logs/sliver.var-log.{}".format(test_sliver.name()))
1868             command = remote + " | tar -C logs/sliver.var-log.{} -xf -".format(test_sliver.name())
1869             utils.system(command)
1870         return True
1871
1872     def gather_var_logs(self):
1873         utils.system("mkdir -p logs/myplc.var-log.{}".format(self.name()))
1874         to_plc = self.actual_command_in_guest("tar -C /var/log/ -cf - .")
1875         command = to_plc + "| tar -C logs/myplc.var-log.{} -xf -".format(self.name())
1876         utils.system(command)
1877         command = "chmod a+r,a+x logs/myplc.var-log.{}/httpd".format(self.name())
1878         utils.system(command)
1879
1880     def gather_pgsql_logs(self):
1881         utils.system("mkdir -p logs/myplc.pgsql-log.{}".format(self.name()))
1882         to_plc = self.actual_command_in_guest("tar -C /var/lib/pgsql/data/pg_log/ -cf - .")
1883         command = to_plc + "| tar -C logs/myplc.pgsql-log.{} -xf -".format(self.name())
1884         utils.system(command)
1885
1886     def gather_root_sfi(self):
1887         utils.system("mkdir -p logs/sfi.{}".format(self.name()))
1888         to_plc = self.actual_command_in_guest("tar -C /root/sfi/ -cf - .")
1889         command = to_plc + "| tar -C logs/sfi.{} -xf -".format(self.name())
1890         utils.system(command)
1891
1892     def gather_nodes_var_logs(self):
1893         for site_spec in self.plc_spec['sites']:
1894             test_site = TestSite(self, site_spec)
1895             for node_spec in site_spec['nodes']:
1896                 test_node = TestNode(self, test_site, node_spec)
1897                 test_ssh = TestSsh(test_node.name(), key="keys/key_admin.rsa")
1898                 command = test_ssh.actual_command("tar -C /var/log -cf - .")
1899                 command = command + "| tar -C logs/node.var-log.{} -xf -".format(test_node.name())
1900                 utils.system("mkdir -p logs/node.var-log.{}".format(test_node.name()))
1901                 utils.system(command)
1902
1903
1904     # returns the filename to use for sql dump/restore, using options.dbname if set
1905     def dbfile(self, database):
1906         # uses options.dbname if it is found
1907         try:
1908             name = self.options.dbname
1909             if not isinstance(name, str):
1910                 raise Exception
1911         except:
1912             t = datetime.now()
1913             d = t.date()
1914             name = str(d)
1915         return "/root/{}-{}.sql".format(database, name)
1916
1917     def plc_db_dump(self):
1918         'dump the planetlab5 DB in /root in the PLC - filename has time'
1919         dump=self.dbfile("planetab5")
1920         self.run_in_guest('pg_dump -U pgsqluser planetlab5 -f '+ dump)
1921         utils.header('Dumped planetlab5 database in {}'.format(dump))
1922         return True
1923
1924     def plc_db_restore(self):
1925         'restore the planetlab5 DB - looks broken, but run -n might help'
1926         dump = self.dbfile("planetab5")
1927         ##stop httpd service
1928         self.run_in_guest('service httpd stop')
1929         # xxx - need another wrapper
1930         self.run_in_guest_piped('echo drop database planetlab5', 'psql --user=pgsqluser template1')
1931         self.run_in_guest('createdb -U postgres --encoding=UNICODE --owner=pgsqluser planetlab5')
1932         self.run_in_guest('psql -U pgsqluser planetlab5 -f ' + dump)
1933         ##starting httpd service
1934         self.run_in_guest('service httpd start')
1935
1936         utils.header('Database restored from ' + dump)
1937
1938     @staticmethod
1939     def create_ignore_steps():
1940         for step in TestPlc.default_steps + TestPlc.other_steps:
1941             # default step can have a plc qualifier
1942             if '@' in step:
1943                 step, qualifier = step.split('@')
1944             # or be defined as forced or ignored by default
1945             for keyword in ['_ignore','_force']:
1946                 if step.endswith(keyword):
1947                     step=step.replace(keyword,'')
1948             if step == SEP or step == SEPSFA :
1949                 continue
1950             method = getattr(TestPlc,step)
1951             name = step + '_ignore'
1952             wrapped = ignore_result(method)
1953 #            wrapped.__doc__ = method.__doc__ + " (run in ignore-result mode)"
1954             setattr(TestPlc, name, wrapped)
1955
1956 #    @ignore_result
1957 #    def ssh_slice_again_ignore (self): pass
1958 #    @ignore_result
1959 #    def check_initscripts_ignore (self): pass
1960
1961     def standby_1_through_20(self):
1962         """convenience function to wait for a specified number of minutes"""
1963         pass
1964     @standby_generic
1965     def standby_1(): pass
1966     @standby_generic
1967     def standby_2(): pass
1968     @standby_generic
1969     def standby_3(): pass
1970     @standby_generic
1971     def standby_4(): pass
1972     @standby_generic
1973     def standby_5(): pass
1974     @standby_generic
1975     def standby_6(): pass
1976     @standby_generic
1977     def standby_7(): pass
1978     @standby_generic
1979     def standby_8(): pass
1980     @standby_generic
1981     def standby_9(): pass
1982     @standby_generic
1983     def standby_10(): pass
1984     @standby_generic
1985     def standby_11(): pass
1986     @standby_generic
1987     def standby_12(): pass
1988     @standby_generic
1989     def standby_13(): pass
1990     @standby_generic
1991     def standby_14(): pass
1992     @standby_generic
1993     def standby_15(): pass
1994     @standby_generic
1995     def standby_16(): pass
1996     @standby_generic
1997     def standby_17(): pass
1998     @standby_generic
1999     def standby_18(): pass
2000     @standby_generic
2001     def standby_19(): pass
2002     @standby_generic
2003     def standby_20(): pass
2004
2005     # convenience for debugging the test logic
2006     def yes(self): return True
2007     def no(self): return False
2008     def fail(self): return False