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         'django_install', '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 out 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',
 165         'bootcd', 'qemu_local_config', SEP,
 166         'qemu_clean_mine', 'qemu_export', 'qemu_cleanlog', SEP,
 167         'qemu_start', 'qemu_timestamp', 'qemu_nodefamily', SEP,
 168         'sfa_install_all', 'sfa_configure', 'cross_sfa_configure',
 169         'sfa_start', 'sfa_import', SEPSFA,
 170         'sfi_configure@1', 'sfa_register_site@1', 'sfa_register_pi@1', SEPSFA,
 171         'sfa_register_user@1', 'sfa_update_user@1',
 172         'sfa_register_slice@1', 'sfa_renew_slice@1', SEPSFA,
 173         'sfa_remove_user_from_slice@1', 'sfi_show_slice_researchers@1',
 174         'sfa_insert_user_in_slice@1', 'sfi_show_slice_researchers@1', SEPSFA,
 175         'sfa_discover@1', 'sfa_rspec@1', SEPSFA,
 176         'sfa_allocate@1', 'sfa_provision@1', 'sfa_describe@1', SEPSFA,
 177         'sfa_check_slice_plc@1', 'sfa_update_slice@1', SEPSFA,
 178         'sfi_list@1', 'sfi_show_site@1', 'sfa_utest@1', SEPSFA,
 179         # we used to run plcsh_stress_test, and then ssh_node_debug and ssh_node_boot
 180         # but as the stress test might take a while, we sometimes missed the debug mode..
 181         'probe_kvm_iptables',
 182         'ping_node', 'ssh_node_debug', 'plcsh_stress_test@1', SEP,
 183         'ssh_node_boot', 'node_bmlogs', 'ssh_slice', 'ssh_slice_basics', SEP,
 184         'ssh_slice_sfa@1', SEPSFA,
 185         'sfa_rspec_empty@1', 'sfa_allocate_empty@1', 'sfa_provision_empty@1',
 186         'sfa_check_slice_plc_empty@1', SEPSFA,
 187         'sfa_delete_slice@1', 'sfa_delete_user@1', SEPSFA,
 188         'cross_check_tcp@1', 'check_system_slice', SEP,
 189         # for inspecting the slice while it runs the first time
 190         #'fail',
 191         # check slices are turned off properly
 192         'debug_nodemanager',
 193         'empty_slices', 'ssh_slice_off', 'slice_fs_deleted_ignore', SEP,
 194         # check they are properly re-created with the same name
 195         'fill_slices', 'ssh_slice_again', SEP,
 196         'gather_logs_force', SEP,
 197         ]
 198     other_steps = [
 199         'export', 'show_boxes', 'super_speed_up_slices', SEP,
 200         'check_hooks', 'plc_stop', 'plcvm_start', 'plcvm_stop', SEP,
 201         'delete_initscripts', 'delete_nodegroups', 'delete_all_sites', SEP,
 202         'delete_sites', 'delete_nodes', 'delete_slices', 'keys_clean', SEP,
 203         'delete_leases', 'list_leases', SEP,
 204         'populate', SEP,
 205         'nodestate_show', 'nodestate_safeboot', 'nodestate_boot', 'nodestate_upgrade', SEP,
 206         'nodedistro_show', 'nodedistro_f14', 'nodedistro_f18', SEP,
 207         'nodedistro_f20', 'nodedistro_f21', 'nodedistro_f22', SEP,
 208         'qemu_list_all', 'qemu_list_mine', 'qemu_kill_all', SEP,
 209         'sfa_install_core', 'sfa_install_sfatables',
 210         'sfa_install_plc', 'sfa_install_client', SEPSFA,
 211         'sfa_plcclean', 'sfa_dbclean', 'sfa_stop', 'sfa_uninstall', 'sfi_clean', SEPSFA,
 212         'sfa_get_expires', SEPSFA,
 213         'plc_db_dump', 'plc_db_restore', SEP,
 214         'check_netflow', 'check_drl', SEP,
 215         'slice_fs_present', 'check_initscripts', SEP,
 216         'standby_1_through_20', 'yes', 'no', SEP,
 217         'install_syslinux6', 'bonding_builds', 'bonding_nodes', SEP,
 218         ]
 219     default_bonding_steps = [
 220         'bonding_init_partial',
 221         'bonding_add_yum',
 222         'bonding_install_rpms', SEP,
 223         ]
 224
 225     @staticmethod
 226     def printable_steps(list):
 227         single_line = " ".join(list) + " "
 228         return single_line.replace(" "+SEP+" ", " \\\n").replace(" "+SEPSFA+" ", " \\\n")
 229     @staticmethod
 230     def valid_step(step):
 231         return step != SEP and step != SEPSFA
 232
 233     # turn off the sfa-related steps when build has skipped SFA
 234     # this was originally for centos5 but is still valid
 235     # for up to f12 as recent SFAs with sqlalchemy won't build before f14
 236     @staticmethod
 237     def _has_sfa_cached(rpms_url):
 238         if os.path.isfile(has_sfa_cache_filename):
 239             with open(has_sfa_cache_filename) as cache:
 240                 cached = cache.read() == "yes"
 241             utils.header("build provides SFA (cached):{}".format(cached))
 242             return cached
 243         # warning, we're now building 'sface' so let's be a bit more picky
 244         # full builds are expected to return with 0 here
 245         utils.header("Checking if build provides SFA package...")
 246         retcod = utils.system("curl --silent {}/ | grep -q sfa-".format(rpms_url)) == 0
 247         encoded = 'yes' if retcod else 'no'
 248         with open(has_sfa_cache_filename,'w') as cache:
 249             cache.write(encoded)
 250         return retcod
 251
 252     @staticmethod
 253     def check_whether_build_has_sfa(rpms_url):
 254         has_sfa = TestPlc._has_sfa_cached(rpms_url)
 255         if has_sfa:
 256             utils.header("build does provide SFA")
 257         else:
 258             # move all steps containing 'sfa' from default_steps to other_steps
 259             utils.header("SFA package not found - removing steps with sfa or sfi")
 260             sfa_steps = [ step for step in TestPlc.default_steps
 261                           if step.find('sfa') >= 0 or step.find("sfi") >= 0 ]
 262             TestPlc.other_steps += sfa_steps
 263             for step in sfa_steps:
 264                 TestPlc.default_steps.remove(step)
 265
 266     def __init__(self, plc_spec, options):
 267         self.plc_spec = plc_spec
 268         self.options = options
 269         self.test_ssh = TestSsh(self.plc_spec['host_box'], self.options.buildname)
 270         self.vserverip = plc_spec['vserverip']
 271         self.vservername = plc_spec['vservername']
 272         self.vplchostname = self.vservername.split('-')[-1]
 273         self.url = "https://{}:443/PLCAPI/".format(plc_spec['vserverip'])
 274         self.apiserver = TestApiserver(self.url, options.dry_run)
 275         (self.ssh_node_boot_timeout, self.ssh_node_boot_silent) = plc_spec['ssh_node_boot_timers']
 276         (self.ssh_node_debug_timeout, self.ssh_node_debug_silent) = plc_spec['ssh_node_debug_timers']
 277
 278     def has_addresses_api(self):
 279         return self.apiserver.has_method('AddIpAddress')
 280
 281     def name(self):
 282         name = self.plc_spec['name']
 283         return "{}.{}".format(name,self.vservername)
 284
 285     def hostname(self):
 286         return self.plc_spec['host_box']
 287
 288     def is_local(self):
 289         return self.test_ssh.is_local()
 290
 291     # define the API methods on this object through xmlrpc
 292     # would help, but not strictly necessary
 293     def connect(self):
 294         pass
 295
 296     def actual_command_in_guest(self,command, backslash=False):
 297         raw1 = self.host_to_guest(command)
 298         raw2 = self.test_ssh.actual_command(raw1, dry_run=self.options.dry_run, backslash=backslash)
 299         return raw2
 300
 301     def start_guest(self):
 302       return utils.system(self.test_ssh.actual_command(self.start_guest_in_host(),
 303                                                        dry_run=self.options.dry_run))
 304
 305     def stop_guest(self):
 306       return utils.system(self.test_ssh.actual_command(self.stop_guest_in_host(),
 307                                                        dry_run=self.options.dry_run))
 308
 309     def run_in_guest(self, command, backslash=False):
 310         raw = self.actual_command_in_guest(command, backslash)
 311         return utils.system(raw)
 312
 313     def run_in_host(self,command):
 314         return self.test_ssh.run_in_buildname(command, dry_run=self.options.dry_run)
 315
 316     # backslashing turned out so awful at some point that I've turned off auto-backslashing
 317     # see e.g. plc_start esp. the version for f14
 318     #command gets run in the plc's vm
 319     def host_to_guest(self, command):
 320         ssh_leg = TestSsh(self.vplchostname)
 321         return ssh_leg.actual_command(command, keep_stdin=True)
 322
 323     # this /vservers thing is legacy...
 324     def vm_root_in_host(self):
 325         return "/vservers/{}/".format(self.vservername)
 326
 327     def vm_timestamp_path(self):
 328         return "/vservers/{}/{}.timestamp".format(self.vservername, self.vservername)
 329
 330     #start/stop the vserver
 331     def start_guest_in_host(self):
 332         return "virsh -c lxc:/// start {}".format(self.vservername)
 333
 334     def stop_guest_in_host(self):
 335         return "virsh -c lxc:/// destroy {}".format(self.vservername)
 336
 337     # xxx quick n dirty
 338     def run_in_guest_piped(self,local,remote):
 339         return utils.system(local+" | "+self.test_ssh.actual_command(self.host_to_guest(remote),
 340                                                                      keep_stdin = True))
 341
 342     def dnf_check_installed(self, rpms):
 343         if isinstance(rpms, list):
 344             rpms=" ".join(rpms)
 345         return self.run_in_guest("rpm -q {}".format(rpms)) == 0
 346
 347     # does a yum install in the vs, ignore yum retcod, check with rpm
 348     def dnf_install(self, rpms):
 349         if isinstance(rpms, list):
 350             rpms=" ".join(rpms)
 351         yum_mode = self.run_in_guest("dnf -y install {}".format(rpms))
 352         if yum_mode != 0:
 353             self.run_in_guest("dnf -y install --allowerasing {}".format(rpms))
 354         # yum-complete-transaction comes with yum-utils, that is in vtest.pkgs
 355         # nothing similar with dnf, forget about this for now
 356         # self.run_in_guest("yum-complete-transaction -y")
 357         return self.dnf_check_installed(rpms)
 358
 359     def pip_install(self, package):
 360         return self.run_in_guest("pip3 install {}".format(package)) == 0
 361
 362     def auth_root(self):
 363         return {'Username'   : self.plc_spec['settings']['PLC_ROOT_USER'],
 364                 'AuthMethod' : 'password',
 365                 'AuthString' : self.plc_spec['settings']['PLC_ROOT_PASSWORD'],
 366                 'Role'       : self.plc_spec['role'],
 367                 }
 368
 369     def locate_site(self,sitename):
 370         for site in self.plc_spec['sites']:
 371             if site['site_fields']['name'] == sitename:
 372                 return site
 373             if site['site_fields']['login_base'] == sitename:
 374                 return site
 375         raise Exception("Cannot locate site {}".format(sitename))
 376
 377     def locate_node(self, nodename):
 378         for site in self.plc_spec['sites']:
 379             for node in site['nodes']:
 380                 if node['name'] == nodename:
 381                     return site, node
 382         raise Exception("Cannot locate node {}".format(nodename))
 383
 384     def locate_hostname(self, hostname):
 385         for site in self.plc_spec['sites']:
 386             for node in site['nodes']:
 387                 if node['node_fields']['hostname'] == hostname:
 388                     return(site, node)
 389         raise Exception("Cannot locate hostname {}".format(hostname))
 390
 391     def locate_key(self, key_name):
 392         for key in self.plc_spec['keys']:
 393             if key['key_name'] == key_name:
 394                 return key
 395         raise Exception("Cannot locate key {}".format(key_name))
 396
 397     def locate_private_key_from_key_names(self, key_names):
 398         # locate the first avail. key
 399         found = False
 400         for key_name in key_names:
 401             key_spec = self.locate_key(key_name)
 402             test_key = TestKey(self,key_spec)
 403             publickey = test_key.publicpath()
 404             privatekey = test_key.privatepath()
 405             if os.path.isfile(publickey) and os.path.isfile(privatekey):
 406                 found = True
 407         if found:
 408             return privatekey
 409         else:
 410             return None
 411
 412     def locate_slice(self, slicename):
 413         for slice in self.plc_spec['slices']:
 414             if slice['slice_fields']['name'] == slicename:
 415                 return slice
 416         raise Exception("Cannot locate slice {}".format(slicename))
 417
 418     def all_sliver_objs(self):
 419         result = []
 420         for slice_spec in self.plc_spec['slices']:
 421             slicename = slice_spec['slice_fields']['name']
 422             for nodename in slice_spec['nodenames']:
 423                 result.append(self.locate_sliver_obj(nodename, slicename))
 424         return result
 425
 426     def locate_sliver_obj(self, nodename, slicename):
 427         site,node = self.locate_node(nodename)
 428         slice = self.locate_slice(slicename)
 429         # build objects
 430         test_site = TestSite(self, site)
 431         test_node = TestNode(self, test_site, node)
 432         # xxx the slice site is assumed to be the node site - mhh - probably harmless
 433         test_slice = TestSlice(self, test_site, slice)
 434         return TestSliver(self, test_node, test_slice)
 435
 436     def locate_first_node(self):
 437         nodename = self.plc_spec['slices'][0]['nodenames'][0]
 438         site,node = self.locate_node(nodename)
 439         test_site = TestSite(self, site)
 440         test_node = TestNode(self, test_site, node)
 441         return test_node
 442
 443     def locate_first_sliver(self):
 444         slice_spec = self.plc_spec['slices'][0]
 445         slicename = slice_spec['slice_fields']['name']
 446         nodename = slice_spec['nodenames'][0]
 447         return self.locate_sliver_obj(nodename,slicename)
 448
 449     # all different hostboxes used in this plc
 450     def get_BoxNodes(self):
 451         # maps on sites and nodes, return [ (host_box,test_node) ]
 452         tuples = []
 453         for site_spec in self.plc_spec['sites']:
 454             test_site = TestSite(self,site_spec)
 455             for node_spec in site_spec['nodes']:
 456                 test_node = TestNode(self, test_site, node_spec)
 457                 if not test_node.is_real():
 458                     tuples.append( (test_node.host_box(),test_node) )
 459         # transform into a dict { 'host_box' -> [ test_node .. ] }
 460         result = {}
 461         for (box,node) in tuples:
 462             if box not in result:
 463                 result[box] = [node]
 464             else:
 465                 result[box].append(node)
 466         return result
 467
 468     # a step for checking this stuff
 469     def show_boxes(self):
 470         'print summary of nodes location'
 471         for box,nodes in self.get_BoxNodes().items():
 472             print(box,":"," + ".join( [ node.name() for node in nodes ] ))
 473         return True
 474
 475     # make this a valid step
 476     def qemu_kill_all(self):
 477         'kill all qemu instances on the qemu boxes involved by this setup'
 478         # this is the brute force version, kill all qemus on that host box
 479         for (box,nodes) in self.get_BoxNodes().items():
 480             # pass the first nodename, as we don't push template-qemu on testboxes
 481             nodedir = nodes[0].nodedir()
 482             TestBoxQemu(box, self.options.buildname).qemu_kill_all(nodedir)
 483         return True
 484
 485     # make this a valid step
 486     def qemu_list_all(self):
 487         'list all qemu instances on the qemu boxes involved by this setup'
 488         for box,nodes in self.get_BoxNodes().items():
 489             # this is the brute force version, kill all qemus on that host box
 490             TestBoxQemu(box, self.options.buildname).qemu_list_all()
 491         return True
 492
 493     # kill only the qemus related to this test
 494     def qemu_list_mine(self):
 495         'list 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.list_qemu()
 500         return True
 501
 502     # kill only the qemus related to this test
 503     def qemu_clean_mine(self):
 504         'cleanup (rm -rf) 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.qemu_clean()
 509         return True
 510
 511     # kill only the right qemus
 512     def qemu_kill_mine(self):
 513         'kill the qemu instances for our nodes'
 514         for box,nodes in self.get_BoxNodes().items():
 515             # the fine-grain version
 516             for node in nodes:
 517                 node.kill_qemu()
 518         return True
 519
 520     #################### display config
 521     def show(self):
 522         "show test configuration after localization"
 523         self.show_pass(1)
 524         self.show_pass(2)
 525         return True
 526
 527     # uggly hack to make sure 'run export' only reports about the 1st plc
 528     # to avoid confusion - also we use 'inri_slice1' in various aliases..
 529     exported_id = 1
 530     def export(self):
 531         "print cut'n paste-able stuff to export env variables to your shell"
 532         # guess local domain from hostname
 533         if TestPlc.exported_id > 1:
 534             print("export GUESTHOSTNAME{:d}={}".format(TestPlc.exported_id, self.plc_spec['vservername']))
 535             return True
 536         TestPlc.exported_id += 1
 537         domain = socket.gethostname().split('.',1)[1]
 538         fqdn   = "{}.{}".format(self.plc_spec['host_box'], domain)
 539         print("export BUILD={}".format(self.options.buildname))
 540         print("export PLCHOSTLXC={}".format(fqdn))
 541         print("export GUESTNAME={}".format(self.vservername))
 542         print("export GUESTHOSTNAME={}.{}".format(self.vplchostname, domain))
 543         # find hostname of first node
 544         hostname, qemubox = self.all_node_infos()[0]
 545         print("export KVMHOST={}.{}".format(qemubox, domain))
 546         print("export NODE={}".format(hostname))
 547         return True
 548
 549     # entry point
 550     always_display_keys=['PLC_WWW_HOST', 'nodes', 'sites']
 551     def show_pass(self, passno):
 552         for (key,val) in self.plc_spec.items():
 553             if not self.options.verbose and key not in TestPlc.always_display_keys:
 554                 continue
 555             if passno == 2:
 556                 if key == 'sites':
 557                     for site in val:
 558                         self.display_site_spec(site)
 559                         for node in site['nodes']:
 560                             self.display_node_spec(node)
 561                 elif key == 'initscripts':
 562                     for initscript in val:
 563                         self.display_initscript_spec(initscript)
 564                 elif key == 'slices':
 565                     for slice in val:
 566                         self.display_slice_spec(slice)
 567                 elif key == 'keys':
 568                     for key in val:
 569                         self.display_key_spec(key)
 570             elif passno == 1:
 571                 if key not in ['sites', 'initscripts', 'slices', 'keys']:
 572                     print('+   ', key, ':', val)
 573
 574     def display_site_spec(self, site):
 575         print('+ ======== site', site['site_fields']['name'])
 576         for k,v in site.items():
 577             if not self.options.verbose and k not in TestPlc.always_display_keys:
 578                 continue
 579             if k == 'nodes':
 580                 if v:
 581                     print('+       ', 'nodes : ', end=' ')
 582                     for node in v:
 583                         print(node['node_fields']['hostname'],'', end=' ')
 584                     print('')
 585             elif k == 'users':
 586                 if v:
 587                     print('+       users : ', end=' ')
 588                     for user in v:
 589                         print(user['name'],'', end=' ')
 590                     print('')
 591             elif k == 'site_fields':
 592                 print('+       login_base', ':', v['login_base'])
 593             elif k == 'address_fields':
 594                 pass
 595             else:
 596                 print('+       ', end=' ')
 597                 utils.pprint(k, v)
 598
 599     def display_initscript_spec(self, initscript):
 600         print('+ ======== initscript', initscript['initscript_fields']['name'])
 601
 602     def display_key_spec(self, key):
 603         print('+ ======== key', key['key_name'])
 604
 605     def display_slice_spec(self, slice):
 606         print('+ ======== slice', slice['slice_fields']['name'])
 607         for k,v in slice.items():
 608             if k == 'nodenames':
 609                 if v:
 610                     print('+       nodes : ', end=' ')
 611                     for nodename in v:
 612                         print(nodename,'', end=' ')
 613                     print('')
 614             elif k == 'usernames':
 615                 if v:
 616                     print('+       users : ', end=' ')
 617                     for username in v:
 618                         print(username,'', end=' ')
 619                     print('')
 620             elif k == 'slice_fields':
 621                 print('+       fields', ':', end=' ')
 622                 print('max_nodes=',v['max_nodes'], end=' ')
 623                 print('')
 624             else:
 625                 print('+       ',k,v)
 626
 627     def display_node_spec(self, node):
 628         print("+           node={} host_box={}".format(node['name'], node['host_box']), end=' ')
 629         print("hostname=", node['node_fields']['hostname'], end=' ')
 630         print("ip=", node['interface_fields']['ip'])
 631         if self.options.verbose:
 632             utils.pprint("node details", node, depth=3)
 633
 634     # another entry point for just showing the boxes involved
 635     def display_mapping(self):
 636         TestPlc.display_mapping_plc(self.plc_spec)
 637         return True
 638
 639     @staticmethod
 640     def display_mapping_plc(plc_spec):
 641         print('+ MyPLC',plc_spec['name'])
 642         # WARNING this would not be right for lxc-based PLC's - should be harmless though
 643         print('+\tvserver address = root@{}:/vservers/{}'.format(plc_spec['host_box'], plc_spec['vservername']))
 644         print('+\tIP = {}/{}'.format(plc_spec['settings']['PLC_API_HOST'], plc_spec['vserverip']))
 645         for site_spec in plc_spec['sites']:
 646             for node_spec in site_spec['nodes']:
 647                 TestPlc.display_mapping_node(node_spec)
 648
 649     @staticmethod
 650     def display_mapping_node(node_spec):
 651         print('+   NODE {}'.format(node_spec['name']))
 652         print('+\tqemu box {}'.format(node_spec['host_box']))
 653         print('+\thostname={}'.format(node_spec['node_fields']['hostname']))
 654
 655     # write a timestamp in /vservers/<>.timestamp
 656     # cannot be inside the vserver, that causes vserver .. build to cough
 657     def plcvm_timestamp(self):
 658         "Create a timestamp to remember creation date for this plc"
 659         now = int(time.time())
 660         # TODO-lxc check this one
 661         # a first approx. is to store the timestamp close to the VM root like vs does
 662         stamp_path = self.vm_timestamp_path()
 663         stamp_dir = os.path.dirname(stamp_path)
 664         utils.system(self.test_ssh.actual_command("mkdir -p {}".format(stamp_dir)))
 665         return utils.system(self.test_ssh.actual_command("echo {:d} > {}".format(now, stamp_path))) == 0
 666
 667     # this is called inconditionnally at the beginning of the test sequence
 668     # just in case this is a rerun, so if the vm is not running it's fine
 669     def plcvm_delete(self):
 670         "vserver delete the test myplc"
 671         stamp_path = self.vm_timestamp_path()
 672         self.run_in_host("rm -f {}".format(stamp_path))
 673         self.run_in_host("virsh -c lxc:/// destroy {}".format(self.vservername))
 674         self.run_in_host("virsh -c lxc:/// undefine {}".format(self.vservername))
 675         self.run_in_host("rm -fr /vservers/{}".format(self.vservername))
 676         return True
 677
 678     ### install
 679     # historically the build was being fetched by the tests
 680     # now the build pushes itself as a subdir of the tests workdir
 681     # so that the tests do not have to worry about extracting the build (svn, git, or whatever)
 682     def plcvm_create(self):
 683         "vserver creation (no install done)"
 684         # push the local build/ dir to the testplc box
 685         if self.is_local():
 686             # a full path for the local calls
 687             build_dir = os.path.dirname(sys.argv[0])
 688             # sometimes this is empty - set to "." in such a case
 689             if not build_dir:
 690                 build_dir="."
 691             build_dir += "/build"
 692         else:
 693             # use a standard name - will be relative to remote buildname
 694             build_dir = "build"
 695             # remove for safety; do *not* mkdir first, otherwise we end up with build/build/
 696             self.test_ssh.rmdir(build_dir)
 697             self.test_ssh.copy(build_dir, recursive=True)
 698         # the repo url is taken from arch-rpms-url
 699         # with the last step (i386) removed
 700         repo_url = self.options.arch_rpms_url
 701         for level in [ 'arch' ]:
 702             repo_url = os.path.dirname(repo_url)
 703
 704         # invoke initvm (drop support for vs)
 705         script = "lbuild-initvm.sh"
 706         script_options = ""
 707         # pass the vbuild-nightly options to [lv]test-initvm
 708         script_options += " -p {}".format(self.options.personality)
 709         script_options += " -d {}".format(self.options.pldistro)
 710         script_options += " -f {}".format(self.options.fcdistro)
 711         script_options += " -r {}".format(repo_url)
 712         vserver_name = self.vservername
 713         try:
 714             vserver_hostname = socket.gethostbyaddr(self.vserverip)[0]
 715             script_options += " -n {}".format(vserver_hostname)
 716         except:
 717             print("Cannot reverse lookup {}".format(self.vserverip))
 718             print("This is considered fatal, as this might pollute the test results")
 719             return False
 720         create_vserver="{build_dir}/{script} {script_options} {vserver_name}".format(**locals())
 721         return self.run_in_host(create_vserver) == 0
 722
 723     ### install django through pip
 724     def django_install(self):
 725         # plcapi requires Django, that is no longer provided py fedora as an rpm
 726         # so we use pip instead
 727         """
 728         pip install Django
 729         """
 730         return self.pip_install('Django')
 731
 732     ### install_rpm
 733     def plc_install(self):
 734         """
 735         yum install myplc, noderepo
 736         """
 737
 738         # compute nodefamily
 739         if self.options.personality == "linux32":
 740             arch = "i386"
 741         elif self.options.personality == "linux64":
 742             arch = "x86_64"
 743         else:
 744             raise Exception("Unsupported personality {}".format(self.options.personality))
 745         nodefamily = "{}-{}-{}".format(self.options.pldistro, self.options.fcdistro, arch)
 746
 747         # check it's possible to install just 'myplc-core' first
 748         if not self.dnf_install("myplc-core"):
 749             return False
 750
 751         pkgs_list = []
 752         pkgs_list.append("myplc")
 753         pkgs_list.append("slicerepo-{}".format(nodefamily))
 754         pkgs_list.append("noderepo-{}".format(nodefamily))
 755         pkgs_string=" ".join(pkgs_list)
 756         return self.dnf_install(pkgs_list)
 757
 758     def install_syslinux6(self):
 759         """
 760         install syslinux6 from the fedora21 release
 761         """
 762         key = 'http://mirror.onelab.eu/keys/RPM-GPG-KEY-fedora-21-primary'
 763
 764         rpms = [
 765             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-6.03-1.fc21.x86_64.rpm',
 766             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-nonlinux-6.03-1.fc21.noarch.rpm',
 767             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-perl-6.03-1.fc21.x86_64.rpm',
 768         ]
 769         # this can be done several times
 770         self.run_in_guest("rpm --import {key}".format(**locals()))
 771         return self.run_in_guest("yum -y localinstall {}".format(" ".join(rpms))) == 0
 772
 773     def bonding_builds(self):
 774         """
 775         list /etc/yum.repos.d on the myplc side
 776         """
 777         self.run_in_guest("ls /etc/yum.repos.d/*partial.repo")
 778         return True
 779
 780     def bonding_nodes(self):
 781         """
 782         List nodes known to the myplc together with their nodefamiliy
 783         """
 784         print("---------------------------------------- nodes")
 785         for node in self.apiserver.GetNodes(self.auth_root()):
 786             print("{} -> {}".format(node['hostname'],
 787                                     self.apiserver.GetNodeFlavour(self.auth_root(),node['hostname'])['nodefamily']))
 788         print("---------------------------------------- nodes")
 789
 790
 791     ###
 792     def mod_python(self):
 793         """yum install mod_python, useful on f18 and above so as to avoid broken wsgi"""
 794         return self.dnf_install( ['mod_python'] )
 795
 796     ###
 797     def plc_configure(self):
 798         "run plc-config-tty"
 799         tmpname = '{}.plc-config-tty'.format(self.name())
 800         with open(tmpname,'w') as fileconf:
 801             for var, value in self.plc_spec['settings'].items():
 802                 fileconf.write('e {}\n{}\n'.format(var, value))
 803             fileconf.write('w\n')
 804             fileconf.write('q\n')
 805         utils.system('cat {}'.format(tmpname))
 806         self.run_in_guest_piped('cat {}'.format(tmpname), 'plc-config-tty')
 807         utils.system('rm {}'.format(tmpname))
 808         return True
 809
 810     # care only about f>=27
 811     def start_stop_systemd(self, service, start_or_stop):
 812         "utility to start/stop a systemd-defined service (sfa)"
 813         return self.run_in_guest("systemctl {} {}".format(start_or_stop, service)) == 0
 814
 815     def plc_start(self):
 816         "start plc through systemclt"
 817         return self.start_stop_systemd('plc', 'start')
 818
 819     def plc_stop(self):
 820         "stop plc through systemctl"
 821         return self.start_stop_systemd('plc', 'stop')
 822
 823     def plcvm_start(self):
 824         "start the PLC vserver"
 825         self.start_guest()
 826         return True
 827
 828     def plcvm_stop(self):
 829         "stop the PLC vserver"
 830         self.stop_guest()
 831         return True
 832
 833     # stores the keys from the config for further use
 834     def keys_store(self):
 835         "stores test users ssh keys in keys/"
 836         for key_spec in self.plc_spec['keys']:
 837                 TestKey(self,key_spec).store_key()
 838         return True
 839
 840     def keys_clean(self):
 841         "removes keys cached in keys/"
 842         utils.system("rm -rf ./keys")
 843         return True
 844
 845     # fetches the ssh keys in the plc's /etc/planetlab and stores them in keys/
 846     # for later direct access to the nodes
 847     def keys_fetch(self):
 848         "gets ssh keys in /etc/planetlab/ and stores them locally in keys/"
 849         dir="./keys"
 850         if not os.path.isdir(dir):
 851             os.mkdir(dir)
 852         vservername = self.vservername
 853         vm_root = self.vm_root_in_host()
 854         overall = True
 855         prefix = 'debug_ssh_key'
 856         for ext in ['pub', 'rsa'] :
 857             src = "{vm_root}/etc/planetlab/{prefix}.{ext}".format(**locals())
 858             dst = "keys/{vservername}-debug.{ext}".format(**locals())
 859             if self.test_ssh.fetch(src, dst) != 0:
 860                 overall=False
 861         return overall
 862
 863     def sites(self):
 864         "create sites with PLCAPI"
 865         return self.do_sites()
 866
 867     def delete_sites(self):
 868         "delete sites with PLCAPI"
 869         return self.do_sites(action="delete")
 870
 871     def do_sites(self, action="add"):
 872         for site_spec in self.plc_spec['sites']:
 873             test_site = TestSite(self,site_spec)
 874             if (action != "add"):
 875                 utils.header("Deleting site {} in {}".format(test_site.name(), self.name()))
 876                 test_site.delete_site()
 877                 # deleted with the site
 878                 #test_site.delete_users()
 879                 continue
 880             else:
 881                 utils.header("Creating site {} & users in {}".format(test_site.name(), self.name()))
 882                 test_site.create_site()
 883                 test_site.create_users()
 884         return True
 885
 886     def delete_all_sites(self):
 887         "Delete all sites in PLC, and related objects"
 888         print('auth_root', self.auth_root())
 889         sites = self.apiserver.GetSites(self.auth_root(), {}, ['site_id', 'login_base'])
 890         for site in sites:
 891             # keep automatic site - otherwise we shoot in our own foot, root_auth is not valid anymore
 892             if site['login_base'] == self.plc_spec['settings']['PLC_SLICE_PREFIX']:
 893                 continue
 894             site_id = site['site_id']
 895             print('Deleting site_id', site_id)
 896             self.apiserver.DeleteSite(self.auth_root(), site_id)
 897         return True
 898
 899     def nodes(self):
 900         "create nodes with PLCAPI"
 901         return self.do_nodes()
 902     def delete_nodes(self):
 903         "delete nodes with PLCAPI"
 904         return self.do_nodes(action="delete")
 905
 906     def do_nodes(self, action="add"):
 907         for site_spec in self.plc_spec['sites']:
 908             test_site = TestSite(self, site_spec)
 909             if action != "add":
 910                 utils.header("Deleting nodes in site {}".format(test_site.name()))
 911                 for node_spec in site_spec['nodes']:
 912                     test_node = TestNode(self, test_site, node_spec)
 913                     utils.header("Deleting {}".format(test_node.name()))
 914                     test_node.delete_node()
 915             else:
 916                 utils.header("Creating nodes for site {} in {}".format(test_site.name(), self.name()))
 917                 for node_spec in site_spec['nodes']:
 918                     utils.pprint('Creating node {}'.format(node_spec), node_spec)
 919                     test_node = TestNode(self, test_site, node_spec)
 920                     test_node.create_node()
 921         return True
 922
 923     def nodegroups(self):
 924         "create nodegroups with PLCAPI"
 925         return self.do_nodegroups("add")
 926     def delete_nodegroups(self):
 927         "delete nodegroups with PLCAPI"
 928         return self.do_nodegroups("delete")
 929
 930     YEAR = 365*24*3600
 931     @staticmethod
 932     def translate_timestamp(start, grain, timestamp):
 933         if timestamp < TestPlc.YEAR:
 934             return start + timestamp*grain
 935         else:
 936             return timestamp
 937
 938     @staticmethod
 939     def timestamp_printable(timestamp):
 940         return time.strftime('%m-%d %H:%M:%S UTC', time.gmtime(timestamp))
 941
 942     def leases(self):
 943         "create leases (on reservable nodes only, use e.g. run -c default -c resa)"
 944         now = int(time.time())
 945         grain = self.apiserver.GetLeaseGranularity(self.auth_root())
 946         print('API answered grain=', grain)
 947         start = (now//grain)*grain
 948         start += grain
 949         # find out all nodes that are reservable
 950         nodes = self.all_reservable_nodenames()
 951         if not nodes:
 952             utils.header("No reservable node found - proceeding without leases")
 953             return True
 954         ok = True
 955         # attach them to the leases as specified in plc_specs
 956         # this is where the 'leases' field gets interpreted as relative of absolute
 957         for lease_spec in self.plc_spec['leases']:
 958             # skip the ones that come with a null slice id
 959             if not lease_spec['slice']:
 960                 continue
 961             lease_spec['t_from']  = TestPlc.translate_timestamp(start, grain, lease_spec['t_from'])
 962             lease_spec['t_until'] = TestPlc.translate_timestamp(start, grain, lease_spec['t_until'])
 963             lease_addition = self.apiserver.AddLeases(self.auth_root(), nodes, lease_spec['slice'],
 964                                                       lease_spec['t_from'], lease_spec['t_until'])
 965             if lease_addition['errors']:
 966                 utils.header("Cannot create leases, {}".format(lease_addition['errors']))
 967                 ok = False
 968             else:
 969                 utils.header('Leases on nodes {} for {} from {:d} ({}) until {:d} ({})'\
 970                              .format(nodes, lease_spec['slice'],
 971                                      lease_spec['t_from'],  TestPlc.timestamp_printable(lease_spec['t_from']),
 972                                      lease_spec['t_until'], TestPlc.timestamp_printable(lease_spec['t_until'])))
 973
 974         return ok
 975
 976     def delete_leases(self):
 977         "remove all leases in the myplc side"
 978         lease_ids = [ l['lease_id'] for l in self.apiserver.GetLeases(self.auth_root())]
 979         utils.header("Cleaning leases {}".format(lease_ids))
 980         self.apiserver.DeleteLeases(self.auth_root(), lease_ids)
 981         return True
 982
 983     def list_leases(self):
 984         "list all leases known to the myplc"
 985         leases = self.apiserver.GetLeases(self.auth_root())
 986         now = int(time.time())
 987         for l in leases:
 988             current = l['t_until'] >= now
 989             if self.options.verbose or current:
 990                 utils.header("{} {} from {} until {}"\
 991                              .format(l['hostname'], l['name'],
 992                                      TestPlc.timestamp_printable(l['t_from']),
 993                                      TestPlc.timestamp_printable(l['t_until'])))
 994         return True
 995
 996     # create nodegroups if needed, and populate
 997     def do_nodegroups(self, action="add"):
 998         # 1st pass to scan contents
 999         groups_dict = {}
1000         for site_spec in self.plc_spec['sites']:
1001             test_site = TestSite(self,site_spec)
1002             for node_spec in site_spec['nodes']:
1003                 test_node = TestNode(self, test_site, node_spec)
1004                 if 'nodegroups' in node_spec:
1005                     nodegroupnames = node_spec['nodegroups']
1006                     if isinstance(nodegroupnames, str):
1007                         nodegroupnames = [ nodegroupnames ]
1008                     for nodegroupname in nodegroupnames:
1009                         if nodegroupname not in groups_dict:
1010                             groups_dict[nodegroupname] = []
1011                         groups_dict[nodegroupname].append(test_node.name())
1012         auth = self.auth_root()
1013         overall = True
1014         for (nodegroupname,group_nodes) in groups_dict.items():
1015             if action == "add":
1016                 print('nodegroups:', 'dealing with nodegroup',\
1017                     nodegroupname, 'on nodes', group_nodes)
1018                 # first, check if the nodetagtype is here
1019                 tag_types = self.apiserver.GetTagTypes(auth, {'tagname':nodegroupname})
1020                 if tag_types:
1021                     tag_type_id = tag_types[0]['tag_type_id']
1022                 else:
1023                     tag_type_id = self.apiserver.AddTagType(auth,
1024                                                             {'tagname' : nodegroupname,
1025                                                              'description' : 'for nodegroup {}'.format(nodegroupname),
1026                                                              'category' : 'test'})
1027                 print('located tag (type)', nodegroupname, 'as', tag_type_id)
1028                 # create nodegroup
1029                 nodegroups = self.apiserver.GetNodeGroups(auth, {'groupname' : nodegroupname})
1030                 if not nodegroups:
1031                     self.apiserver.AddNodeGroup(auth, nodegroupname, tag_type_id, 'yes')
1032                     print('created nodegroup', nodegroupname, \
1033                         'from tagname', nodegroupname, 'and value', 'yes')
1034                 # set node tag on all nodes, value='yes'
1035                 for nodename in group_nodes:
1036                     try:
1037                         self.apiserver.AddNodeTag(auth, nodename, nodegroupname, "yes")
1038                     except:
1039                         traceback.print_exc()
1040                         print('node', nodename, 'seems to already have tag', nodegroupname)
1041                     # check anyway
1042                     try:
1043                         expect_yes = self.apiserver.GetNodeTags(auth,
1044                                                                 {'hostname' : nodename,
1045                                                                  'tagname'  : nodegroupname},
1046                                                                 ['value'])[0]['value']
1047                         if expect_yes != "yes":
1048                             print('Mismatch node tag on node',nodename,'got',expect_yes)
1049                             overall = False
1050                     except:
1051                         if not self.options.dry_run:
1052                             print('Cannot find tag', nodegroupname, 'on node', nodename)
1053                             overall = False
1054             else:
1055                 try:
1056                     print('cleaning nodegroup', nodegroupname)
1057                     self.apiserver.DeleteNodeGroup(auth, nodegroupname)
1058                 except:
1059                     traceback.print_exc()
1060                     overall = False
1061         return overall
1062
1063     # a list of TestNode objs
1064     def all_nodes(self):
1065         nodes=[]
1066         for site_spec in self.plc_spec['sites']:
1067             test_site = TestSite(self,site_spec)
1068             for node_spec in site_spec['nodes']:
1069                 nodes.append(TestNode(self, test_site, node_spec))
1070         return nodes
1071
1072     # return a list of tuples (nodename,qemuname)
1073     def all_node_infos(self) :
1074         node_infos = []
1075         for site_spec in self.plc_spec['sites']:
1076             node_infos += [ (node_spec['node_fields']['hostname'], node_spec['host_box']) \
1077                                 for node_spec in site_spec['nodes'] ]
1078         return node_infos
1079
1080     def all_nodenames(self):
1081         return [ x[0] for x in self.all_node_infos() ]
1082     def all_reservable_nodenames(self):
1083         res = []
1084         for site_spec in self.plc_spec['sites']:
1085             for node_spec in site_spec['nodes']:
1086                 node_fields = node_spec['node_fields']
1087                 if 'node_type' in node_fields and node_fields['node_type'] == 'reservable':
1088                     res.append(node_fields['hostname'])
1089         return res
1090
1091     # silent_minutes : during the first <silent_minutes> minutes nothing gets printed
1092     def nodes_check_boot_state(self, target_boot_state, timeout_minutes,
1093                                silent_minutes, period_seconds = 15):
1094         if self.options.dry_run:
1095             print('dry_run')
1096             return True
1097
1098         class CompleterTaskBootState(CompleterTask):
1099             def __init__(self, test_plc, hostname):
1100                 self.test_plc = test_plc
1101                 self.hostname = hostname
1102                 self.last_boot_state = 'undef'
1103             def actual_run(self):
1104                 try:
1105                     node = self.test_plc.apiserver.GetNodes(self.test_plc.auth_root(),
1106                                                             [ self.hostname ],
1107                                                             ['boot_state'])[0]
1108                     self.last_boot_state = node['boot_state']
1109                     return self.last_boot_state == target_boot_state
1110                 except:
1111                     return False
1112             def message(self):
1113                 return "CompleterTaskBootState with node {}".format(self.hostname)
1114             def failure_epilogue(self):
1115                 print("node {} in state {} - expected {}"\
1116                     .format(self.hostname, self.last_boot_state, target_boot_state))
1117
1118         timeout = timedelta(minutes=timeout_minutes)
1119         graceout = timedelta(minutes=silent_minutes)
1120         period   = timedelta(seconds=period_seconds)
1121         # the nodes that haven't checked yet - start with a full list and shrink over time
1122         utils.header("checking nodes boot state (expected {})".format(target_boot_state))
1123         tasks = [ CompleterTaskBootState(self,hostname) \
1124                       for (hostname,_) in self.all_node_infos() ]
1125         message = 'check_boot_state={}'.format(target_boot_state)
1126         return Completer(tasks, message=message).run(timeout, graceout, period)
1127
1128     def nodes_booted(self):
1129         return self.nodes_check_boot_state('boot', timeout_minutes=30, silent_minutes=28)
1130
1131     def probe_kvm_iptables(self):
1132         (_,kvmbox) = self.all_node_infos()[0]
1133         TestSsh(kvmbox).run("iptables-save")
1134         return True
1135
1136     # probing nodes
1137     def check_nodes_ping(self, timeout_seconds=60, period_seconds=10):
1138         class CompleterTaskPingNode(CompleterTask):
1139             def __init__(self, hostname):
1140                 self.hostname = hostname
1141             def run(self, silent):
1142                 command="ping -c 1 -w 1 {} >& /dev/null".format(self.hostname)
1143                 return utils.system(command, silent=silent) == 0
1144             def failure_epilogue(self):
1145                 print("Cannot ping node with name {}".format(self.hostname))
1146         timeout = timedelta(seconds = timeout_seconds)
1147         graceout = timeout
1148         period = timedelta(seconds = period_seconds)
1149         node_infos = self.all_node_infos()
1150         tasks = [ CompleterTaskPingNode(h) for (h,_) in node_infos ]
1151         return Completer(tasks, message='ping_node').run(timeout, graceout, period)
1152
1153     # ping node before we try to reach ssh, helpful for troubleshooting failing bootCDs
1154     def ping_node(self):
1155         "Ping nodes"
1156         return self.check_nodes_ping()
1157
1158     def check_nodes_ssh(self, debug, timeout_minutes, silent_minutes, period_seconds=15):
1159         # various delays
1160         timeout  = timedelta(minutes=timeout_minutes)
1161         graceout = timedelta(minutes=silent_minutes)
1162         period   = timedelta(seconds=period_seconds)
1163         vservername = self.vservername
1164         if debug:
1165             message = "debug"
1166             completer_message = 'ssh_node_debug'
1167             local_key = "keys/{vservername}-debug.rsa".format(**locals())
1168         else:
1169             message = "boot"
1170             completer_message = 'ssh_node_boot'
1171             local_key = "keys/key_admin.rsa"
1172         utils.header("checking ssh access to nodes (expected in {} mode)".format(message))
1173         node_infos = self.all_node_infos()
1174         tasks = [ CompleterTaskNodeSsh(nodename, qemuname, local_key,
1175                                         boot_state=message, dry_run=self.options.dry_run) \
1176                       for (nodename, qemuname) in node_infos ]
1177         return Completer(tasks, message=completer_message).run(timeout, graceout, period)
1178
1179     def ssh_node_debug(self):
1180         "Tries to ssh into nodes in debug mode with the debug ssh key"
1181         return self.check_nodes_ssh(debug = True,
1182                                     timeout_minutes = self.ssh_node_debug_timeout,
1183                                     silent_minutes = self.ssh_node_debug_silent)
1184
1185     def ssh_node_boot(self):
1186         "Tries to ssh into nodes in production mode with the root ssh key"
1187         return self.check_nodes_ssh(debug = False,
1188                                     timeout_minutes = self.ssh_node_boot_timeout,
1189                                     silent_minutes = self.ssh_node_boot_silent)
1190
1191     def node_bmlogs(self):
1192         "Checks that there's a non-empty dir. /var/log/bm/raw"
1193         return utils.system(self.actual_command_in_guest("ls /var/log/bm/raw")) == 0
1194
1195     @node_mapper
1196     def qemu_local_init(self): pass
1197     @node_mapper
1198     def bootcd(self): pass
1199     @node_mapper
1200     def qemu_local_config(self): pass
1201     @node_mapper
1202     def qemu_export(self): pass
1203     @node_mapper
1204     def qemu_cleanlog(self): pass
1205     @node_mapper
1206     def nodestate_reinstall(self): pass
1207     @node_mapper
1208     def nodestate_upgrade(self): pass
1209     @node_mapper
1210     def nodestate_safeboot(self): pass
1211     @node_mapper
1212     def nodestate_boot(self): pass
1213     @node_mapper
1214     def nodestate_show(self): pass
1215     @node_mapper
1216     def nodedistro_f14(self): pass
1217     @node_mapper
1218     def nodedistro_f18(self): pass
1219     @node_mapper
1220     def nodedistro_f20(self): pass
1221     @node_mapper
1222     def nodedistro_f21(self): pass
1223     @node_mapper
1224     def nodedistro_f22(self): pass
1225     @node_mapper
1226     def nodedistro_show(self): pass
1227
1228     ### check hooks : invoke scripts from hooks/{node,slice}
1229     def check_hooks_node(self):
1230         return self.locate_first_node().check_hooks()
1231     def check_hooks_sliver(self) :
1232         return self.locate_first_sliver().check_hooks()
1233
1234     def check_hooks(self):
1235         "runs unit tests in the node and slice contexts - see hooks/{node,slice}"
1236         return self.check_hooks_node() and self.check_hooks_sliver()
1237
1238     ### initscripts
1239     def do_check_initscripts(self):
1240         class CompleterTaskInitscript(CompleterTask):
1241             def __init__(self, test_sliver, stamp):
1242                 self.test_sliver = test_sliver
1243                 self.stamp = stamp
1244             def actual_run(self):
1245                 return self.test_sliver.check_initscript_stamp(self.stamp)
1246             def message(self):
1247                 return "initscript checker for {}".format(self.test_sliver.name())
1248             def failure_epilogue(self):
1249                 print("initscript stamp {} not found in sliver {}"\
1250                     .format(self.stamp, self.test_sliver.name()))
1251
1252         tasks = []
1253         for slice_spec in self.plc_spec['slices']:
1254             if 'initscriptstamp' not in slice_spec:
1255                 continue
1256             stamp = slice_spec['initscriptstamp']
1257             slicename = slice_spec['slice_fields']['name']
1258             for nodename in slice_spec['nodenames']:
1259                 print('nodename', nodename, 'slicename', slicename, 'stamp', stamp)
1260                 site,node = self.locate_node(nodename)
1261                 # xxx - passing the wrong site - probably harmless
1262                 test_site = TestSite(self, site)
1263                 test_slice = TestSlice(self, test_site, slice_spec)
1264                 test_node = TestNode(self, test_site, node)
1265                 test_sliver = TestSliver(self, test_node, test_slice)
1266                 tasks.append(CompleterTaskInitscript(test_sliver, stamp))
1267         return Completer(tasks, message='check_initscripts').\
1268             run (timedelta(minutes=5), timedelta(minutes=4), timedelta(seconds=10))
1269
1270     def check_initscripts(self):
1271         "check that the initscripts have triggered"
1272         return self.do_check_initscripts()
1273
1274     def initscripts(self):
1275         "create initscripts with PLCAPI"
1276         for initscript in self.plc_spec['initscripts']:
1277             utils.pprint('Adding Initscript in plc {}'.format(self.plc_spec['name']), initscript)
1278             self.apiserver.AddInitScript(self.auth_root(), initscript['initscript_fields'])
1279         return True
1280
1281     def delete_initscripts(self):
1282         "delete initscripts with PLCAPI"
1283         for initscript in self.plc_spec['initscripts']:
1284             initscript_name = initscript['initscript_fields']['name']
1285             print(('Attempting to delete {} in plc {}'.format(initscript_name, self.plc_spec['name'])))
1286             try:
1287                 self.apiserver.DeleteInitScript(self.auth_root(), initscript_name)
1288                 print(initscript_name, 'deleted')
1289             except:
1290                 print('deletion went wrong - probably did not exist')
1291         return True
1292
1293     ### manage slices
1294     def slices(self):
1295         "create slices with PLCAPI"
1296         return self.do_slices(action="add")
1297
1298     def delete_slices(self):
1299         "delete slices with PLCAPI"
1300         return self.do_slices(action="delete")
1301
1302     def fill_slices(self):
1303         "add nodes in slices with PLCAPI"
1304         return self.do_slices(action="fill")
1305
1306     def empty_slices(self):
1307         "remove nodes from slices with PLCAPI"
1308         return self.do_slices(action="empty")
1309
1310     def do_slices(self,  action="add"):
1311         for slice in self.plc_spec['slices']:
1312             site_spec = self.locate_site(slice['sitename'])
1313             test_site = TestSite(self,site_spec)
1314             test_slice=TestSlice(self,test_site,slice)
1315             if action == "delete":
1316                 test_slice.delete_slice()
1317             elif action == "fill":
1318                 test_slice.add_nodes()
1319             elif action == "empty":
1320                 test_slice.delete_nodes()
1321             else:
1322                 test_slice.create_slice()
1323         return True
1324
1325     @slice_mapper__tasks(20, 10, 15)
1326     def ssh_slice(self): pass
1327     @slice_mapper__tasks(20, 19, 15)
1328     def ssh_slice_off(self): pass
1329     @slice_mapper__tasks(1, 1, 15)
1330     def slice_fs_present(self): pass
1331     @slice_mapper__tasks(1, 1, 15)
1332     def slice_fs_deleted(self): pass
1333
1334     # use another name so we can exclude/ignore it from the tests on the nightly command line
1335     def ssh_slice_again(self): return self.ssh_slice()
1336     # note that simply doing ssh_slice_again=ssh_slice would kind of work too
1337     # but for some reason the ignore-wrapping thing would not
1338
1339     @slice_mapper
1340     def ssh_slice_basics(self): pass
1341     @slice_mapper
1342     def check_vsys_defaults(self): pass
1343
1344     @node_mapper
1345     def keys_clear_known_hosts(self): pass
1346
1347     def plcapi_urls(self):
1348         """
1349         attempts to reach the PLCAPI with various forms for the URL
1350         """
1351         return PlcapiUrlScanner(self.auth_root(), ip=self.vserverip).scan()
1352
1353     def speed_up_slices(self):
1354         "tweak nodemanager cycle (wait time) to 30+/-10 s"
1355         return self._speed_up_slices (30, 10)
1356     def super_speed_up_slices(self):
1357         "dev mode: tweak nodemanager cycle (wait time) to 5+/-1 s"
1358         return self._speed_up_slices(5, 1)
1359
1360     def _speed_up_slices(self, p, r):
1361         # create the template on the server-side
1362         template = "{}.nodemanager".format(self.name())
1363         with open(template,"w") as template_file:
1364             template_file.write('OPTIONS="-p {} -r {} -d"\n'.format(p, r))
1365         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1366         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1367         self.test_ssh.copy_abs(template, remote)
1368         # Add a conf file
1369         if not self.apiserver.GetConfFiles(self.auth_root(),
1370                                            {'dest' : '/etc/sysconfig/nodemanager'}):
1371             self.apiserver.AddConfFile(self.auth_root(),
1372                                         {'dest' : '/etc/sysconfig/nodemanager',
1373                                          'source' : 'PlanetLabConf/nodemanager',
1374                                          'postinstall_cmd' : 'service nm restart',})
1375         return True
1376
1377     def debug_nodemanager(self):
1378         "sets verbose mode for nodemanager, and speeds up cycle even more (needs speed_up_slices first)"
1379         template = "{}.nodemanager".format(self.name())
1380         with open(template,"w") as template_file:
1381             template_file.write('OPTIONS="-p 10 -r 6 -v -d"\n')
1382         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1383         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1384         self.test_ssh.copy_abs(template, remote)
1385         return True
1386
1387     @node_mapper
1388     def qemu_start(self) : pass
1389
1390     @node_mapper
1391     def qemu_timestamp(self) : pass
1392
1393     @node_mapper
1394     def qemu_nodefamily(self): pass
1395
1396     # when a spec refers to a node possibly on another plc
1397     def locate_sliver_obj_cross(self, nodename, slicename, other_plcs):
1398         for plc in [ self ] + other_plcs:
1399             try:
1400                 return plc.locate_sliver_obj(nodename, slicename)
1401             except:
1402                 pass
1403         raise Exception("Cannot locate sliver {}@{} among all PLCs".format(nodename, slicename))
1404
1405     # implement this one as a cross step so that we can take advantage of different nodes
1406     # in multi-plcs mode
1407     def cross_check_tcp(self, other_plcs):
1408         "check TCP connectivity between 2 slices (or in loopback if only one is defined)"
1409         if 'tcp_specs' not in self.plc_spec or not self.plc_spec['tcp_specs']:
1410             utils.header("check_tcp: no/empty config found")
1411             return True
1412         specs = self.plc_spec['tcp_specs']
1413         overall = True
1414
1415         # first wait for the network to be up and ready from the slices
1416         class CompleterTaskNetworkReadyInSliver(CompleterTask):
1417             def __init__(self, test_sliver):
1418                 self.test_sliver = test_sliver
1419             def actual_run(self):
1420                 return self.test_sliver.check_tcp_ready(port = 9999)
1421             def message(self):
1422                 return "network ready checker for {}".format(self.test_sliver.name())
1423             def failure_epilogue(self):
1424                 print("could not bind port from sliver {}".format(self.test_sliver.name()))
1425
1426         sliver_specs = {}
1427         tasks = []
1428         managed_sliver_names = set()
1429         for spec in specs:
1430             # locate the TestSliver instances involved, and cache them in the spec instance
1431             spec['s_sliver'] = self.locate_sliver_obj_cross(spec['server_node'], spec['server_slice'], other_plcs)
1432             spec['c_sliver'] = self.locate_sliver_obj_cross(spec['client_node'], spec['client_slice'], other_plcs)
1433             message = "Will check TCP between s={} and c={}"\
1434                       .format(spec['s_sliver'].name(), spec['c_sliver'].name())
1435             if 'client_connect' in spec:
1436                 message += " (using {})".format(spec['client_connect'])
1437             utils.header(message)
1438             # we need to check network presence in both slivers, but also
1439             # avoid to insert a sliver several times
1440             for sliver in [ spec['s_sliver'], spec['c_sliver'] ]:
1441                 if sliver.name() not in managed_sliver_names:
1442                     tasks.append(CompleterTaskNetworkReadyInSliver(sliver))
1443                     # add this sliver's name in the set
1444                     managed_sliver_names .update( {sliver.name()} )
1445
1446         # wait for the netork to be OK in all server sides
1447         if not Completer(tasks, message='check for network readiness in slivers').\
1448            run(timedelta(seconds=30), timedelta(seconds=24), period=timedelta(seconds=5)):
1449             return False
1450
1451         # run server and client
1452         for spec in specs:
1453             port = spec['port']
1454             # server side
1455             # the issue here is that we have the server run in background
1456             # and so we have no clue if it took off properly or not
1457             # looks like in some cases it does not
1458             address = spec['s_sliver'].test_node.name()
1459             if not spec['s_sliver'].run_tcp_server(address, port, timeout=20):
1460                 overall = False
1461                 break
1462
1463             # idem for the client side
1464             # use nodename from located sliver, unless 'client_connect' is set
1465             if 'client_connect' in spec:
1466                 destination = spec['client_connect']
1467             else:
1468                 destination = spec['s_sliver'].test_node.name()
1469             if not spec['c_sliver'].run_tcp_client(destination, port):
1470                 overall = False
1471         return overall
1472
1473     # painfully enough, we need to allow for some time as netflow might show up last
1474     def check_system_slice(self):
1475         "all nodes: check that a system slice is alive"
1476         # netflow currently not working in the lxc distro
1477         # drl not built at all in the wtx distro
1478         # if we find either of them we're happy
1479         return self.check_netflow() or self.check_drl()
1480
1481     # expose these
1482     def check_netflow(self): return self._check_system_slice('netflow')
1483     def check_drl(self): return self._check_system_slice('drl')
1484
1485     # we have the slices up already here, so it should not take too long
1486     def _check_system_slice(self, slicename, timeout_minutes=5, period_seconds=15):
1487         class CompleterTaskSystemSlice(CompleterTask):
1488             def __init__(self, test_node, dry_run):
1489                 self.test_node = test_node
1490                 self.dry_run = dry_run
1491             def actual_run(self):
1492                 return self.test_node._check_system_slice(slicename, dry_run=self.dry_run)
1493             def message(self):
1494                 return "System slice {} @ {}".format(slicename, self.test_node.name())
1495             def failure_epilogue(self):
1496                 print("COULD not find system slice {} @ {}".format(slicename, self.test_node.name()))
1497         timeout = timedelta(minutes=timeout_minutes)
1498         silent  = timedelta(0)
1499         period  = timedelta(seconds=period_seconds)
1500         tasks = [ CompleterTaskSystemSlice(test_node, self.options.dry_run) \
1501                       for test_node in self.all_nodes() ]
1502         return Completer(tasks, message='_check_system_slice').run(timeout, silent, period)
1503
1504     def plcsh_stress_test(self):
1505         "runs PLCAPI stress test, that checks Add/Update/Delete on all types - preserves contents"
1506         # install the stress-test in the plc image
1507         location = "/usr/share/plc_api/plcsh_stress_test.py"
1508         remote = "{}/{}".format(self.vm_root_in_host(), location)
1509         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1510         command = location
1511         command += " -- --check"
1512         if self.options.size == 1:
1513             command +=  " --tiny"
1514         return self.run_in_guest(command) == 0
1515
1516     # populate runs the same utility without slightly different options
1517     # in particular runs with --preserve (dont cleanup) and without --check
1518     # also it gets run twice, once with the --foreign option for creating fake foreign entries
1519
1520     def sfa_install_all(self):
1521         "yum install sfa sfa-plc sfa-sfatables sfa-client"
1522         return (self.dnf_install("sfa sfa-plc sfa-sfatables sfa-client") and
1523                 self.run_in_guest("systemctl enable sfa-registry")==0 and
1524                 self.run_in_guest("systemctl enable sfa-aggregate")==0)
1525
1526     def sfa_install_core(self):
1527         "yum install sfa"
1528         return self.dnf_install("sfa")
1529
1530     def sfa_install_plc(self):
1531         "yum install sfa-plc"
1532         return self.dnf_install("sfa-plc")
1533
1534     def sfa_install_sfatables(self):
1535         "yum install sfa-sfatables"
1536         return self.dnf_install("sfa-sfatables")
1537
1538     # for some very odd reason, this sometimes fails with the following symptom
1539     # # yum install sfa-client
1540     # Setting up Install Process
1541     # ...
1542     # Downloading Packages:
1543     # Running rpm_check_debug
1544     # Running Transaction Test
1545     # Transaction Test Succeeded
1546     # Running Transaction
1547     # Transaction couldn't start:
1548     # installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem
1549     # [('installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem', (9, '/', 69632L))]
1550     # even though in the same context I have
1551     # [2012.05.23--f14-32-sfastd1-1-vplc07] / # df -h
1552     # Filesystem            Size  Used Avail Use% Mounted on
1553     # /dev/hdv1             806G  264G  501G  35% /
1554     # none                   16M   36K   16M   1% /tmp
1555     #
1556     # so as a workaround, we first try yum install, and then invoke rpm on the cached rpm...
1557     def sfa_install_client(self):
1558         "yum install sfa-client"
1559         first_try = self.dnf_install("sfa-client")
1560         if first_try:
1561             return True
1562         utils.header("********** Regular yum failed - special workaround in place, 2nd chance")
1563         code, cached_rpm_path = \
1564                 utils.output_of(self.actual_command_in_guest('find /var/cache/yum -name sfa-client\*.rpm'))
1565         utils.header("rpm_path=<<{}>>".format(rpm_path))
1566         # just for checking
1567         self.run_in_guest("rpm -i {}".format(cached_rpm_path))
1568         return self.dnf_check_installed("sfa-client")
1569
1570     def sfa_dbclean(self):
1571         "thoroughly wipes off the SFA database"
1572         return self.run_in_guest("sfaadmin reg nuke") == 0 or \
1573             self.run_in_guest("sfa-nuke.py") == 0 or \
1574             self.run_in_guest("sfa-nuke-plc.py") == 0 or \
1575             self.run_in_guest("sfaadmin registry nuke") == 0
1576
1577     def sfa_fsclean(self):
1578         "cleanup /etc/sfa/trusted_roots and /var/lib/sfa"
1579         self.run_in_guest("rm -rf /etc/sfa/trusted_roots /var/lib/sfa/authorities")
1580         return True
1581
1582     def sfa_plcclean(self):
1583         "cleans the PLC entries that were created as a side effect of running the script"
1584         # ignore result
1585         sfa_spec = self.plc_spec['sfa']
1586
1587         for auth_sfa_spec in sfa_spec['auth_sfa_specs']:
1588             login_base = auth_sfa_spec['login_base']
1589             try:
1590                 self.apiserver.DeleteSite(self.auth_root(),login_base)
1591             except:
1592                 print("Site {} already absent from PLC db".format(login_base))
1593
1594             for spec_name in ['pi_spec', 'user_spec']:
1595                 user_spec = auth_sfa_spec[spec_name]
1596                 username = user_spec['email']
1597                 try:
1598                     self.apiserver.DeletePerson(self.auth_root(),username)
1599                 except:
1600                     # this in fact is expected as sites delete their members
1601                     #print "User {} already absent from PLC db".format(username)
1602                     pass
1603
1604         print("REMEMBER TO RUN sfa_import AGAIN")
1605         return True
1606
1607     def sfa_uninstall(self):
1608         "uses rpm to uninstall sfa - ignore result"
1609         self.run_in_guest("rpm -e sfa sfa-sfatables sfa-client sfa-plc")
1610         self.run_in_guest("rm -rf /var/lib/sfa")
1611         self.run_in_guest("rm -rf /etc/sfa")
1612         self.run_in_guest("rm -rf /var/log/sfa_access.log /var/log/sfa_import_plc.log /var/log/sfa.daemon")
1613         # xxx tmp
1614         self.run_in_guest("rpm -e --noscripts sfa-plc")
1615         return True
1616
1617     ### run unit tests for SFA
1618     # NOTE: for some reason on f14/i386, yum install sfa-tests fails for no reason
1619     # Running Transaction
1620     # Transaction couldn't start:
1621     # installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem
1622     # [('installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem', (9, '/', 208896L))]
1623     # no matter how many Gbs are available on the testplc
1624     # could not figure out what's wrong, so...
1625     # if the yum install phase fails, consider the test is successful
1626     # other combinations will eventually run it hopefully
1627     def sfa_utest(self):
1628         "dnf install sfa-tests and run SFA unittests"
1629         self.run_in_guest("dnf -y install sfa-tests")
1630         # failed to install - forget it
1631         if self.run_in_guest("rpm -q sfa-tests") != 0:
1632             utils.header("WARNING: SFA unit tests failed to install, ignoring")
1633             return True
1634         return self.run_in_guest("/usr/share/sfa/tests/testAll.py") == 0
1635
1636     ###
1637     def confdir(self):
1638         dirname = "conf.{}".format(self.plc_spec['name'])
1639         if not os.path.isdir(dirname):
1640             utils.system("mkdir -p {}".format(dirname))
1641         if not os.path.isdir(dirname):
1642             raise Exception("Cannot create config dir for plc {}".format(self.name()))
1643         return dirname
1644
1645     def conffile(self, filename):
1646         return "{}/{}".format(self.confdir(), filename)
1647     def confsubdir(self, dirname, clean, dry_run=False):
1648         subdirname = "{}/{}".format(self.confdir(), dirname)
1649         if clean:
1650             utils.system("rm -rf {}".format(subdirname))
1651         if not os.path.isdir(subdirname):
1652             utils.system("mkdir -p {}".format(subdirname))
1653         if not dry_run and not os.path.isdir(subdirname):
1654             raise "Cannot create config subdir {} for plc {}".format(dirname, self.name())
1655         return subdirname
1656
1657     def conffile_clean(self, filename):
1658         filename=self.conffile(filename)
1659         return utils.system("rm -rf {}".format(filename))==0
1660
1661     ###
1662     def sfa_configure(self):
1663         "run sfa-config-tty"
1664         tmpname = self.conffile("sfa-config-tty")
1665         with open(tmpname,'w') as fileconf:
1666             for var, value in self.plc_spec['sfa']['settings'].items():
1667                 fileconf.write('e {}\n{}\n'.format(var, value))
1668             fileconf.write('w\n')
1669             fileconf.write('R\n')
1670             fileconf.write('q\n')
1671         utils.system('cat {}'.format(tmpname))
1672         self.run_in_guest_piped('cat {}'.format(tmpname), 'sfa-config-tty')
1673         return True
1674
1675     def aggregate_xml_line(self):
1676         port = self.plc_spec['sfa']['neighbours-port']
1677         return '<aggregate addr="{}" hrn="{}" port="{}"/>'\
1678             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'], port)
1679
1680     def registry_xml_line(self):
1681         return '<registry addr="{}" hrn="{}" port="12345"/>'\
1682             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'])
1683
1684
1685     # a cross step that takes all other plcs in argument
1686     def cross_sfa_configure(self, other_plcs):
1687         "writes aggregates.xml and registries.xml that point to all other PLCs in the test"
1688         # of course with a single plc, other_plcs is an empty list
1689         if not other_plcs:
1690             return True
1691         agg_fname = self.conffile("agg.xml")
1692         with open(agg_fname,"w") as out:
1693             out.write("<aggregates>{}</aggregates>\n"\
1694                       .format(" ".join([ plc.aggregate_xml_line() for plc in other_plcs ])))
1695         utils.header("(Over)wrote {}".format(agg_fname))
1696         reg_fname=self.conffile("reg.xml")
1697         with open(reg_fname,"w") as out:
1698             out.write("<registries>{}</registries>\n"\
1699                       .format(" ".join([ plc.registry_xml_line() for plc in other_plcs ])))
1700         utils.header("(Over)wrote {}".format(reg_fname))
1701         return self.test_ssh.copy_abs(agg_fname,
1702                                       '/{}/etc/sfa/aggregates.xml'.format(self.vm_root_in_host())) == 0 \
1703            and self.test_ssh.copy_abs(reg_fname,
1704                                       '/{}/etc/sfa/registries.xml'.format(self.vm_root_in_host())) == 0
1705
1706     def sfa_import(self):
1707         "use sfaadmin to import from plc"
1708         auth = self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH']
1709         return self.run_in_guest('sfaadmin reg import_registry') == 0
1710
1711     def sfa_start(self):
1712         "start SFA through systemctl"
1713         return (self.start_stop_systemd('sfa-registry', 'start') and
1714                 self.start_stop_systemd('sfa-aggregate', 'start'))
1715
1716
1717     def sfi_configure(self):
1718         "Create /root/sfi on the plc side for sfi client configuration"
1719         if self.options.dry_run:
1720             utils.header("DRY RUN - skipping step")
1721             return True
1722         sfa_spec = self.plc_spec['sfa']
1723         # cannot use auth_sfa_mapper to pass dir_name
1724         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1725             test_slice = TestAuthSfa(self, slice_spec)
1726             dir_basename = os.path.basename(test_slice.sfi_path())
1727             dir_name = self.confsubdir("dot-sfi/{}".format(dir_basename),
1728                                        clean=True, dry_run=self.options.dry_run)
1729             test_slice.sfi_configure(dir_name)
1730             # push into the remote /root/sfi area
1731             location = test_slice.sfi_path()
1732             remote = "{}/{}".format(self.vm_root_in_host(), location)
1733             self.test_ssh.mkdir(remote, abs=True)
1734             # need to strip last level or remote otherwise we get an extra dir level
1735             self.test_ssh.copy_abs(dir_name, os.path.dirname(remote), recursive=True)
1736
1737         return True
1738
1739     def sfi_clean(self):
1740         "clean up /root/sfi on the plc side"
1741         self.run_in_guest("rm -rf /root/sfi")
1742         return True
1743
1744     def sfa_rspec_empty(self):
1745         "expose a static empty rspec (ships with the tests module) in the sfi directory"
1746         filename = "empty-rspec.xml"
1747         overall = True
1748         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1749             test_slice = TestAuthSfa(self, slice_spec)
1750             in_vm = test_slice.sfi_path()
1751             remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1752             if self.test_ssh.copy_abs(filename, remote) !=0:
1753                 overall = False
1754         return overall
1755
1756     @auth_sfa_mapper
1757     def sfa_register_site(self): pass
1758     @auth_sfa_mapper
1759     def sfa_register_pi(self): pass
1760     @auth_sfa_mapper
1761     def sfa_register_user(self): pass
1762     @auth_sfa_mapper
1763     def sfa_update_user(self): pass
1764     @auth_sfa_mapper
1765     def sfa_register_slice(self): pass
1766     @auth_sfa_mapper
1767     def sfa_renew_slice(self): pass
1768     @auth_sfa_mapper
1769     def sfa_get_expires(self): pass
1770     @auth_sfa_mapper
1771     def sfa_discover(self): pass
1772     @auth_sfa_mapper
1773     def sfa_rspec(self): pass
1774     @auth_sfa_mapper
1775     def sfa_allocate(self): pass
1776     @auth_sfa_mapper
1777     def sfa_allocate_empty(self): pass
1778     @auth_sfa_mapper
1779     def sfa_provision(self): pass
1780     @auth_sfa_mapper
1781     def sfa_provision_empty(self): pass
1782     @auth_sfa_mapper
1783     def sfa_describe(self): pass
1784     @auth_sfa_mapper
1785     def sfa_check_slice_plc(self): pass
1786     @auth_sfa_mapper
1787     def sfa_check_slice_plc_empty(self): pass
1788     @auth_sfa_mapper
1789     def sfa_update_slice(self): pass
1790     @auth_sfa_mapper
1791     def sfa_remove_user_from_slice(self): pass
1792     @auth_sfa_mapper
1793     def sfa_insert_user_in_slice(self): pass
1794     @auth_sfa_mapper
1795     def sfi_list(self): pass
1796     @auth_sfa_mapper
1797     def sfi_show_site(self): pass
1798     @auth_sfa_mapper
1799     def sfi_show_slice(self): pass
1800     @auth_sfa_mapper
1801     def sfi_show_slice_researchers(self): pass
1802     @auth_sfa_mapper
1803     def ssh_slice_sfa(self): pass
1804     @auth_sfa_mapper
1805     def sfa_delete_user(self): pass
1806     @auth_sfa_mapper
1807     def sfa_delete_slice(self): pass
1808
1809     def sfa_stop(self):
1810         "stop sfa through systemclt"
1811         return (self.start_stop_systemd('sfa-aggregate', 'stop') and
1812                 self.start_stop_systemd('sfa-registry', 'stop'))
1813
1814     def populate(self):
1815         "creates random entries in the PLCAPI"
1816         # install the stress-test in the plc image
1817         location = "/usr/share/plc_api/plcsh_stress_test.py"
1818         remote = "{}/{}".format(self.vm_root_in_host(), location)
1819         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1820         command = location
1821         command += " -- --preserve --short-names"
1822         local = (self.run_in_guest(command) == 0);
1823         # second run with --foreign
1824         command += ' --foreign'
1825         remote = (self.run_in_guest(command) == 0);
1826         return local and remote
1827
1828
1829     ####################
1830     @bonding_redirector
1831     def bonding_init_partial(self): pass
1832
1833     @bonding_redirector
1834     def bonding_add_yum(self): pass
1835
1836     @bonding_redirector
1837     def bonding_install_rpms(self): pass
1838
1839     ####################
1840
1841     def gather_logs(self):
1842         "gets all possible logs from plc's/qemu node's/slice's for future reference"
1843         # (1.a) get the plc's /var/log/ and store it locally in logs/myplc.var-log.<plcname>/*
1844         # (1.b) get the plc's  /var/lib/pgsql/data/pg_log/ -> logs/myplc.pgsql-log.<plcname>/*
1845         # (1.c) get the plc's /root/sfi -> logs/sfi.<plcname>/
1846         # (2) get all the nodes qemu log and store it as logs/node.qemu.<node>.log
1847         # (3) get the nodes /var/log and store is as logs/node.var-log.<node>/*
1848         # (4) as far as possible get the slice's /var/log as logs/sliver.var-log.<sliver>/*
1849         # (1.a)
1850         print("-------------------- TestPlc.gather_logs : PLC's /var/log")
1851         self.gather_var_logs()
1852         # (1.b)
1853         print("-------------------- TestPlc.gather_logs : PLC's /var/lib/psql/data/pg_log/")
1854         self.gather_pgsql_logs()
1855         # (1.c)
1856         print("-------------------- TestPlc.gather_logs : PLC's /root/sfi/")
1857         self.gather_root_sfi()
1858         # (2)
1859         print("-------------------- TestPlc.gather_logs : nodes's QEMU logs")
1860         for site_spec in self.plc_spec['sites']:
1861             test_site = TestSite(self,site_spec)
1862             for node_spec in site_spec['nodes']:
1863                 test_node = TestNode(self, test_site, node_spec)
1864                 test_node.gather_qemu_logs()
1865         # (3)
1866         print("-------------------- TestPlc.gather_logs : nodes's /var/log")
1867         self.gather_nodes_var_logs()
1868         # (4)
1869         print("-------------------- TestPlc.gather_logs : sample sliver's /var/log")
1870         self.gather_slivers_var_logs()
1871         return True
1872
1873     def gather_slivers_var_logs(self):
1874         for test_sliver in self.all_sliver_objs():
1875             remote = test_sliver.tar_var_logs()
1876             utils.system("mkdir -p logs/sliver.var-log.{}".format(test_sliver.name()))
1877             command = remote + " | tar -C logs/sliver.var-log.{} -xf -".format(test_sliver.name())
1878             utils.system(command)
1879         return True
1880
1881     def gather_var_logs(self):
1882         utils.system("mkdir -p logs/myplc.var-log.{}".format(self.name()))
1883         to_plc = self.actual_command_in_guest("tar -C /var/log/ -cf - .")
1884         command = to_plc + "| tar -C logs/myplc.var-log.{} -xf -".format(self.name())
1885         utils.system(command)
1886         command = "chmod a+r,a+x logs/myplc.var-log.{}/httpd".format(self.name())
1887         utils.system(command)
1888
1889     def gather_pgsql_logs(self):
1890         utils.system("mkdir -p logs/myplc.pgsql-log.{}".format(self.name()))
1891         to_plc = self.actual_command_in_guest("tar -C /var/lib/pgsql/data/pg_log/ -cf - .")
1892         command = to_plc + "| tar -C logs/myplc.pgsql-log.{} -xf -".format(self.name())
1893         utils.system(command)
1894
1895     def gather_root_sfi(self):
1896         utils.system("mkdir -p logs/sfi.{}".format(self.name()))
1897         to_plc = self.actual_command_in_guest("tar -C /root/sfi/ -cf - .")
1898         command = to_plc + "| tar -C logs/sfi.{} -xf -".format(self.name())
1899         utils.system(command)
1900
1901     def gather_nodes_var_logs(self):
1902         for site_spec in self.plc_spec['sites']:
1903             test_site = TestSite(self, site_spec)
1904             for node_spec in site_spec['nodes']:
1905                 test_node = TestNode(self, test_site, node_spec)
1906                 test_ssh = TestSsh(test_node.name(), key="keys/key_admin.rsa")
1907                 command = test_ssh.actual_command("tar -C /var/log -cf - .")
1908                 command = command + "| tar -C logs/node.var-log.{} -xf -".format(test_node.name())
1909                 utils.system("mkdir -p logs/node.var-log.{}".format(test_node.name()))
1910                 utils.system(command)
1911
1912
1913     # returns the filename to use for sql dump/restore, using options.dbname if set
1914     def dbfile(self, database):
1915         # uses options.dbname if it is found
1916         try:
1917             name = self.options.dbname
1918             if not isinstance(name, str):
1919                 raise Exception
1920         except:
1921             t = datetime.now()
1922             d = t.date()
1923             name = str(d)
1924         return "/root/{}-{}.sql".format(database, name)
1925
1926     def plc_db_dump(self):
1927         'dump the planetlab5 DB in /root in the PLC - filename has time'
1928         dump=self.dbfile("planetab5")
1929         self.run_in_guest('pg_dump -U pgsqluser planetlab5 -f '+ dump)
1930         utils.header('Dumped planetlab5 database in {}'.format(dump))
1931         return True
1932
1933     def plc_db_restore(self):
1934         'restore the planetlab5 DB - looks broken, but run -n might help'
1935         dump = self.dbfile("planetab5")
1936         self.run_in_guest('systemctl stop httpd')
1937         # xxx - need another wrapper
1938         self.run_in_guest_piped('echo drop database planetlab5', 'psql --user=pgsqluser template1')
1939         self.run_in_guest('createdb -U postgres --encoding=UNICODE --owner=pgsqluser planetlab5')
1940         self.run_in_guest('psql -U pgsqluser planetlab5 -f ' + dump)
1941         ##starting httpd service
1942         self.run_in_guest('systemctl start httpd')
1943
1944         utils.header('Database restored from ' + dump)
1945
1946     @staticmethod
1947     def create_ignore_steps():
1948         for step in TestPlc.default_steps + TestPlc.other_steps:
1949             # default step can have a plc qualifier
1950             if '@' in step:
1951                 step, qualifier = step.split('@')
1952             # or be defined as forced or ignored by default
1953             for keyword in ['_ignore', '_force']:
1954                 if step.endswith(keyword):
1955                     step=step.replace(keyword,'')
1956             if step == SEP or step == SEPSFA :
1957                 continue
1958             method = getattr(TestPlc,step)
1959             name = step + '_ignore'
1960             wrapped = ignore_result(method)
1961 #            wrapped.__doc__ = method.__doc__ + " (run in ignore-result mode)"
1962             setattr(TestPlc, name, wrapped)
1963
1964 #    @ignore_result
1965 #    def ssh_slice_again_ignore (self): pass
1966 #    @ignore_result
1967 #    def check_initscripts_ignore (self): pass
1968
1969     def standby_1_through_20(self):
1970         """convenience function to wait for a specified number of minutes"""
1971         pass
1972     @standby_generic
1973     def standby_1(): pass
1974     @standby_generic
1975     def standby_2(): pass
1976     @standby_generic
1977     def standby_3(): pass
1978     @standby_generic
1979     def standby_4(): pass
1980     @standby_generic
1981     def standby_5(): pass
1982     @standby_generic
1983     def standby_6(): pass
1984     @standby_generic
1985     def standby_7(): pass
1986     @standby_generic
1987     def standby_8(): pass
1988     @standby_generic
1989     def standby_9(): pass
1990     @standby_generic
1991     def standby_10(): pass
1992     @standby_generic
1993     def standby_11(): pass
1994     @standby_generic
1995     def standby_12(): pass
1996     @standby_generic
1997     def standby_13(): pass
1998     @standby_generic
1999     def standby_14(): pass
2000     @standby_generic
2001     def standby_15(): pass
2002     @standby_generic
2003     def standby_16(): pass
2004     @standby_generic
2005     def standby_17(): pass
2006     @standby_generic
2007     def standby_18(): pass
2008     @standby_generic
2009     def standby_19(): pass
2010     @standby_generic
2011     def standby_20(): pass
2012
2013     # convenience for debugging the test logic
2014     def yes(self): return True
2015     def no(self): return False
2016     def fail(self): return False