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 # ss # slices created under plcsh interactively seem to be fine but these ones don't have the tags
 161 # ss # keep this out of the way for now
 162 # ss         'check_vsys_defaults_ignore', SEP,
 163 # ss # run this first off so it's easier to re-run on another qemu box
 164 # ss         'qemu_kill_mine', 'nodestate_reinstall', 'qemu_local_init',
 165 # ss         'bootcd', 'qemu_local_config', SEP,
 166 # ss         'qemu_clean_mine', 'qemu_export', 'qemu_cleanlog', SEP,
 167 # ss         '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 # ss        'probe_kvm_iptables',
 182 # ss        'ping_node', 'ssh_node_debug', 'plcsh_stress_test@1', SEP,
 183 # ss        'ssh_node_boot', 'node_bmlogs', 'ssh_slice', 'ssh_slice_basics', SEP,
 184 # ss        '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 # ss        '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 # ss        'debug_nodemanager',
 193 # ss        'empty_slices', 'ssh_slice_off', 'slice_fs_deleted_ignore', SEP,
 194 # ss        # check they are properly re-created with the same name
 195 # ss        '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         # used to be part of default steps but won't work since f27
 216         'cross_check_tcp@1',
 217         'slice_fs_present', 'check_initscripts', SEP,
 218         'standby_1_through_20', 'yes', 'no', SEP,
 219         'install_syslinux6', 'bonding_builds', 'bonding_nodes', SEP,
 220         ]
 221     default_bonding_steps = [
 222         'bonding_init_partial',
 223         'bonding_add_yum',
 224         'bonding_install_rpms', SEP,
 225         ]
 226
 227     @staticmethod
 228     def printable_steps(list):
 229         single_line = " ".join(list) + " "
 230         return single_line.replace(" "+SEP+" ", " \\\n").replace(" "+SEPSFA+" ", " \\\n")
 231     @staticmethod
 232     def valid_step(step):
 233         return step != SEP and step != SEPSFA
 234
 235     # turn off the sfa-related steps when build has skipped SFA
 236     # this was originally for centos5 but is still valid
 237     # for up to f12 as recent SFAs with sqlalchemy won't build before f14
 238     @staticmethod
 239     def _has_sfa_cached(rpms_url):
 240         if os.path.isfile(has_sfa_cache_filename):
 241             with open(has_sfa_cache_filename) as cache:
 242                 cached = cache.read() == "yes"
 243             utils.header("build provides SFA (cached):{}".format(cached))
 244             return cached
 245         # warning, we're now building 'sface' so let's be a bit more picky
 246         # full builds are expected to return with 0 here
 247         utils.header("Checking if build provides SFA package...")
 248         retcod = utils.system("curl --silent {}/ | grep -q sfa-".format(rpms_url)) == 0
 249         encoded = 'yes' if retcod else 'no'
 250         with open(has_sfa_cache_filename,'w') as cache:
 251             cache.write(encoded)
 252         return retcod
 253
 254     @staticmethod
 255     def check_whether_build_has_sfa(rpms_url):
 256         has_sfa = TestPlc._has_sfa_cached(rpms_url)
 257         if has_sfa:
 258             utils.header("build does provide SFA")
 259         else:
 260             # move all steps containing 'sfa' from default_steps to other_steps
 261             utils.header("SFA package not found - removing steps with sfa or sfi")
 262             sfa_steps = [ step for step in TestPlc.default_steps
 263                           if step.find('sfa') >= 0 or step.find("sfi") >= 0 ]
 264             TestPlc.other_steps += sfa_steps
 265             for step in sfa_steps:
 266                 TestPlc.default_steps.remove(step)
 267
 268     def __init__(self, plc_spec, options):
 269         self.plc_spec = plc_spec
 270         self.options = options
 271         self.test_ssh = TestSsh(self.plc_spec['host_box'], self.options.buildname)
 272         self.vserverip = plc_spec['vserverip']
 273         self.vservername = plc_spec['vservername']
 274         self.vplchostname = self.vservername.split('-')[-1]
 275         self.url = "https://{}:443/PLCAPI/".format(plc_spec['vserverip'])
 276         self.apiserver = TestApiserver(self.url, options.dry_run)
 277         (self.ssh_node_boot_timeout, self.ssh_node_boot_silent) = plc_spec['ssh_node_boot_timers']
 278         (self.ssh_node_debug_timeout, self.ssh_node_debug_silent) = plc_spec['ssh_node_debug_timers']
 279
 280     def has_addresses_api(self):
 281         return self.apiserver.has_method('AddIpAddress')
 282
 283     def name(self):
 284         name = self.plc_spec['name']
 285         return "{}.{}".format(name,self.vservername)
 286
 287     def hostname(self):
 288         return self.plc_spec['host_box']
 289
 290     def is_local(self):
 291         return self.test_ssh.is_local()
 292
 293     # define the API methods on this object through xmlrpc
 294     # would help, but not strictly necessary
 295     def connect(self):
 296         pass
 297
 298     def actual_command_in_guest(self,command, backslash=False):
 299         raw1 = self.host_to_guest(command)
 300         raw2 = self.test_ssh.actual_command(raw1, dry_run=self.options.dry_run, backslash=backslash)
 301         return raw2
 302
 303     def start_guest(self):
 304       return utils.system(self.test_ssh.actual_command(self.start_guest_in_host(),
 305                                                        dry_run=self.options.dry_run))
 306
 307     def stop_guest(self):
 308       return utils.system(self.test_ssh.actual_command(self.stop_guest_in_host(),
 309                                                        dry_run=self.options.dry_run))
 310
 311     def run_in_guest(self, command, backslash=False):
 312         raw = self.actual_command_in_guest(command, backslash)
 313         return utils.system(raw)
 314
 315     def run_in_host(self,command):
 316         return self.test_ssh.run_in_buildname(command, dry_run=self.options.dry_run)
 317
 318     # backslashing turned out so awful at some point that I've turned off auto-backslashing
 319     # see e.g. plc_start esp. the version for f14
 320     #command gets run in the plc's vm
 321     def host_to_guest(self, command):
 322         ssh_leg = TestSsh(self.vplchostname)
 323         return ssh_leg.actual_command(command, keep_stdin=True)
 324
 325     # this /vservers thing is legacy...
 326     def vm_root_in_host(self):
 327         return "/vservers/{}/".format(self.vservername)
 328
 329     def vm_timestamp_path(self):
 330         return "/vservers/{}/{}.timestamp".format(self.vservername, self.vservername)
 331
 332     #start/stop the vserver
 333     def start_guest_in_host(self):
 334         return "virsh -c lxc:/// start {}".format(self.vservername)
 335
 336     def stop_guest_in_host(self):
 337         return "virsh -c lxc:/// destroy {}".format(self.vservername)
 338
 339     # xxx quick n dirty
 340     def run_in_guest_piped(self,local,remote):
 341         return utils.system(local+" | "+self.test_ssh.actual_command(self.host_to_guest(remote),
 342                                                                      keep_stdin = True))
 343
 344     def dnf_check_installed(self, rpms):
 345         if isinstance(rpms, list):
 346             rpms=" ".join(rpms)
 347         return self.run_in_guest("rpm -q {}".format(rpms)) == 0
 348
 349     # does a yum install in the vs, ignore yum retcod, check with rpm
 350     def dnf_install(self, rpms):
 351         if isinstance(rpms, list):
 352             rpms=" ".join(rpms)
 353         yum_mode = self.run_in_guest("dnf -y install {}".format(rpms))
 354         if yum_mode != 0:
 355             self.run_in_guest("dnf -y install --allowerasing {}".format(rpms))
 356         # yum-complete-transaction comes with yum-utils, that is in vtest.pkgs
 357         # nothing similar with dnf, forget about this for now
 358         # self.run_in_guest("yum-complete-transaction -y")
 359         return self.dnf_check_installed(rpms)
 360
 361     def pip3_install(self, package):
 362         return self.run_in_guest("pip3 install {}".format(package)) == 0
 363
 364     def auth_root(self):
 365         return {'Username'   : self.plc_spec['settings']['PLC_ROOT_USER'],
 366                 'AuthMethod' : 'password',
 367                 'AuthString' : self.plc_spec['settings']['PLC_ROOT_PASSWORD'],
 368                 'Role'       : self.plc_spec['role'],
 369                 }
 370
 371     def locate_site(self,sitename):
 372         for site in self.plc_spec['sites']:
 373             if site['site_fields']['name'] == sitename:
 374                 return site
 375             if site['site_fields']['login_base'] == sitename:
 376                 return site
 377         raise Exception("Cannot locate site {}".format(sitename))
 378
 379     def locate_node(self, nodename):
 380         for site in self.plc_spec['sites']:
 381             for node in site['nodes']:
 382                 if node['name'] == nodename:
 383                     return site, node
 384         raise Exception("Cannot locate node {}".format(nodename))
 385
 386     def locate_hostname(self, hostname):
 387         for site in self.plc_spec['sites']:
 388             for node in site['nodes']:
 389                 if node['node_fields']['hostname'] == hostname:
 390                     return(site, node)
 391         raise Exception("Cannot locate hostname {}".format(hostname))
 392
 393     def locate_key(self, key_name):
 394         for key in self.plc_spec['keys']:
 395             if key['key_name'] == key_name:
 396                 return key
 397         raise Exception("Cannot locate key {}".format(key_name))
 398
 399     def locate_private_key_from_key_names(self, key_names):
 400         # locate the first avail. key
 401         found = False
 402         for key_name in key_names:
 403             key_spec = self.locate_key(key_name)
 404             test_key = TestKey(self,key_spec)
 405             publickey = test_key.publicpath()
 406             privatekey = test_key.privatepath()
 407             if os.path.isfile(publickey) and os.path.isfile(privatekey):
 408                 found = True
 409         if found:
 410             return privatekey
 411         else:
 412             return None
 413
 414     def locate_slice(self, slicename):
 415         for slice in self.plc_spec['slices']:
 416             if slice['slice_fields']['name'] == slicename:
 417                 return slice
 418         raise Exception("Cannot locate slice {}".format(slicename))
 419
 420     def all_sliver_objs(self):
 421         result = []
 422         for slice_spec in self.plc_spec['slices']:
 423             slicename = slice_spec['slice_fields']['name']
 424             for nodename in slice_spec['nodenames']:
 425                 result.append(self.locate_sliver_obj(nodename, slicename))
 426         return result
 427
 428     def locate_sliver_obj(self, nodename, slicename):
 429         site,node = self.locate_node(nodename)
 430         slice = self.locate_slice(slicename)
 431         # build objects
 432         test_site = TestSite(self, site)
 433         test_node = TestNode(self, test_site, node)
 434         # xxx the slice site is assumed to be the node site - mhh - probably harmless
 435         test_slice = TestSlice(self, test_site, slice)
 436         return TestSliver(self, test_node, test_slice)
 437
 438     def locate_first_node(self):
 439         nodename = self.plc_spec['slices'][0]['nodenames'][0]
 440         site,node = self.locate_node(nodename)
 441         test_site = TestSite(self, site)
 442         test_node = TestNode(self, test_site, node)
 443         return test_node
 444
 445     def locate_first_sliver(self):
 446         slice_spec = self.plc_spec['slices'][0]
 447         slicename = slice_spec['slice_fields']['name']
 448         nodename = slice_spec['nodenames'][0]
 449         return self.locate_sliver_obj(nodename,slicename)
 450
 451     # all different hostboxes used in this plc
 452     def get_BoxNodes(self):
 453         # maps on sites and nodes, return [ (host_box,test_node) ]
 454         tuples = []
 455         for site_spec in self.plc_spec['sites']:
 456             test_site = TestSite(self,site_spec)
 457             for node_spec in site_spec['nodes']:
 458                 test_node = TestNode(self, test_site, node_spec)
 459                 if not test_node.is_real():
 460                     tuples.append( (test_node.host_box(),test_node) )
 461         # transform into a dict { 'host_box' -> [ test_node .. ] }
 462         result = {}
 463         for (box,node) in tuples:
 464             if box not in result:
 465                 result[box] = [node]
 466             else:
 467                 result[box].append(node)
 468         return result
 469
 470     # a step for checking this stuff
 471     def show_boxes(self):
 472         'print summary of nodes location'
 473         for box,nodes in self.get_BoxNodes().items():
 474             print(box,":"," + ".join( [ node.name() for node in nodes ] ))
 475         return True
 476
 477     # make this a valid step
 478     def qemu_kill_all(self):
 479         'kill all qemu instances on the qemu boxes involved by this setup'
 480         # this is the brute force version, kill all qemus on that host box
 481         for (box,nodes) in self.get_BoxNodes().items():
 482             # pass the first nodename, as we don't push template-qemu on testboxes
 483             nodedir = nodes[0].nodedir()
 484             TestBoxQemu(box, self.options.buildname).qemu_kill_all(nodedir)
 485         return True
 486
 487     # make this a valid step
 488     def qemu_list_all(self):
 489         'list all qemu instances on the qemu boxes involved by this setup'
 490         for box,nodes in self.get_BoxNodes().items():
 491             # this is the brute force version, kill all qemus on that host box
 492             TestBoxQemu(box, self.options.buildname).qemu_list_all()
 493         return True
 494
 495     # kill only the qemus related to this test
 496     def qemu_list_mine(self):
 497         'list qemu instances for our nodes'
 498         for (box,nodes) in self.get_BoxNodes().items():
 499             # the fine-grain version
 500             for node in nodes:
 501                 node.list_qemu()
 502         return True
 503
 504     # kill only the qemus related to this test
 505     def qemu_clean_mine(self):
 506         'cleanup (rm -rf) qemu instances for our nodes'
 507         for box,nodes in self.get_BoxNodes().items():
 508             # the fine-grain version
 509             for node in nodes:
 510                 node.qemu_clean()
 511         return True
 512
 513     # kill only the right qemus
 514     def qemu_kill_mine(self):
 515         'kill the qemu instances for our nodes'
 516         for box,nodes in self.get_BoxNodes().items():
 517             # the fine-grain version
 518             for node in nodes:
 519                 node.kill_qemu()
 520         return True
 521
 522     #################### display config
 523     def show(self):
 524         "show test configuration after localization"
 525         self.show_pass(1)
 526         self.show_pass(2)
 527         return True
 528
 529     # uggly hack to make sure 'run export' only reports about the 1st plc
 530     # to avoid confusion - also we use 'inri_slice1' in various aliases..
 531     exported_id = 1
 532     def export(self):
 533         "print cut'n paste-able stuff to export env variables to your shell"
 534         # guess local domain from hostname
 535         if TestPlc.exported_id > 1:
 536             print("export GUESTHOSTNAME{:d}={}".format(TestPlc.exported_id, self.plc_spec['vservername']))
 537             return True
 538         TestPlc.exported_id += 1
 539         domain = socket.gethostname().split('.',1)[1]
 540         fqdn   = "{}.{}".format(self.plc_spec['host_box'], domain)
 541         print("export BUILD={}".format(self.options.buildname))
 542         print("export PLCHOSTLXC={}".format(fqdn))
 543         print("export GUESTNAME={}".format(self.vservername))
 544         print("export GUESTHOSTNAME={}.{}".format(self.vplchostname, domain))
 545         # find hostname of first node
 546         hostname, qemubox = self.all_node_infos()[0]
 547         print("export KVMHOST={}.{}".format(qemubox, domain))
 548         print("export NODE={}".format(hostname))
 549         return True
 550
 551     # entry point
 552     always_display_keys=['PLC_WWW_HOST', 'nodes', 'sites']
 553     def show_pass(self, passno):
 554         for (key,val) in self.plc_spec.items():
 555             if not self.options.verbose and key not in TestPlc.always_display_keys:
 556                 continue
 557             if passno == 2:
 558                 if key == 'sites':
 559                     for site in val:
 560                         self.display_site_spec(site)
 561                         for node in site['nodes']:
 562                             self.display_node_spec(node)
 563                 elif key == 'initscripts':
 564                     for initscript in val:
 565                         self.display_initscript_spec(initscript)
 566                 elif key == 'slices':
 567                     for slice in val:
 568                         self.display_slice_spec(slice)
 569                 elif key == 'keys':
 570                     for key in val:
 571                         self.display_key_spec(key)
 572             elif passno == 1:
 573                 if key not in ['sites', 'initscripts', 'slices', 'keys']:
 574                     print('+   ', key, ':', val)
 575
 576     def display_site_spec(self, site):
 577         print('+ ======== site', site['site_fields']['name'])
 578         for k,v in site.items():
 579             if not self.options.verbose and k not in TestPlc.always_display_keys:
 580                 continue
 581             if k == 'nodes':
 582                 if v:
 583                     print('+       ', 'nodes : ', end=' ')
 584                     for node in v:
 585                         print(node['node_fields']['hostname'],'', end=' ')
 586                     print('')
 587             elif k == 'users':
 588                 if v:
 589                     print('+       users : ', end=' ')
 590                     for user in v:
 591                         print(user['name'],'', end=' ')
 592                     print('')
 593             elif k == 'site_fields':
 594                 print('+       login_base', ':', v['login_base'])
 595             elif k == 'address_fields':
 596                 pass
 597             else:
 598                 print('+       ', end=' ')
 599                 utils.pprint(k, v)
 600
 601     def display_initscript_spec(self, initscript):
 602         print('+ ======== initscript', initscript['initscript_fields']['name'])
 603
 604     def display_key_spec(self, key):
 605         print('+ ======== key', key['key_name'])
 606
 607     def display_slice_spec(self, slice):
 608         print('+ ======== slice', slice['slice_fields']['name'])
 609         for k,v in slice.items():
 610             if k == 'nodenames':
 611                 if v:
 612                     print('+       nodes : ', end=' ')
 613                     for nodename in v:
 614                         print(nodename,'', end=' ')
 615                     print('')
 616             elif k == 'usernames':
 617                 if v:
 618                     print('+       users : ', end=' ')
 619                     for username in v:
 620                         print(username,'', end=' ')
 621                     print('')
 622             elif k == 'slice_fields':
 623                 print('+       fields', ':', end=' ')
 624                 print('max_nodes=',v['max_nodes'], end=' ')
 625                 print('')
 626             else:
 627                 print('+       ',k,v)
 628
 629     def display_node_spec(self, node):
 630         print("+           node={} host_box={}".format(node['name'], node['host_box']), end=' ')
 631         print("hostname=", node['node_fields']['hostname'], end=' ')
 632         print("ip=", node['interface_fields']['ip'])
 633         if self.options.verbose:
 634             utils.pprint("node details", node, depth=3)
 635
 636     # another entry point for just showing the boxes involved
 637     def display_mapping(self):
 638         TestPlc.display_mapping_plc(self.plc_spec)
 639         return True
 640
 641     @staticmethod
 642     def display_mapping_plc(plc_spec):
 643         print('+ MyPLC',plc_spec['name'])
 644         # WARNING this would not be right for lxc-based PLC's - should be harmless though
 645         print('+\tvserver address = root@{}:/vservers/{}'.format(plc_spec['host_box'], plc_spec['vservername']))
 646         print('+\tIP = {}/{}'.format(plc_spec['settings']['PLC_API_HOST'], plc_spec['vserverip']))
 647         for site_spec in plc_spec['sites']:
 648             for node_spec in site_spec['nodes']:
 649                 TestPlc.display_mapping_node(node_spec)
 650
 651     @staticmethod
 652     def display_mapping_node(node_spec):
 653         print('+   NODE {}'.format(node_spec['name']))
 654         print('+\tqemu box {}'.format(node_spec['host_box']))
 655         print('+\thostname={}'.format(node_spec['node_fields']['hostname']))
 656
 657     # write a timestamp in /vservers/<>.timestamp
 658     # cannot be inside the vserver, that causes vserver .. build to cough
 659     def plcvm_timestamp(self):
 660         "Create a timestamp to remember creation date for this plc"
 661         now = int(time.time())
 662         # TODO-lxc check this one
 663         # a first approx. is to store the timestamp close to the VM root like vs does
 664         stamp_path = self.vm_timestamp_path()
 665         stamp_dir = os.path.dirname(stamp_path)
 666         utils.system(self.test_ssh.actual_command("mkdir -p {}".format(stamp_dir)))
 667         return utils.system(self.test_ssh.actual_command("echo {:d} > {}".format(now, stamp_path))) == 0
 668
 669     # this is called inconditionnally at the beginning of the test sequence
 670     # just in case this is a rerun, so if the vm is not running it's fine
 671     def plcvm_delete(self):
 672         "vserver delete the test myplc"
 673         stamp_path = self.vm_timestamp_path()
 674         self.run_in_host("rm -f {}".format(stamp_path))
 675         self.run_in_host("virsh -c lxc:/// destroy {}".format(self.vservername))
 676         self.run_in_host("virsh -c lxc:/// undefine {}".format(self.vservername))
 677         self.run_in_host("rm -fr /vservers/{}".format(self.vservername))
 678         return True
 679
 680     ### install
 681     # historically the build was being fetched by the tests
 682     # now the build pushes itself as a subdir of the tests workdir
 683     # so that the tests do not have to worry about extracting the build (svn, git, or whatever)
 684     def plcvm_create(self):
 685         "vserver creation (no install done)"
 686         # push the local build/ dir to the testplc box
 687         if self.is_local():
 688             # a full path for the local calls
 689             build_dir = os.path.dirname(sys.argv[0])
 690             # sometimes this is empty - set to "." in such a case
 691             if not build_dir:
 692                 build_dir="."
 693             build_dir += "/build"
 694         else:
 695             # use a standard name - will be relative to remote buildname
 696             build_dir = "build"
 697             # remove for safety; do *not* mkdir first, otherwise we end up with build/build/
 698             self.test_ssh.rmdir(build_dir)
 699             self.test_ssh.copy(build_dir, recursive=True)
 700         # the repo url is taken from arch-rpms-url
 701         # with the last step (i386) removed
 702         repo_url = self.options.arch_rpms_url
 703         for level in [ 'arch' ]:
 704             repo_url = os.path.dirname(repo_url)
 705
 706         # invoke initvm (drop support for vs)
 707         script = "lbuild-initvm.sh"
 708         script_options = ""
 709         # pass the vbuild-nightly options to [lv]test-initvm
 710         script_options += " -p {}".format(self.options.personality)
 711         script_options += " -d {}".format(self.options.pldistro)
 712         script_options += " -f {}".format(self.options.fcdistro)
 713         script_options += " -r {}".format(repo_url)
 714         vserver_name = self.vservername
 715         try:
 716             vserver_hostname = socket.gethostbyaddr(self.vserverip)[0]
 717             script_options += " -n {}".format(vserver_hostname)
 718         except:
 719             print("Cannot reverse lookup {}".format(self.vserverip))
 720             print("This is considered fatal, as this might pollute the test results")
 721             return False
 722         create_vserver="{build_dir}/{script} {script_options} {vserver_name}".format(**locals())
 723         return self.run_in_host(create_vserver) == 0
 724
 725     ### install django through pip
 726     def django_install(self):
 727         # plcapi requires Django, that is no longer provided py fedora as an rpm
 728         # so we use pip instead
 729         """
 730         pip install Django
 731         """
 732         return self.pip3_install('Django')
 733
 734     ### install_rpm
 735     def plc_install(self):
 736         """
 737         yum install myplc, noderepo
 738         """
 739
 740         # compute nodefamily
 741         if self.options.personality == "linux32":
 742             arch = "i386"
 743         elif self.options.personality == "linux64":
 744             arch = "x86_64"
 745         else:
 746             raise Exception("Unsupported personality {}".format(self.options.personality))
 747         nodefamily = "{}-{}-{}".format(self.options.pldistro, self.options.fcdistro, arch)
 748
 749         # check it's possible to install just 'myplc-core' first
 750         if not self.dnf_install("myplc-core"):
 751             return False
 752
 753         pkgs_list = []
 754         pkgs_list.append("myplc")
 755         # pkgs_list.append("slicerepo-{}".format(nodefamily))
 756         # pkgs_list.append("noderepo-{}".format(nodefamily))
 757         pkgs_string=" ".join(pkgs_list)
 758         return self.dnf_install(pkgs_list)
 759
 760     def install_syslinux6(self):
 761         """
 762         install syslinux6 from the fedora21 release
 763         """
 764         key = 'http://mirror.onelab.eu/keys/RPM-GPG-KEY-fedora-21-primary'
 765
 766         rpms = [
 767             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-6.03-1.fc21.x86_64.rpm',
 768             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-nonlinux-6.03-1.fc21.noarch.rpm',
 769             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-perl-6.03-1.fc21.x86_64.rpm',
 770         ]
 771         # this can be done several times
 772         self.run_in_guest("rpm --import {key}".format(**locals()))
 773         return self.run_in_guest("yum -y localinstall {}".format(" ".join(rpms))) == 0
 774
 775     def bonding_builds(self):
 776         """
 777         list /etc/yum.repos.d on the myplc side
 778         """
 779         self.run_in_guest("ls /etc/yum.repos.d/*partial.repo")
 780         return True
 781
 782     def bonding_nodes(self):
 783         """
 784         List nodes known to the myplc together with their nodefamiliy
 785         """
 786         print("---------------------------------------- nodes")
 787         for node in self.apiserver.GetNodes(self.auth_root()):
 788             print("{} -> {}".format(node['hostname'],
 789                                     self.apiserver.GetNodeFlavour(self.auth_root(),node['hostname'])['nodefamily']))
 790         print("---------------------------------------- nodes")
 791
 792
 793     ###
 794     def mod_python(self):
 795         """yum install mod_python, useful on f18 and above so as to avoid broken wsgi"""
 796         return self.dnf_install( ['mod_python'] )
 797
 798     ###
 799     def plc_configure(self):
 800         "run plc-config-tty"
 801         tmpname = '{}.plc-config-tty'.format(self.name())
 802         with open(tmpname,'w') as fileconf:
 803             for var, value in self.plc_spec['settings'].items():
 804                 fileconf.write('e {}\n{}\n'.format(var, value))
 805             fileconf.write('w\n')
 806             fileconf.write('q\n')
 807         utils.system('cat {}'.format(tmpname))
 808         self.run_in_guest_piped('cat {}'.format(tmpname), 'plc-config-tty')
 809         utils.system('rm {}'.format(tmpname))
 810         return True
 811
 812     # care only about f>=27
 813     def start_stop_systemd(self, service, start_or_stop):
 814         "utility to start/stop a systemd-defined service (sfa)"
 815         return self.run_in_guest("systemctl {} {}".format(start_or_stop, service)) == 0
 816
 817     def plc_start(self):
 818         "start plc through systemclt"
 819         return self.start_stop_systemd('plc', 'start')
 820
 821     def plc_stop(self):
 822         "stop plc through systemctl"
 823         return self.start_stop_systemd('plc', 'stop')
 824
 825     def plcvm_start(self):
 826         "start the PLC vserver"
 827         self.start_guest()
 828         return True
 829
 830     def plcvm_stop(self):
 831         "stop the PLC vserver"
 832         self.stop_guest()
 833         return True
 834
 835     # stores the keys from the config for further use
 836     def keys_store(self):
 837         "stores test users ssh keys in keys/"
 838         for key_spec in self.plc_spec['keys']:
 839                 TestKey(self,key_spec).store_key()
 840         return True
 841
 842     def keys_clean(self):
 843         "removes keys cached in keys/"
 844         utils.system("rm -rf ./keys")
 845         return True
 846
 847     # fetches the ssh keys in the plc's /etc/planetlab and stores them in keys/
 848     # for later direct access to the nodes
 849     def keys_fetch(self):
 850         "gets ssh keys in /etc/planetlab/ and stores them locally in keys/"
 851         dir="./keys"
 852         if not os.path.isdir(dir):
 853             os.mkdir(dir)
 854         vservername = self.vservername
 855         vm_root = self.vm_root_in_host()
 856         overall = True
 857         prefix = 'debug_ssh_key'
 858         for ext in ['pub', 'rsa'] :
 859             src = "{vm_root}/etc/planetlab/{prefix}.{ext}".format(**locals())
 860             dst = "keys/{vservername}-debug.{ext}".format(**locals())
 861             if self.test_ssh.fetch(src, dst) != 0:
 862                 overall=False
 863         return overall
 864
 865     def sites(self):
 866         "create sites with PLCAPI"
 867         return self.do_sites()
 868
 869     def delete_sites(self):
 870         "delete sites with PLCAPI"
 871         return self.do_sites(action="delete")
 872
 873     def do_sites(self, action="add"):
 874         for site_spec in self.plc_spec['sites']:
 875             test_site = TestSite(self,site_spec)
 876             if (action != "add"):
 877                 utils.header("Deleting site {} in {}".format(test_site.name(), self.name()))
 878                 test_site.delete_site()
 879                 # deleted with the site
 880                 #test_site.delete_users()
 881                 continue
 882             else:
 883                 utils.header("Creating site {} & users in {}".format(test_site.name(), self.name()))
 884                 test_site.create_site()
 885                 test_site.create_users()
 886         return True
 887
 888     def delete_all_sites(self):
 889         "Delete all sites in PLC, and related objects"
 890         print('auth_root', self.auth_root())
 891         sites = self.apiserver.GetSites(self.auth_root(), {}, ['site_id', 'login_base'])
 892         for site in sites:
 893             # keep automatic site - otherwise we shoot in our own foot, root_auth is not valid anymore
 894             if site['login_base'] == self.plc_spec['settings']['PLC_SLICE_PREFIX']:
 895                 continue
 896             site_id = site['site_id']
 897             print('Deleting site_id', site_id)
 898             self.apiserver.DeleteSite(self.auth_root(), site_id)
 899         return True
 900
 901     def nodes(self):
 902         "create nodes with PLCAPI"
 903         return self.do_nodes()
 904     def delete_nodes(self):
 905         "delete nodes with PLCAPI"
 906         return self.do_nodes(action="delete")
 907
 908     def do_nodes(self, action="add"):
 909         for site_spec in self.plc_spec['sites']:
 910             test_site = TestSite(self, site_spec)
 911             if action != "add":
 912                 utils.header("Deleting nodes in site {}".format(test_site.name()))
 913                 for node_spec in site_spec['nodes']:
 914                     test_node = TestNode(self, test_site, node_spec)
 915                     utils.header("Deleting {}".format(test_node.name()))
 916                     test_node.delete_node()
 917             else:
 918                 utils.header("Creating nodes for site {} in {}".format(test_site.name(), self.name()))
 919                 for node_spec in site_spec['nodes']:
 920                     utils.pprint('Creating node {}'.format(node_spec), node_spec)
 921                     test_node = TestNode(self, test_site, node_spec)
 922                     test_node.create_node()
 923         return True
 924
 925     def nodegroups(self):
 926         "create nodegroups with PLCAPI"
 927         return self.do_nodegroups("add")
 928     def delete_nodegroups(self):
 929         "delete nodegroups with PLCAPI"
 930         return self.do_nodegroups("delete")
 931
 932     YEAR = 365*24*3600
 933     @staticmethod
 934     def translate_timestamp(start, grain, timestamp):
 935         if timestamp < TestPlc.YEAR:
 936             return start + timestamp*grain
 937         else:
 938             return timestamp
 939
 940     @staticmethod
 941     def timestamp_printable(timestamp):
 942         return time.strftime('%m-%d %H:%M:%S UTC', time.gmtime(timestamp))
 943
 944     def leases(self):
 945         "create leases (on reservable nodes only, use e.g. run -c default -c resa)"
 946         now = int(time.time())
 947         grain = self.apiserver.GetLeaseGranularity(self.auth_root())
 948         print('API answered grain=', grain)
 949         start = (now//grain)*grain
 950         start += grain
 951         # find out all nodes that are reservable
 952         nodes = self.all_reservable_nodenames()
 953         if not nodes:
 954             utils.header("No reservable node found - proceeding without leases")
 955             return True
 956         ok = True
 957         # attach them to the leases as specified in plc_specs
 958         # this is where the 'leases' field gets interpreted as relative of absolute
 959         for lease_spec in self.plc_spec['leases']:
 960             # skip the ones that come with a null slice id
 961             if not lease_spec['slice']:
 962                 continue
 963             lease_spec['t_from']  = TestPlc.translate_timestamp(start, grain, lease_spec['t_from'])
 964             lease_spec['t_until'] = TestPlc.translate_timestamp(start, grain, lease_spec['t_until'])
 965             lease_addition = self.apiserver.AddLeases(self.auth_root(), nodes, lease_spec['slice'],
 966                                                       lease_spec['t_from'], lease_spec['t_until'])
 967             if lease_addition['errors']:
 968                 utils.header("Cannot create leases, {}".format(lease_addition['errors']))
 969                 ok = False
 970             else:
 971                 utils.header('Leases on nodes {} for {} from {:d} ({}) until {:d} ({})'\
 972                              .format(nodes, lease_spec['slice'],
 973                                      lease_spec['t_from'],  TestPlc.timestamp_printable(lease_spec['t_from']),
 974                                      lease_spec['t_until'], TestPlc.timestamp_printable(lease_spec['t_until'])))
 975
 976         return ok
 977
 978     def delete_leases(self):
 979         "remove all leases in the myplc side"
 980         lease_ids = [ l['lease_id'] for l in self.apiserver.GetLeases(self.auth_root())]
 981         utils.header("Cleaning leases {}".format(lease_ids))
 982         self.apiserver.DeleteLeases(self.auth_root(), lease_ids)
 983         return True
 984
 985     def list_leases(self):
 986         "list all leases known to the myplc"
 987         leases = self.apiserver.GetLeases(self.auth_root())
 988         now = int(time.time())
 989         for l in leases:
 990             current = l['t_until'] >= now
 991             if self.options.verbose or current:
 992                 utils.header("{} {} from {} until {}"\
 993                              .format(l['hostname'], l['name'],
 994                                      TestPlc.timestamp_printable(l['t_from']),
 995                                      TestPlc.timestamp_printable(l['t_until'])))
 996         return True
 997
 998     # create nodegroups if needed, and populate
 999     def do_nodegroups(self, action="add"):
1000         # 1st pass to scan contents
1001         groups_dict = {}
1002         for site_spec in self.plc_spec['sites']:
1003             test_site = TestSite(self,site_spec)
1004             for node_spec in site_spec['nodes']:
1005                 test_node = TestNode(self, test_site, node_spec)
1006                 if 'nodegroups' in node_spec:
1007                     nodegroupnames = node_spec['nodegroups']
1008                     if isinstance(nodegroupnames, str):
1009                         nodegroupnames = [ nodegroupnames ]
1010                     for nodegroupname in nodegroupnames:
1011                         if nodegroupname not in groups_dict:
1012                             groups_dict[nodegroupname] = []
1013                         groups_dict[nodegroupname].append(test_node.name())
1014         auth = self.auth_root()
1015         overall = True
1016         for (nodegroupname,group_nodes) in groups_dict.items():
1017             if action == "add":
1018                 print('nodegroups:', 'dealing with nodegroup',\
1019                     nodegroupname, 'on nodes', group_nodes)
1020                 # first, check if the nodetagtype is here
1021                 tag_types = self.apiserver.GetTagTypes(auth, {'tagname':nodegroupname})
1022                 if tag_types:
1023                     tag_type_id = tag_types[0]['tag_type_id']
1024                 else:
1025                     tag_type_id = self.apiserver.AddTagType(auth,
1026                                                             {'tagname' : nodegroupname,
1027                                                              'description' : 'for nodegroup {}'.format(nodegroupname),
1028                                                              'category' : 'test'})
1029                 print('located tag (type)', nodegroupname, 'as', tag_type_id)
1030                 # create nodegroup
1031                 nodegroups = self.apiserver.GetNodeGroups(auth, {'groupname' : nodegroupname})
1032                 if not nodegroups:
1033                     self.apiserver.AddNodeGroup(auth, nodegroupname, tag_type_id, 'yes')
1034                     print('created nodegroup', nodegroupname, \
1035                         'from tagname', nodegroupname, 'and value', 'yes')
1036                 # set node tag on all nodes, value='yes'
1037                 for nodename in group_nodes:
1038                     try:
1039                         self.apiserver.AddNodeTag(auth, nodename, nodegroupname, "yes")
1040                     except:
1041                         traceback.print_exc()
1042                         print('node', nodename, 'seems to already have tag', nodegroupname)
1043                     # check anyway
1044                     try:
1045                         expect_yes = self.apiserver.GetNodeTags(auth,
1046                                                                 {'hostname' : nodename,
1047                                                                  'tagname'  : nodegroupname},
1048                                                                 ['value'])[0]['value']
1049                         if expect_yes != "yes":
1050                             print('Mismatch node tag on node',nodename,'got',expect_yes)
1051                             overall = False
1052                     except:
1053                         if not self.options.dry_run:
1054                             print('Cannot find tag', nodegroupname, 'on node', nodename)
1055                             overall = False
1056             else:
1057                 try:
1058                     print('cleaning nodegroup', nodegroupname)
1059                     self.apiserver.DeleteNodeGroup(auth, nodegroupname)
1060                 except:
1061                     traceback.print_exc()
1062                     overall = False
1063         return overall
1064
1065     # a list of TestNode objs
1066     def all_nodes(self):
1067         nodes=[]
1068         for site_spec in self.plc_spec['sites']:
1069             test_site = TestSite(self,site_spec)
1070             for node_spec in site_spec['nodes']:
1071                 nodes.append(TestNode(self, test_site, node_spec))
1072         return nodes
1073
1074     # return a list of tuples (nodename,qemuname)
1075     def all_node_infos(self) :
1076         node_infos = []
1077         for site_spec in self.plc_spec['sites']:
1078             node_infos += [ (node_spec['node_fields']['hostname'], node_spec['host_box']) \
1079                                 for node_spec in site_spec['nodes'] ]
1080         return node_infos
1081
1082     def all_nodenames(self):
1083         return [ x[0] for x in self.all_node_infos() ]
1084     def all_reservable_nodenames(self):
1085         res = []
1086         for site_spec in self.plc_spec['sites']:
1087             for node_spec in site_spec['nodes']:
1088                 node_fields = node_spec['node_fields']
1089                 if 'node_type' in node_fields and node_fields['node_type'] == 'reservable':
1090                     res.append(node_fields['hostname'])
1091         return res
1092
1093     # silent_minutes : during the first <silent_minutes> minutes nothing gets printed
1094     def nodes_check_boot_state(self, target_boot_state, timeout_minutes,
1095                                silent_minutes, period_seconds = 15):
1096         if self.options.dry_run:
1097             print('dry_run')
1098             return True
1099
1100         class CompleterTaskBootState(CompleterTask):
1101             def __init__(self, test_plc, hostname):
1102                 self.test_plc = test_plc
1103                 self.hostname = hostname
1104                 self.last_boot_state = 'undef'
1105             def actual_run(self):
1106                 try:
1107                     node = self.test_plc.apiserver.GetNodes(self.test_plc.auth_root(),
1108                                                             [ self.hostname ],
1109                                                             ['boot_state'])[0]
1110                     self.last_boot_state = node['boot_state']
1111                     return self.last_boot_state == target_boot_state
1112                 except:
1113                     return False
1114             def message(self):
1115                 return "CompleterTaskBootState with node {}".format(self.hostname)
1116             def failure_epilogue(self):
1117                 print("node {} in state {} - expected {}"\
1118                     .format(self.hostname, self.last_boot_state, target_boot_state))
1119
1120         timeout = timedelta(minutes=timeout_minutes)
1121         graceout = timedelta(minutes=silent_minutes)
1122         period   = timedelta(seconds=period_seconds)
1123         # the nodes that haven't checked yet - start with a full list and shrink over time
1124         utils.header("checking nodes boot state (expected {})".format(target_boot_state))
1125         tasks = [ CompleterTaskBootState(self,hostname) \
1126                       for (hostname,_) in self.all_node_infos() ]
1127         message = 'check_boot_state={}'.format(target_boot_state)
1128         return Completer(tasks, message=message).run(timeout, graceout, period)
1129
1130     def nodes_booted(self):
1131         return self.nodes_check_boot_state('boot', timeout_minutes=30, silent_minutes=28)
1132
1133     def probe_kvm_iptables(self):
1134         (_,kvmbox) = self.all_node_infos()[0]
1135         TestSsh(kvmbox).run("iptables-save")
1136         return True
1137
1138     # probing nodes
1139     def check_nodes_ping(self, timeout_seconds=60, period_seconds=10):
1140         class CompleterTaskPingNode(CompleterTask):
1141             def __init__(self, hostname):
1142                 self.hostname = hostname
1143             def run(self, silent):
1144                 command="ping -c 1 -w 1 {} >& /dev/null".format(self.hostname)
1145                 return utils.system(command, silent=silent) == 0
1146             def failure_epilogue(self):
1147                 print("Cannot ping node with name {}".format(self.hostname))
1148         timeout = timedelta(seconds = timeout_seconds)
1149         graceout = timeout
1150         period = timedelta(seconds = period_seconds)
1151         node_infos = self.all_node_infos()
1152         tasks = [ CompleterTaskPingNode(h) for (h,_) in node_infos ]
1153         return Completer(tasks, message='ping_node').run(timeout, graceout, period)
1154
1155     # ping node before we try to reach ssh, helpful for troubleshooting failing bootCDs
1156     def ping_node(self):
1157         "Ping nodes"
1158         return self.check_nodes_ping()
1159
1160     def check_nodes_ssh(self, debug, timeout_minutes, silent_minutes, period_seconds=15):
1161         # various delays
1162         timeout  = timedelta(minutes=timeout_minutes)
1163         graceout = timedelta(minutes=silent_minutes)
1164         period   = timedelta(seconds=period_seconds)
1165         vservername = self.vservername
1166         if debug:
1167             message = "debug"
1168             completer_message = 'ssh_node_debug'
1169             local_key = "keys/{vservername}-debug.rsa".format(**locals())
1170         else:
1171             message = "boot"
1172             completer_message = 'ssh_node_boot'
1173             local_key = "keys/key_admin.rsa"
1174         utils.header("checking ssh access to nodes (expected in {} mode)".format(message))
1175         node_infos = self.all_node_infos()
1176         tasks = [ CompleterTaskNodeSsh(nodename, qemuname, local_key,
1177                                         boot_state=message, dry_run=self.options.dry_run) \
1178                       for (nodename, qemuname) in node_infos ]
1179         return Completer(tasks, message=completer_message).run(timeout, graceout, period)
1180
1181     def ssh_node_debug(self):
1182         "Tries to ssh into nodes in debug mode with the debug ssh key"
1183         return self.check_nodes_ssh(debug = True,
1184                                     timeout_minutes = self.ssh_node_debug_timeout,
1185                                     silent_minutes = self.ssh_node_debug_silent)
1186
1187     def ssh_node_boot(self):
1188         "Tries to ssh into nodes in production mode with the root ssh key"
1189         return self.check_nodes_ssh(debug = False,
1190                                     timeout_minutes = self.ssh_node_boot_timeout,
1191                                     silent_minutes = self.ssh_node_boot_silent)
1192
1193     def node_bmlogs(self):
1194         "Checks that there's a non-empty dir. /var/log/bm/raw"
1195         return utils.system(self.actual_command_in_guest("ls /var/log/bm/raw")) == 0
1196
1197     @node_mapper
1198     def qemu_local_init(self): pass
1199     @node_mapper
1200     def bootcd(self): pass
1201     @node_mapper
1202     def qemu_local_config(self): pass
1203     @node_mapper
1204     def qemu_export(self): pass
1205     @node_mapper
1206     def qemu_cleanlog(self): pass
1207     @node_mapper
1208     def nodestate_reinstall(self): pass
1209     @node_mapper
1210     def nodestate_upgrade(self): pass
1211     @node_mapper
1212     def nodestate_safeboot(self): pass
1213     @node_mapper
1214     def nodestate_boot(self): pass
1215     @node_mapper
1216     def nodestate_show(self): pass
1217     @node_mapper
1218     def nodedistro_f14(self): pass
1219     @node_mapper
1220     def nodedistro_f18(self): pass
1221     @node_mapper
1222     def nodedistro_f20(self): pass
1223     @node_mapper
1224     def nodedistro_f21(self): pass
1225     @node_mapper
1226     def nodedistro_f22(self): pass
1227     @node_mapper
1228     def nodedistro_show(self): pass
1229
1230     ### check hooks : invoke scripts from hooks/{node,slice}
1231     def check_hooks_node(self):
1232         return self.locate_first_node().check_hooks()
1233     def check_hooks_sliver(self) :
1234         return self.locate_first_sliver().check_hooks()
1235
1236     def check_hooks(self):
1237         "runs unit tests in the node and slice contexts - see hooks/{node,slice}"
1238         return self.check_hooks_node() and self.check_hooks_sliver()
1239
1240     ### initscripts
1241     def do_check_initscripts(self):
1242         class CompleterTaskInitscript(CompleterTask):
1243             def __init__(self, test_sliver, stamp):
1244                 self.test_sliver = test_sliver
1245                 self.stamp = stamp
1246             def actual_run(self):
1247                 return self.test_sliver.check_initscript_stamp(self.stamp)
1248             def message(self):
1249                 return "initscript checker for {}".format(self.test_sliver.name())
1250             def failure_epilogue(self):
1251                 print("initscript stamp {} not found in sliver {}"\
1252                     .format(self.stamp, self.test_sliver.name()))
1253
1254         tasks = []
1255         for slice_spec in self.plc_spec['slices']:
1256             if 'initscriptstamp' not in slice_spec:
1257                 continue
1258             stamp = slice_spec['initscriptstamp']
1259             slicename = slice_spec['slice_fields']['name']
1260             for nodename in slice_spec['nodenames']:
1261                 print('nodename', nodename, 'slicename', slicename, 'stamp', stamp)
1262                 site,node = self.locate_node(nodename)
1263                 # xxx - passing the wrong site - probably harmless
1264                 test_site = TestSite(self, site)
1265                 test_slice = TestSlice(self, test_site, slice_spec)
1266                 test_node = TestNode(self, test_site, node)
1267                 test_sliver = TestSliver(self, test_node, test_slice)
1268                 tasks.append(CompleterTaskInitscript(test_sliver, stamp))
1269         return Completer(tasks, message='check_initscripts').\
1270             run (timedelta(minutes=5), timedelta(minutes=4), timedelta(seconds=10))
1271
1272     def check_initscripts(self):
1273         "check that the initscripts have triggered"
1274         return self.do_check_initscripts()
1275
1276     def initscripts(self):
1277         "create initscripts with PLCAPI"
1278         for initscript in self.plc_spec['initscripts']:
1279             utils.pprint('Adding Initscript in plc {}'.format(self.plc_spec['name']), initscript)
1280             self.apiserver.AddInitScript(self.auth_root(), initscript['initscript_fields'])
1281         return True
1282
1283     def delete_initscripts(self):
1284         "delete initscripts with PLCAPI"
1285         for initscript in self.plc_spec['initscripts']:
1286             initscript_name = initscript['initscript_fields']['name']
1287             print(('Attempting to delete {} in plc {}'.format(initscript_name, self.plc_spec['name'])))
1288             try:
1289                 self.apiserver.DeleteInitScript(self.auth_root(), initscript_name)
1290                 print(initscript_name, 'deleted')
1291             except:
1292                 print('deletion went wrong - probably did not exist')
1293         return True
1294
1295     ### manage slices
1296     def slices(self):
1297         "create slices with PLCAPI"
1298         return self.do_slices(action="add")
1299
1300     def delete_slices(self):
1301         "delete slices with PLCAPI"
1302         return self.do_slices(action="delete")
1303
1304     def fill_slices(self):
1305         "add nodes in slices with PLCAPI"
1306         return self.do_slices(action="fill")
1307
1308     def empty_slices(self):
1309         "remove nodes from slices with PLCAPI"
1310         return self.do_slices(action="empty")
1311
1312     def do_slices(self,  action="add"):
1313         for slice in self.plc_spec['slices']:
1314             site_spec = self.locate_site(slice['sitename'])
1315             test_site = TestSite(self,site_spec)
1316             test_slice=TestSlice(self,test_site,slice)
1317             if action == "delete":
1318                 test_slice.delete_slice()
1319             elif action == "fill":
1320                 test_slice.add_nodes()
1321             elif action == "empty":
1322                 test_slice.delete_nodes()
1323             else:
1324                 test_slice.create_slice()
1325         return True
1326
1327     @slice_mapper__tasks(20, 10, 15)
1328     def ssh_slice(self): pass
1329     @slice_mapper__tasks(20, 19, 15)
1330     def ssh_slice_off(self): pass
1331     @slice_mapper__tasks(1, 1, 15)
1332     def slice_fs_present(self): pass
1333     @slice_mapper__tasks(1, 1, 15)
1334     def slice_fs_deleted(self): pass
1335
1336     # use another name so we can exclude/ignore it from the tests on the nightly command line
1337     def ssh_slice_again(self): return self.ssh_slice()
1338     # note that simply doing ssh_slice_again=ssh_slice would kind of work too
1339     # but for some reason the ignore-wrapping thing would not
1340
1341     @slice_mapper
1342     def ssh_slice_basics(self): pass
1343     @slice_mapper
1344     def check_vsys_defaults(self): pass
1345
1346     @node_mapper
1347     def keys_clear_known_hosts(self): pass
1348
1349     def plcapi_urls(self):
1350         """
1351         attempts to reach the PLCAPI with various forms for the URL
1352         """
1353         return PlcapiUrlScanner(self.auth_root(), ip=self.vserverip).scan()
1354
1355     def speed_up_slices(self):
1356         "tweak nodemanager cycle (wait time) to 30+/-10 s"
1357         return self._speed_up_slices (30, 10)
1358     def super_speed_up_slices(self):
1359         "dev mode: tweak nodemanager cycle (wait time) to 5+/-1 s"
1360         return self._speed_up_slices(5, 1)
1361
1362     def _speed_up_slices(self, p, r):
1363         # create the template on the server-side
1364         template = "{}.nodemanager".format(self.name())
1365         with open(template,"w") as template_file:
1366             template_file.write('OPTIONS="-p {} -r {} -d"\n'.format(p, r))
1367         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1368         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1369         self.test_ssh.copy_abs(template, remote)
1370         # Add a conf file
1371         if not self.apiserver.GetConfFiles(self.auth_root(),
1372                                            {'dest' : '/etc/sysconfig/nodemanager'}):
1373             self.apiserver.AddConfFile(self.auth_root(),
1374                                         {'dest' : '/etc/sysconfig/nodemanager',
1375                                          'source' : 'PlanetLabConf/nodemanager',
1376                                          'postinstall_cmd' : 'service nm restart',})
1377         return True
1378
1379     def debug_nodemanager(self):
1380         "sets verbose mode for nodemanager, and speeds up cycle even more (needs speed_up_slices first)"
1381         template = "{}.nodemanager".format(self.name())
1382         with open(template,"w") as template_file:
1383             template_file.write('OPTIONS="-p 10 -r 6 -v -d"\n')
1384         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1385         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1386         self.test_ssh.copy_abs(template, remote)
1387         return True
1388
1389     @node_mapper
1390     def qemu_start(self) : pass
1391
1392     @node_mapper
1393     def qemu_timestamp(self) : pass
1394
1395     @node_mapper
1396     def qemu_nodefamily(self): pass
1397
1398     # when a spec refers to a node possibly on another plc
1399     def locate_sliver_obj_cross(self, nodename, slicename, other_plcs):
1400         for plc in [ self ] + other_plcs:
1401             try:
1402                 return plc.locate_sliver_obj(nodename, slicename)
1403             except:
1404                 pass
1405         raise Exception("Cannot locate sliver {}@{} among all PLCs".format(nodename, slicename))
1406
1407     # implement this one as a cross step so that we can take advantage of different nodes
1408     # in multi-plcs mode
1409     def cross_check_tcp(self, other_plcs):
1410         "check TCP connectivity between 2 slices (or in loopback if only one is defined)"
1411         if 'tcp_specs' not in self.plc_spec or not self.plc_spec['tcp_specs']:
1412             utils.header("check_tcp: no/empty config found")
1413             return True
1414         specs = self.plc_spec['tcp_specs']
1415         overall = True
1416
1417         # first wait for the network to be up and ready from the slices
1418         class CompleterTaskNetworkReadyInSliver(CompleterTask):
1419             def __init__(self, test_sliver):
1420                 self.test_sliver = test_sliver
1421             def actual_run(self):
1422                 return self.test_sliver.check_tcp_ready(port = 9999)
1423             def message(self):
1424                 return "network ready checker for {}".format(self.test_sliver.name())
1425             def failure_epilogue(self):
1426                 print("could not bind port from sliver {}".format(self.test_sliver.name()))
1427
1428         sliver_specs = {}
1429         tasks = []
1430         managed_sliver_names = set()
1431         for spec in specs:
1432             # locate the TestSliver instances involved, and cache them in the spec instance
1433             spec['s_sliver'] = self.locate_sliver_obj_cross(spec['server_node'], spec['server_slice'], other_plcs)
1434             spec['c_sliver'] = self.locate_sliver_obj_cross(spec['client_node'], spec['client_slice'], other_plcs)
1435             message = "Will check TCP between s={} and c={}"\
1436                       .format(spec['s_sliver'].name(), spec['c_sliver'].name())
1437             if 'client_connect' in spec:
1438                 message += " (using {})".format(spec['client_connect'])
1439             utils.header(message)
1440             # we need to check network presence in both slivers, but also
1441             # avoid to insert a sliver several times
1442             for sliver in [ spec['s_sliver'], spec['c_sliver'] ]:
1443                 if sliver.name() not in managed_sliver_names:
1444                     tasks.append(CompleterTaskNetworkReadyInSliver(sliver))
1445                     # add this sliver's name in the set
1446                     managed_sliver_names .update( {sliver.name()} )
1447
1448         # wait for the netork to be OK in all server sides
1449         if not Completer(tasks, message='check for network readiness in slivers').\
1450            run(timedelta(seconds=30), timedelta(seconds=24), period=timedelta(seconds=5)):
1451             return False
1452
1453         # run server and client
1454         for spec in specs:
1455             port = spec['port']
1456             # server side
1457             # the issue here is that we have the server run in background
1458             # and so we have no clue if it took off properly or not
1459             # looks like in some cases it does not
1460             address = spec['s_sliver'].test_node.name()
1461             if not spec['s_sliver'].run_tcp_server(address, port, timeout=20):
1462                 overall = False
1463                 break
1464
1465             # idem for the client side
1466             # use nodename from located sliver, unless 'client_connect' is set
1467             if 'client_connect' in spec:
1468                 destination = spec['client_connect']
1469             else:
1470                 destination = spec['s_sliver'].test_node.name()
1471             if not spec['c_sliver'].run_tcp_client(destination, port):
1472                 overall = False
1473         return overall
1474
1475     # painfully enough, we need to allow for some time as netflow might show up last
1476     def check_system_slice(self):
1477         "all nodes: check that a system slice is alive"
1478         # netflow currently not working in the lxc distro
1479         # drl not built at all in the wtx distro
1480         # if we find either of them we're happy
1481         return self.check_netflow() or self.check_drl()
1482
1483     # expose these
1484     def check_netflow(self): return self._check_system_slice('netflow')
1485     def check_drl(self): return self._check_system_slice('drl')
1486
1487     # we have the slices up already here, so it should not take too long
1488     def _check_system_slice(self, slicename, timeout_minutes=5, period_seconds=15):
1489         class CompleterTaskSystemSlice(CompleterTask):
1490             def __init__(self, test_node, dry_run):
1491                 self.test_node = test_node
1492                 self.dry_run = dry_run
1493             def actual_run(self):
1494                 return self.test_node._check_system_slice(slicename, dry_run=self.dry_run)
1495             def message(self):
1496                 return "System slice {} @ {}".format(slicename, self.test_node.name())
1497             def failure_epilogue(self):
1498                 print("COULD not find system slice {} @ {}".format(slicename, self.test_node.name()))
1499         timeout = timedelta(minutes=timeout_minutes)
1500         silent  = timedelta(0)
1501         period  = timedelta(seconds=period_seconds)
1502         tasks = [ CompleterTaskSystemSlice(test_node, self.options.dry_run) \
1503                       for test_node in self.all_nodes() ]
1504         return Completer(tasks, message='_check_system_slice').run(timeout, silent, period)
1505
1506     def plcsh_stress_test(self):
1507         "runs PLCAPI stress test, that checks Add/Update/Delete on all types - preserves contents"
1508         # install the stress-test in the plc image
1509         location = "/usr/share/plc_api/plcsh_stress_test.py"
1510         remote = "{}/{}".format(self.vm_root_in_host(), location)
1511         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1512         command = location
1513         command += " -- --check"
1514         if self.options.size == 1:
1515             command +=  " --tiny"
1516         return self.run_in_guest(command) == 0
1517
1518     # populate runs the same utility without slightly different options
1519     # in particular runs with --preserve (dont cleanup) and without --check
1520     # also it gets run twice, once with the --foreign option for creating fake foreign entries
1521
1522     def install_pip2(self):
1523
1524         replacements = [
1525             "https://acc.dl.osdn.jp/storage/g/u/un/unitedrpms/32/x86_64/python2-pip-19.1.1-7.fc32.noarch.rpm",
1526         ]
1527
1528         return (
1529                self.run_in_guest("pip2 --version") == 0
1530             or self.run_in_guest("dnf install python2-pip") == 0
1531             or self.run_in_guest("dnf localinstall -y " + " ".join(replacements)) == 0)
1532
1533
1534     def install_m2crypto(self):
1535
1536         # installing m2crypto for python2 is increasingly difficult
1537         # f29 and f31: dnf install python2-m2crypto
1538         # f33: no longer available but the f31 repos below do the job just fine
1539         # note that using pip2 does not look like a viable option because it does
1540         # an install from sources and that's quite awkward
1541
1542         replacements = [
1543             "http://mirror.onelab.eu/fedora/releases/31/Everything/x86_64/os/Packages/p/python2-typing-3.6.2-5.fc31.noarch.rpm",
1544             "http://mirror.onelab.eu/fedora/releases/31/Everything/x86_64/os/Packages/p/python2-m2crypto-0.35.2-2.fc31.x86_64.rpm",
1545         ]
1546
1547         return (
1548                self.run_in_guest('python2 -c "import M2Crypto"', backslash=True) == 0
1549             or self.run_in_guest("pip2 install python2-m2crypto") == 0
1550             or self.run_in_guest("dnf localinstall -y " + " ".join(replacements)) == 0)
1551
1552         return attempt
1553
1554         # about pip2:
1555         # we can try and use
1556         # that qould then need to be mirrored
1557         # so the logic goes like this
1558         # check for pip2 command
1559         # if not, try dnf install python2-pip
1560         # if still not, dnf localinstall the above
1561
1562
1563     def sfa_install_all(self):
1564         "yum install sfa sfa-plc sfa-sfatables sfa-client"
1565
1566         # the rpm/dnf packages named in python2-* are getting deprecated
1567         # we use pip2 instead
1568         # but that's not good for m2crypto
1569
1570         pip_dependencies = [
1571             'sqlalchemy-migrate',
1572             'lxml',
1573             'python-dateutil',
1574             'psycopg2-binary',
1575         ]
1576
1577         return (
1578                     self.install_pip2()
1579                 and self.install_m2crypto()
1580                 and all((self.run_in_guest(f"pip2 install {dep}") == 0)
1581                         for dep in pip_dependencies)
1582                 and self.dnf_install("sfa sfa-plc sfa-sfatables sfa-client")
1583                 and self.run_in_guest("systemctl enable sfa-registry")==0
1584                 and self.run_in_guest("systemctl enable sfa-aggregate")==0)
1585
1586     def sfa_install_core(self):
1587         "yum install sfa"
1588         return self.dnf_install("sfa")
1589
1590     def sfa_install_plc(self):
1591         "yum install sfa-plc"
1592         return self.dnf_install("sfa-plc")
1593
1594     def sfa_install_sfatables(self):
1595         "yum install sfa-sfatables"
1596         return self.dnf_install("sfa-sfatables")
1597
1598     # for some very odd reason, this sometimes fails with the following symptom
1599     # # yum install sfa-client
1600     # Setting up Install Process
1601     # ...
1602     # Downloading Packages:
1603     # Running rpm_check_debug
1604     # Running Transaction Test
1605     # Transaction Test Succeeded
1606     # Running Transaction
1607     # Transaction couldn't start:
1608     # installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem
1609     # [('installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem', (9, '/', 69632L))]
1610     # even though in the same context I have
1611     # [2012.05.23--f14-32-sfastd1-1-vplc07] / # df -h
1612     # Filesystem            Size  Used Avail Use% Mounted on
1613     # /dev/hdv1             806G  264G  501G  35% /
1614     # none                   16M   36K   16M   1% /tmp
1615     #
1616     # so as a workaround, we first try yum install, and then invoke rpm on the cached rpm...
1617     def sfa_install_client(self):
1618         "yum install sfa-client"
1619         first_try = self.dnf_install("sfa-client")
1620         if first_try:
1621             return True
1622         utils.header("********** Regular yum failed - special workaround in place, 2nd chance")
1623         code, cached_rpm_path = \
1624                 utils.output_of(self.actual_command_in_guest('find /var/cache/yum -name sfa-client\*.rpm'))
1625         utils.header("rpm_path=<<{}>>".format(rpm_path))
1626         # just for checking
1627         self.run_in_guest("rpm -i {}".format(cached_rpm_path))
1628         return self.dnf_check_installed("sfa-client")
1629
1630     def sfa_dbclean(self):
1631         "thoroughly wipes off the SFA database"
1632         return self.run_in_guest("sfaadmin reg nuke") == 0 or \
1633             self.run_in_guest("sfa-nuke.py") == 0 or \
1634             self.run_in_guest("sfa-nuke-plc.py") == 0 or \
1635             self.run_in_guest("sfaadmin registry nuke") == 0
1636
1637     def sfa_fsclean(self):
1638         "cleanup /etc/sfa/trusted_roots and /var/lib/sfa"
1639         self.run_in_guest("rm -rf /etc/sfa/trusted_roots /var/lib/sfa/authorities")
1640         return True
1641
1642     def sfa_plcclean(self):
1643         "cleans the PLC entries that were created as a side effect of running the script"
1644         # ignore result
1645         sfa_spec = self.plc_spec['sfa']
1646
1647         for auth_sfa_spec in sfa_spec['auth_sfa_specs']:
1648             login_base = auth_sfa_spec['login_base']
1649             try:
1650                 self.apiserver.DeleteSite(self.auth_root(),login_base)
1651             except:
1652                 print("Site {} already absent from PLC db".format(login_base))
1653
1654             for spec_name in ['pi_spec', 'user_spec']:
1655                 user_spec = auth_sfa_spec[spec_name]
1656                 username = user_spec['email']
1657                 try:
1658                     self.apiserver.DeletePerson(self.auth_root(),username)
1659                 except:
1660                     # this in fact is expected as sites delete their members
1661                     #print "User {} already absent from PLC db".format(username)
1662                     pass
1663
1664         print("REMEMBER TO RUN sfa_import AGAIN")
1665         return True
1666
1667     def sfa_uninstall(self):
1668         "uses rpm to uninstall sfa - ignore result"
1669         self.run_in_guest("rpm -e sfa sfa-sfatables sfa-client sfa-plc")
1670         self.run_in_guest("rm -rf /var/lib/sfa")
1671         self.run_in_guest("rm -rf /etc/sfa")
1672         self.run_in_guest("rm -rf /var/log/sfa_access.log /var/log/sfa_import_plc.log /var/log/sfa.daemon")
1673         # xxx tmp
1674         self.run_in_guest("rpm -e --noscripts sfa-plc")
1675         return True
1676
1677     ### run unit tests for SFA
1678     # NOTE: for some reason on f14/i386, yum install sfa-tests fails for no reason
1679     # Running Transaction
1680     # Transaction couldn't start:
1681     # installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem
1682     # [('installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem', (9, '/', 208896L))]
1683     # no matter how many Gbs are available on the testplc
1684     # could not figure out what's wrong, so...
1685     # if the yum install phase fails, consider the test is successful
1686     # other combinations will eventually run it hopefully
1687     def sfa_utest(self):
1688         "dnf install sfa-tests and run SFA unittests"
1689         self.run_in_guest("dnf -y install sfa-tests")
1690         # failed to install - forget it
1691         if self.run_in_guest("rpm -q sfa-tests") != 0:
1692             utils.header("WARNING: SFA unit tests failed to install, ignoring")
1693             return True
1694         return self.run_in_guest("/usr/share/sfa/tests/testAll.py") == 0
1695
1696     ###
1697     def confdir(self):
1698         dirname = "conf.{}".format(self.plc_spec['name'])
1699         if not os.path.isdir(dirname):
1700             utils.system("mkdir -p {}".format(dirname))
1701         if not os.path.isdir(dirname):
1702             raise Exception("Cannot create config dir for plc {}".format(self.name()))
1703         return dirname
1704
1705     def conffile(self, filename):
1706         return "{}/{}".format(self.confdir(), filename)
1707     def confsubdir(self, dirname, clean, dry_run=False):
1708         subdirname = "{}/{}".format(self.confdir(), dirname)
1709         if clean:
1710             utils.system("rm -rf {}".format(subdirname))
1711         if not os.path.isdir(subdirname):
1712             utils.system("mkdir -p {}".format(subdirname))
1713         if not dry_run and not os.path.isdir(subdirname):
1714             raise "Cannot create config subdir {} for plc {}".format(dirname, self.name())
1715         return subdirname
1716
1717     def conffile_clean(self, filename):
1718         filename=self.conffile(filename)
1719         return utils.system("rm -rf {}".format(filename))==0
1720
1721     ###
1722     def sfa_configure(self):
1723         "run sfa-config-tty"
1724         tmpname = self.conffile("sfa-config-tty")
1725         with open(tmpname,'w') as fileconf:
1726             for var, value in self.plc_spec['sfa']['settings'].items():
1727                 fileconf.write('e {}\n{}\n'.format(var, value))
1728             fileconf.write('w\n')
1729             fileconf.write('R\n')
1730             fileconf.write('q\n')
1731         utils.system('cat {}'.format(tmpname))
1732         self.run_in_guest_piped('cat {}'.format(tmpname), 'sfa-config-tty')
1733         return True
1734
1735     def aggregate_xml_line(self):
1736         port = self.plc_spec['sfa']['neighbours-port']
1737         return '<aggregate addr="{}" hrn="{}" port="{}"/>'\
1738             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'], port)
1739
1740     def registry_xml_line(self):
1741         return '<registry addr="{}" hrn="{}" port="12345"/>'\
1742             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'])
1743
1744
1745     # a cross step that takes all other plcs in argument
1746     def cross_sfa_configure(self, other_plcs):
1747         "writes aggregates.xml and registries.xml that point to all other PLCs in the test"
1748         # of course with a single plc, other_plcs is an empty list
1749         if not other_plcs:
1750             return True
1751         agg_fname = self.conffile("agg.xml")
1752         with open(agg_fname,"w") as out:
1753             out.write("<aggregates>{}</aggregates>\n"\
1754                       .format(" ".join([ plc.aggregate_xml_line() for plc in other_plcs ])))
1755         utils.header("(Over)wrote {}".format(agg_fname))
1756         reg_fname=self.conffile("reg.xml")
1757         with open(reg_fname,"w") as out:
1758             out.write("<registries>{}</registries>\n"\
1759                       .format(" ".join([ plc.registry_xml_line() for plc in other_plcs ])))
1760         utils.header("(Over)wrote {}".format(reg_fname))
1761         return self.test_ssh.copy_abs(agg_fname,
1762                                       '/{}/etc/sfa/aggregates.xml'.format(self.vm_root_in_host())) == 0 \
1763            and self.test_ssh.copy_abs(reg_fname,
1764                                       '/{}/etc/sfa/registries.xml'.format(self.vm_root_in_host())) == 0
1765
1766     def sfa_import(self):
1767         "use sfaadmin to import from plc"
1768         auth = self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH']
1769         return self.run_in_guest('sfaadmin reg import_registry') == 0
1770
1771     def sfa_start(self):
1772         "start SFA through systemctl - also install dependencies"
1773
1774         return (self.start_stop_systemd('sfa-registry', 'start')
1775             and self.start_stop_systemd('sfa-aggregate', 'start'))
1776
1777
1778     def sfi_configure(self):
1779         "Create /root/sfi on the plc side for sfi client configuration"
1780         if self.options.dry_run:
1781             utils.header("DRY RUN - skipping step")
1782             return True
1783         sfa_spec = self.plc_spec['sfa']
1784         # cannot use auth_sfa_mapper to pass dir_name
1785         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1786             test_slice = TestAuthSfa(self, slice_spec)
1787             dir_basename = os.path.basename(test_slice.sfi_path())
1788             dir_name = self.confsubdir("dot-sfi/{}".format(dir_basename),
1789                                        clean=True, dry_run=self.options.dry_run)
1790             test_slice.sfi_configure(dir_name)
1791             # push into the remote /root/sfi area
1792             location = test_slice.sfi_path()
1793             remote = "{}/{}".format(self.vm_root_in_host(), location)
1794             self.test_ssh.mkdir(remote, abs=True)
1795             # need to strip last level or remote otherwise we get an extra dir level
1796             self.test_ssh.copy_abs(dir_name, os.path.dirname(remote), recursive=True)
1797
1798         return True
1799
1800     def sfi_clean(self):
1801         "clean up /root/sfi on the plc side"
1802         self.run_in_guest("rm -rf /root/sfi")
1803         return True
1804
1805     def sfa_rspec_empty(self):
1806         "expose a static empty rspec (ships with the tests module) in the sfi directory"
1807         filename = "empty-rspec.xml"
1808         overall = True
1809         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1810             test_slice = TestAuthSfa(self, slice_spec)
1811             in_vm = test_slice.sfi_path()
1812             remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1813             if self.test_ssh.copy_abs(filename, remote) !=0:
1814                 overall = False
1815         return overall
1816
1817     @auth_sfa_mapper
1818     def sfa_register_site(self): pass
1819     @auth_sfa_mapper
1820     def sfa_register_pi(self): pass
1821     @auth_sfa_mapper
1822     def sfa_register_user(self): pass
1823     @auth_sfa_mapper
1824     def sfa_update_user(self): pass
1825     @auth_sfa_mapper
1826     def sfa_register_slice(self): pass
1827     @auth_sfa_mapper
1828     def sfa_renew_slice(self): pass
1829     @auth_sfa_mapper
1830     def sfa_get_expires(self): pass
1831     @auth_sfa_mapper
1832     def sfa_discover(self): pass
1833     @auth_sfa_mapper
1834     def sfa_rspec(self): pass
1835     @auth_sfa_mapper
1836     def sfa_allocate(self): pass
1837     @auth_sfa_mapper
1838     def sfa_allocate_empty(self): pass
1839     @auth_sfa_mapper
1840     def sfa_provision(self): pass
1841     @auth_sfa_mapper
1842     def sfa_provision_empty(self): pass
1843     @auth_sfa_mapper
1844     def sfa_describe(self): pass
1845     @auth_sfa_mapper
1846     def sfa_check_slice_plc(self): pass
1847     @auth_sfa_mapper
1848     def sfa_check_slice_plc_empty(self): pass
1849     @auth_sfa_mapper
1850     def sfa_update_slice(self): pass
1851     @auth_sfa_mapper
1852     def sfa_remove_user_from_slice(self): pass
1853     @auth_sfa_mapper
1854     def sfa_insert_user_in_slice(self): pass
1855     @auth_sfa_mapper
1856     def sfi_list(self): pass
1857     @auth_sfa_mapper
1858     def sfi_show_site(self): pass
1859     @auth_sfa_mapper
1860     def sfi_show_slice(self): pass
1861     @auth_sfa_mapper
1862     def sfi_show_slice_researchers(self): pass
1863     @auth_sfa_mapper
1864     def ssh_slice_sfa(self): pass
1865     @auth_sfa_mapper
1866     def sfa_delete_user(self): pass
1867     @auth_sfa_mapper
1868     def sfa_delete_slice(self): pass
1869
1870     def sfa_stop(self):
1871         "stop sfa through systemclt"
1872         return (self.start_stop_systemd('sfa-aggregate', 'stop') and
1873                 self.start_stop_systemd('sfa-registry', 'stop'))
1874
1875     def populate(self):
1876         "creates random entries in the PLCAPI"
1877         # install the stress-test in the plc image
1878         location = "/usr/share/plc_api/plcsh_stress_test.py"
1879         remote = "{}/{}".format(self.vm_root_in_host(), location)
1880         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1881         command = location
1882         command += " -- --preserve --short-names"
1883         local = (self.run_in_guest(command) == 0);
1884         # second run with --foreign
1885         command += ' --foreign'
1886         remote = (self.run_in_guest(command) == 0);
1887         return local and remote
1888
1889
1890     ####################
1891     @bonding_redirector
1892     def bonding_init_partial(self): pass
1893
1894     @bonding_redirector
1895     def bonding_add_yum(self): pass
1896
1897     @bonding_redirector
1898     def bonding_install_rpms(self): pass
1899
1900     ####################
1901
1902     def gather_logs(self):
1903         "gets all possible logs from plc's/qemu node's/slice's for future reference"
1904         # (1.a) get the plc's /var/log/ and store it locally in logs/myplc.var-log.<plcname>/*
1905         # (1.b) get the plc's  /var/lib/pgsql/data/pg_log/ -> logs/myplc.pgsql-log.<plcname>/*
1906         # (1.c) get the plc's /root/sfi -> logs/sfi.<plcname>/
1907         # (2) get all the nodes qemu log and store it as logs/node.qemu.<node>.log
1908         # (3) get the nodes /var/log and store is as logs/node.var-log.<node>/*
1909         # (4) as far as possible get the slice's /var/log as logs/sliver.var-log.<sliver>/*
1910         # (1.a)
1911         print("-------------------- TestPlc.gather_logs : PLC's /var/log")
1912         self.gather_var_logs()
1913         # (1.b)
1914         print("-------------------- TestPlc.gather_logs : PLC's /var/lib/psql/data/pg_log/")
1915         self.gather_pgsql_logs()
1916         # (1.c)
1917         print("-------------------- TestPlc.gather_logs : PLC's /root/sfi/")
1918         self.gather_root_sfi()
1919         # (2)
1920         print("-------------------- TestPlc.gather_logs : nodes's QEMU logs")
1921         for site_spec in self.plc_spec['sites']:
1922             test_site = TestSite(self,site_spec)
1923             for node_spec in site_spec['nodes']:
1924                 test_node = TestNode(self, test_site, node_spec)
1925                 test_node.gather_qemu_logs()
1926         # (3)
1927         print("-------------------- TestPlc.gather_logs : nodes's /var/log")
1928         self.gather_nodes_var_logs()
1929         # (4)
1930         print("-------------------- TestPlc.gather_logs : sample sliver's /var/log")
1931         self.gather_slivers_var_logs()
1932         return True
1933
1934     def gather_slivers_var_logs(self):
1935         for test_sliver in self.all_sliver_objs():
1936             remote = test_sliver.tar_var_logs()
1937             utils.system("mkdir -p logs/sliver.var-log.{}".format(test_sliver.name()))
1938             command = remote + " | tar -C logs/sliver.var-log.{} -xf -".format(test_sliver.name())
1939             utils.system(command)
1940         return True
1941
1942     def gather_var_logs(self):
1943         utils.system("mkdir -p logs/myplc.var-log.{}".format(self.name()))
1944         to_plc = self.actual_command_in_guest("tar -C /var/log/ -cf - .")
1945         command = to_plc + "| tar -C logs/myplc.var-log.{} -xf -".format(self.name())
1946         utils.system(command)
1947         command = "chmod a+r,a+x logs/myplc.var-log.{}/httpd".format(self.name())
1948         utils.system(command)
1949
1950     def gather_pgsql_logs(self):
1951         utils.system("mkdir -p logs/myplc.pgsql-log.{}".format(self.name()))
1952         to_plc = self.actual_command_in_guest("tar -C /var/lib/pgsql/data/pg_log/ -cf - .")
1953         command = to_plc + "| tar -C logs/myplc.pgsql-log.{} -xf -".format(self.name())
1954         utils.system(command)
1955
1956     def gather_root_sfi(self):
1957         utils.system("mkdir -p logs/sfi.{}".format(self.name()))
1958         to_plc = self.actual_command_in_guest("tar -C /root/sfi/ -cf - .")
1959         command = to_plc + "| tar -C logs/sfi.{} -xf -".format(self.name())
1960         utils.system(command)
1961
1962     def gather_nodes_var_logs(self):
1963         for site_spec in self.plc_spec['sites']:
1964             test_site = TestSite(self, site_spec)
1965             for node_spec in site_spec['nodes']:
1966                 test_node = TestNode(self, test_site, node_spec)
1967                 test_ssh = TestSsh(test_node.name(), key="keys/key_admin.rsa")
1968                 command = test_ssh.actual_command("tar -C /var/log -cf - .")
1969                 command = command + "| tar -C logs/node.var-log.{} -xf -".format(test_node.name())
1970                 utils.system("mkdir -p logs/node.var-log.{}".format(test_node.name()))
1971                 utils.system(command)
1972
1973
1974     # returns the filename to use for sql dump/restore, using options.dbname if set
1975     def dbfile(self, database):
1976         # uses options.dbname if it is found
1977         try:
1978             name = self.options.dbname
1979             if not isinstance(name, str):
1980                 raise Exception
1981         except:
1982             t = datetime.now()
1983             d = t.date()
1984             name = str(d)
1985         return "/root/{}-{}.sql".format(database, name)
1986
1987     def plc_db_dump(self):
1988         'dump the planetlab5 DB in /root in the PLC - filename has time'
1989         dump=self.dbfile("planetab5")
1990         self.run_in_guest('pg_dump -U pgsqluser planetlab5 -f '+ dump)
1991         utils.header('Dumped planetlab5 database in {}'.format(dump))
1992         return True
1993
1994     def plc_db_restore(self):
1995         'restore the planetlab5 DB - looks broken, but run -n might help'
1996         dump = self.dbfile("planetab5")
1997         self.run_in_guest('systemctl stop httpd')
1998         # xxx - need another wrapper
1999         self.run_in_guest_piped('echo drop database planetlab5', 'psql --user=pgsqluser template1')
2000         self.run_in_guest('createdb -U postgres --encoding=UNICODE --owner=pgsqluser planetlab5')
2001         self.run_in_guest('psql -U pgsqluser planetlab5 -f ' + dump)
2002         ##starting httpd service
2003         self.run_in_guest('systemctl start httpd')
2004
2005         utils.header('Database restored from ' + dump)
2006
2007     @staticmethod
2008     def create_ignore_steps():
2009         for step in TestPlc.default_steps + TestPlc.other_steps:
2010             # default step can have a plc qualifier
2011             if '@' in step:
2012                 step, qualifier = step.split('@')
2013             # or be defined as forced or ignored by default
2014             for keyword in ['_ignore', '_force']:
2015                 if step.endswith(keyword):
2016                     step=step.replace(keyword,'')
2017             if step == SEP or step == SEPSFA :
2018                 continue
2019             method = getattr(TestPlc,step)
2020             name = step + '_ignore'
2021             wrapped = ignore_result(method)
2022 #            wrapped.__doc__ = method.__doc__ + " (run in ignore-result mode)"
2023             setattr(TestPlc, name, wrapped)
2024
2025 #    @ignore_result
2026 #    def ssh_slice_again_ignore (self): pass
2027 #    @ignore_result
2028 #    def check_initscripts_ignore (self): pass
2029
2030     def standby_1_through_20(self):
2031         """convenience function to wait for a specified number of minutes"""
2032         pass
2033     @standby_generic
2034     def standby_1(): pass
2035     @standby_generic
2036     def standby_2(): pass
2037     @standby_generic
2038     def standby_3(): pass
2039     @standby_generic
2040     def standby_4(): pass
2041     @standby_generic
2042     def standby_5(): pass
2043     @standby_generic
2044     def standby_6(): pass
2045     @standby_generic
2046     def standby_7(): pass
2047     @standby_generic
2048     def standby_8(): pass
2049     @standby_generic
2050     def standby_9(): pass
2051     @standby_generic
2052     def standby_10(): pass
2053     @standby_generic
2054     def standby_11(): pass
2055     @standby_generic
2056     def standby_12(): pass
2057     @standby_generic
2058     def standby_13(): pass
2059     @standby_generic
2060     def standby_14(): pass
2061     @standby_generic
2062     def standby_15(): pass
2063     @standby_generic
2064     def standby_16(): pass
2065     @standby_generic
2066     def standby_17(): pass
2067     @standby_generic
2068     def standby_18(): pass
2069     @standby_generic
2070     def standby_19(): pass
2071     @standby_generic
2072     def standby_20(): pass
2073
2074     # convenience for debugging the test logic
2075     def yes(self): return True
2076     def no(self): return False
2077     def fail(self): return False