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-4".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(f"pip3 install {package} || pip install {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         ##########
 707         # on the virsh containers, DNS resolution using gethostbyaddr
 708         # won't work fine, for the hosts under .pl.sophia.inria.fr
 709         # although these IPs can be reversed from virtually everywhere else
 710         #
 711         # this has started with something around fedora35 so I am suspecting python-3.10
 712         #
 713         # in any case, here's a quick and dirty workaround, as I have bumped my head
 714         # against the screen for two good hours and not found any single clue
 715         # about how to deal with this properly
 716
 717         import subprocess
 718
 719         def workaround_gethostaddr(ip):
 720             command = f"host {ip} 8.8.8.8"
 721             completed = subprocess.run(command, shell=True, capture_output=True)
 722             pieces = completed.stdout.decode().split("domain name pointer ")
 723             if len(pieces) == 2:
 724                 return pieces[1].replace(".\n", "")
 725             else:
 726                 return None
 727
 728         # invoke initvm (drop support for vs)
 729         script = "lbuild-initvm.sh"
 730         script_options = ""
 731         # pass the vbuild-nightly options to [lv]test-initvm
 732         script_options += " -p {}".format(self.options.personality)
 733         script_options += " -d {}".format(self.options.pldistro)
 734         script_options += " -f {}".format(self.options.fcdistro)
 735         script_options += " -r {}".format(repo_url)
 736         vserver_name = self.vservername
 737         try:
 738             vserver_hostname = socket.gethostbyaddr(self.vserverip)[0]
 739         except:
 740             # read more above about this workaround
 741             vserver_hostname = workaround_gethostaddr(self.vserverip)
 742             if not vserver_hostname:
 743                 print("Cannot reverse lookup {}".format(self.vserverip))
 744                 print("This is considered fatal, as this might pollute the test results")
 745                 return False
 746         script_options += " -n {}".format(vserver_hostname)
 747         create_vserver="{build_dir}/{script} {script_options} {vserver_name}".format(**locals())
 748         return self.run_in_host(create_vserver) == 0
 749
 750     ### install django through pip
 751     def django_install(self):
 752         # plcapi requires Django, that is no longer provided py fedora as an rpm
 753         # so we use pip instead
 754         """
 755         pip install Django
 756         """
 757         return self.pip3_install('Django')
 758
 759     ### install_rpm
 760     def plc_install(self):
 761         """
 762         yum install myplc, noderepo
 763         """
 764
 765         # compute nodefamily
 766         if self.options.personality == "linux32":
 767             arch = "i386"
 768         elif self.options.personality == "linux64":
 769             arch = "x86_64"
 770         else:
 771             raise Exception("Unsupported personality {}".format(self.options.personality))
 772         nodefamily = "{}-{}-{}".format(self.options.pldistro, self.options.fcdistro, arch)
 773
 774         # check it's possible to install just 'myplc-core' first
 775         if not self.dnf_install("myplc-core"):
 776             return False
 777
 778         pkgs_list = []
 779         pkgs_list.append("myplc")
 780         # pkgs_list.append("slicerepo-{}".format(nodefamily))
 781         # pkgs_list.append("noderepo-{}".format(nodefamily))
 782         pkgs_string=" ".join(pkgs_list)
 783         return self.dnf_install(pkgs_list)
 784
 785     def install_syslinux6(self):
 786         """
 787         install syslinux6 from the fedora21 release
 788         """
 789         key = 'http://mirror.onelab.eu/keys/RPM-GPG-KEY-fedora-21-primary'
 790
 791         rpms = [
 792             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-6.03-1.fc21.x86_64.rpm',
 793             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-nonlinux-6.03-1.fc21.noarch.rpm',
 794             'http://mirror.onelab.eu/fedora/releases/21/Everything/x86_64/os/Packages/s/syslinux-perl-6.03-1.fc21.x86_64.rpm',
 795         ]
 796         # this can be done several times
 797         self.run_in_guest("rpm --import {key}".format(**locals()))
 798         return self.run_in_guest("yum -y localinstall {}".format(" ".join(rpms))) == 0
 799
 800     def bonding_builds(self):
 801         """
 802         list /etc/yum.repos.d on the myplc side
 803         """
 804         self.run_in_guest("ls /etc/yum.repos.d/*partial.repo")
 805         return True
 806
 807     def bonding_nodes(self):
 808         """
 809         List nodes known to the myplc together with their nodefamiliy
 810         """
 811         print("---------------------------------------- nodes")
 812         for node in self.apiserver.GetNodes(self.auth_root()):
 813             print("{} -> {}".format(node['hostname'],
 814                                     self.apiserver.GetNodeFlavour(self.auth_root(),node['hostname'])['nodefamily']))
 815         print("---------------------------------------- nodes")
 816
 817
 818     ###
 819     def mod_python(self):
 820         """yum install mod_python, useful on f18 and above so as to avoid broken wsgi"""
 821         return self.dnf_install( ['mod_python'] )
 822
 823     ###
 824     def plc_configure(self):
 825         "run plc-config-tty"
 826         tmpname = '{}.plc-config-tty'.format(self.name())
 827         with open(tmpname,'w') as fileconf:
 828             for var, value in self.plc_spec['settings'].items():
 829                 fileconf.write('e {}\n{}\n'.format(var, value))
 830             fileconf.write('w\n')
 831             fileconf.write('q\n')
 832         utils.system('cat {}'.format(tmpname))
 833         self.run_in_guest_piped('cat {}'.format(tmpname), 'plc-config-tty')
 834         utils.system('rm {}'.format(tmpname))
 835         return True
 836
 837     # care only about f>=27
 838     def start_stop_systemd(self, service, start_or_stop):
 839         "utility to start/stop a systemd-defined service (sfa)"
 840         return self.run_in_guest("systemctl {} {}".format(start_or_stop, service)) == 0
 841
 842     def plc_start(self):
 843         "start plc through systemclt"
 844         return self.start_stop_systemd('plc', 'start')
 845
 846     def plc_stop(self):
 847         "stop plc through systemctl"
 848         return self.start_stop_systemd('plc', 'stop')
 849
 850     def plcvm_start(self):
 851         "start the PLC vserver"
 852         self.start_guest()
 853         return True
 854
 855     def plcvm_stop(self):
 856         "stop the PLC vserver"
 857         self.stop_guest()
 858         return True
 859
 860     # stores the keys from the config for further use
 861     def keys_store(self):
 862         "stores test users ssh keys in keys/"
 863         for key_spec in self.plc_spec['keys']:
 864                 TestKey(self,key_spec).store_key()
 865         return True
 866
 867     def keys_clean(self):
 868         "removes keys cached in keys/"
 869         utils.system("rm -rf ./keys")
 870         return True
 871
 872     # fetches the ssh keys in the plc's /etc/planetlab and stores them in keys/
 873     # for later direct access to the nodes
 874     def keys_fetch(self):
 875         "gets ssh keys in /etc/planetlab/ and stores them locally in keys/"
 876         dir="./keys"
 877         if not os.path.isdir(dir):
 878             os.mkdir(dir)
 879         vservername = self.vservername
 880         vm_root = self.vm_root_in_host()
 881         overall = True
 882         prefix = 'debug_ssh_key'
 883         for ext in ['pub', 'rsa'] :
 884             src = "{vm_root}/etc/planetlab/{prefix}.{ext}".format(**locals())
 885             dst = "keys/{vservername}-debug.{ext}".format(**locals())
 886             if self.test_ssh.fetch(src, dst) != 0:
 887                 overall=False
 888         return overall
 889
 890     def sites(self):
 891         "create sites with PLCAPI"
 892         return self.do_sites()
 893
 894     def delete_sites(self):
 895         "delete sites with PLCAPI"
 896         return self.do_sites(action="delete")
 897
 898     def do_sites(self, action="add"):
 899         for site_spec in self.plc_spec['sites']:
 900             test_site = TestSite(self,site_spec)
 901             if (action != "add"):
 902                 utils.header("Deleting site {} in {}".format(test_site.name(), self.name()))
 903                 test_site.delete_site()
 904                 # deleted with the site
 905                 #test_site.delete_users()
 906                 continue
 907             else:
 908                 utils.header("Creating site {} & users in {}".format(test_site.name(), self.name()))
 909                 test_site.create_site()
 910                 test_site.create_users()
 911         return True
 912
 913     def delete_all_sites(self):
 914         "Delete all sites in PLC, and related objects"
 915         print('auth_root', self.auth_root())
 916         sites = self.apiserver.GetSites(self.auth_root(), {}, ['site_id', 'login_base'])
 917         for site in sites:
 918             # keep automatic site - otherwise we shoot in our own foot, root_auth is not valid anymore
 919             if site['login_base'] == self.plc_spec['settings']['PLC_SLICE_PREFIX']:
 920                 continue
 921             site_id = site['site_id']
 922             print('Deleting site_id', site_id)
 923             self.apiserver.DeleteSite(self.auth_root(), site_id)
 924         return True
 925
 926     def nodes(self):
 927         "create nodes with PLCAPI"
 928         return self.do_nodes()
 929     def delete_nodes(self):
 930         "delete nodes with PLCAPI"
 931         return self.do_nodes(action="delete")
 932
 933     def do_nodes(self, action="add"):
 934         for site_spec in self.plc_spec['sites']:
 935             test_site = TestSite(self, site_spec)
 936             if action != "add":
 937                 utils.header("Deleting nodes in site {}".format(test_site.name()))
 938                 for node_spec in site_spec['nodes']:
 939                     test_node = TestNode(self, test_site, node_spec)
 940                     utils.header("Deleting {}".format(test_node.name()))
 941                     test_node.delete_node()
 942             else:
 943                 utils.header("Creating nodes for site {} in {}".format(test_site.name(), self.name()))
 944                 for node_spec in site_spec['nodes']:
 945                     utils.pprint('Creating node {}'.format(node_spec), node_spec)
 946                     test_node = TestNode(self, test_site, node_spec)
 947                     test_node.create_node()
 948         return True
 949
 950     def nodegroups(self):
 951         "create nodegroups with PLCAPI"
 952         return self.do_nodegroups("add")
 953     def delete_nodegroups(self):
 954         "delete nodegroups with PLCAPI"
 955         return self.do_nodegroups("delete")
 956
 957     YEAR = 365*24*3600
 958     @staticmethod
 959     def translate_timestamp(start, grain, timestamp):
 960         if timestamp < TestPlc.YEAR:
 961             return start + timestamp*grain
 962         else:
 963             return timestamp
 964
 965     @staticmethod
 966     def timestamp_printable(timestamp):
 967         return time.strftime('%m-%d %H:%M:%S UTC', time.gmtime(timestamp))
 968
 969     def leases(self):
 970         "create leases (on reservable nodes only, use e.g. run -c default -c resa)"
 971         now = int(time.time())
 972         grain = self.apiserver.GetLeaseGranularity(self.auth_root())
 973         print('API answered grain=', grain)
 974         start = (now//grain)*grain
 975         start += grain
 976         # find out all nodes that are reservable
 977         nodes = self.all_reservable_nodenames()
 978         if not nodes:
 979             utils.header("No reservable node found - proceeding without leases")
 980             return True
 981         ok = True
 982         # attach them to the leases as specified in plc_specs
 983         # this is where the 'leases' field gets interpreted as relative of absolute
 984         for lease_spec in self.plc_spec['leases']:
 985             # skip the ones that come with a null slice id
 986             if not lease_spec['slice']:
 987                 continue
 988             lease_spec['t_from']  = TestPlc.translate_timestamp(start, grain, lease_spec['t_from'])
 989             lease_spec['t_until'] = TestPlc.translate_timestamp(start, grain, lease_spec['t_until'])
 990             lease_addition = self.apiserver.AddLeases(self.auth_root(), nodes, lease_spec['slice'],
 991                                                       lease_spec['t_from'], lease_spec['t_until'])
 992             if lease_addition['errors']:
 993                 utils.header("Cannot create leases, {}".format(lease_addition['errors']))
 994                 ok = False
 995             else:
 996                 utils.header('Leases on nodes {} for {} from {:d} ({}) until {:d} ({})'\
 997                              .format(nodes, lease_spec['slice'],
 998                                      lease_spec['t_from'],  TestPlc.timestamp_printable(lease_spec['t_from']),
 999                                      lease_spec['t_until'], TestPlc.timestamp_printable(lease_spec['t_until'])))
1000
1001         return ok
1002
1003     def delete_leases(self):
1004         "remove all leases in the myplc side"
1005         lease_ids = [ l['lease_id'] for l in self.apiserver.GetLeases(self.auth_root())]
1006         utils.header("Cleaning leases {}".format(lease_ids))
1007         self.apiserver.DeleteLeases(self.auth_root(), lease_ids)
1008         return True
1009
1010     def list_leases(self):
1011         "list all leases known to the myplc"
1012         leases = self.apiserver.GetLeases(self.auth_root())
1013         now = int(time.time())
1014         for l in leases:
1015             current = l['t_until'] >= now
1016             if self.options.verbose or current:
1017                 utils.header("{} {} from {} until {}"\
1018                              .format(l['hostname'], l['name'],
1019                                      TestPlc.timestamp_printable(l['t_from']),
1020                                      TestPlc.timestamp_printable(l['t_until'])))
1021         return True
1022
1023     # create nodegroups if needed, and populate
1024     def do_nodegroups(self, action="add"):
1025         # 1st pass to scan contents
1026         groups_dict = {}
1027         for site_spec in self.plc_spec['sites']:
1028             test_site = TestSite(self,site_spec)
1029             for node_spec in site_spec['nodes']:
1030                 test_node = TestNode(self, test_site, node_spec)
1031                 if 'nodegroups' in node_spec:
1032                     nodegroupnames = node_spec['nodegroups']
1033                     if isinstance(nodegroupnames, str):
1034                         nodegroupnames = [ nodegroupnames ]
1035                     for nodegroupname in nodegroupnames:
1036                         if nodegroupname not in groups_dict:
1037                             groups_dict[nodegroupname] = []
1038                         groups_dict[nodegroupname].append(test_node.name())
1039         auth = self.auth_root()
1040         overall = True
1041         for (nodegroupname,group_nodes) in groups_dict.items():
1042             if action == "add":
1043                 print('nodegroups:', 'dealing with nodegroup',\
1044                     nodegroupname, 'on nodes', group_nodes)
1045                 # first, check if the nodetagtype is here
1046                 tag_types = self.apiserver.GetTagTypes(auth, {'tagname':nodegroupname})
1047                 if tag_types:
1048                     tag_type_id = tag_types[0]['tag_type_id']
1049                 else:
1050                     tag_type_id = self.apiserver.AddTagType(auth,
1051                                                             {'tagname' : nodegroupname,
1052                                                              'description' : 'for nodegroup {}'.format(nodegroupname),
1053                                                              'category' : 'test'})
1054                 print('located tag (type)', nodegroupname, 'as', tag_type_id)
1055                 # create nodegroup
1056                 nodegroups = self.apiserver.GetNodeGroups(auth, {'groupname' : nodegroupname})
1057                 if not nodegroups:
1058                     self.apiserver.AddNodeGroup(auth, nodegroupname, tag_type_id, 'yes')
1059                     print('created nodegroup', nodegroupname, \
1060                         'from tagname', nodegroupname, 'and value', 'yes')
1061                 # set node tag on all nodes, value='yes'
1062                 for nodename in group_nodes:
1063                     try:
1064                         self.apiserver.AddNodeTag(auth, nodename, nodegroupname, "yes")
1065                     except:
1066                         traceback.print_exc()
1067                         print('node', nodename, 'seems to already have tag', nodegroupname)
1068                     # check anyway
1069                     try:
1070                         expect_yes = self.apiserver.GetNodeTags(auth,
1071                                                                 {'hostname' : nodename,
1072                                                                  'tagname'  : nodegroupname},
1073                                                                 ['value'])[0]['value']
1074                         if expect_yes != "yes":
1075                             print('Mismatch node tag on node',nodename,'got',expect_yes)
1076                             overall = False
1077                     except:
1078                         if not self.options.dry_run:
1079                             print('Cannot find tag', nodegroupname, 'on node', nodename)
1080                             overall = False
1081             else:
1082                 try:
1083                     print('cleaning nodegroup', nodegroupname)
1084                     self.apiserver.DeleteNodeGroup(auth, nodegroupname)
1085                 except:
1086                     traceback.print_exc()
1087                     overall = False
1088         return overall
1089
1090     # a list of TestNode objs
1091     def all_nodes(self):
1092         nodes=[]
1093         for site_spec in self.plc_spec['sites']:
1094             test_site = TestSite(self,site_spec)
1095             for node_spec in site_spec['nodes']:
1096                 nodes.append(TestNode(self, test_site, node_spec))
1097         return nodes
1098
1099     # return a list of tuples (nodename,qemuname)
1100     def all_node_infos(self) :
1101         node_infos = []
1102         for site_spec in self.plc_spec['sites']:
1103             node_infos += [ (node_spec['node_fields']['hostname'], node_spec['host_box']) \
1104                                 for node_spec in site_spec['nodes'] ]
1105         return node_infos
1106
1107     def all_nodenames(self):
1108         return [ x[0] for x in self.all_node_infos() ]
1109     def all_reservable_nodenames(self):
1110         res = []
1111         for site_spec in self.plc_spec['sites']:
1112             for node_spec in site_spec['nodes']:
1113                 node_fields = node_spec['node_fields']
1114                 if 'node_type' in node_fields and node_fields['node_type'] == 'reservable':
1115                     res.append(node_fields['hostname'])
1116         return res
1117
1118     # silent_minutes : during the first <silent_minutes> minutes nothing gets printed
1119     def nodes_check_boot_state(self, target_boot_state, timeout_minutes,
1120                                silent_minutes, period_seconds = 15):
1121         if self.options.dry_run:
1122             print('dry_run')
1123             return True
1124
1125         class CompleterTaskBootState(CompleterTask):
1126             def __init__(self, test_plc, hostname):
1127                 self.test_plc = test_plc
1128                 self.hostname = hostname
1129                 self.last_boot_state = 'undef'
1130             def actual_run(self):
1131                 try:
1132                     node = self.test_plc.apiserver.GetNodes(self.test_plc.auth_root(),
1133                                                             [ self.hostname ],
1134                                                             ['boot_state'])[0]
1135                     self.last_boot_state = node['boot_state']
1136                     return self.last_boot_state == target_boot_state
1137                 except:
1138                     return False
1139             def message(self):
1140                 return "CompleterTaskBootState with node {}".format(self.hostname)
1141             def failure_epilogue(self):
1142                 print("node {} in state {} - expected {}"\
1143                     .format(self.hostname, self.last_boot_state, target_boot_state))
1144
1145         timeout = timedelta(minutes=timeout_minutes)
1146         graceout = timedelta(minutes=silent_minutes)
1147         period   = timedelta(seconds=period_seconds)
1148         # the nodes that haven't checked yet - start with a full list and shrink over time
1149         utils.header("checking nodes boot state (expected {})".format(target_boot_state))
1150         tasks = [ CompleterTaskBootState(self,hostname) \
1151                       for (hostname,_) in self.all_node_infos() ]
1152         message = 'check_boot_state={}'.format(target_boot_state)
1153         return Completer(tasks, message=message).run(timeout, graceout, period)
1154
1155     def nodes_booted(self):
1156         return self.nodes_check_boot_state('boot', timeout_minutes=30, silent_minutes=28)
1157
1158     def probe_kvm_iptables(self):
1159         (_,kvmbox) = self.all_node_infos()[0]
1160         TestSsh(kvmbox).run("iptables-save")
1161         return True
1162
1163     # probing nodes
1164     def check_nodes_ping(self, timeout_seconds=60, period_seconds=10):
1165         class CompleterTaskPingNode(CompleterTask):
1166             def __init__(self, hostname):
1167                 self.hostname = hostname
1168             def run(self, silent):
1169                 command="ping -c 1 -w 1 {} >& /dev/null".format(self.hostname)
1170                 return utils.system(command, silent=silent) == 0
1171             def failure_epilogue(self):
1172                 print("Cannot ping node with name {}".format(self.hostname))
1173         timeout = timedelta(seconds = timeout_seconds)
1174         graceout = timeout
1175         period = timedelta(seconds = period_seconds)
1176         node_infos = self.all_node_infos()
1177         tasks = [ CompleterTaskPingNode(h) for (h,_) in node_infos ]
1178         return Completer(tasks, message='ping_node').run(timeout, graceout, period)
1179
1180     # ping node before we try to reach ssh, helpful for troubleshooting failing bootCDs
1181     def ping_node(self):
1182         "Ping nodes"
1183         return self.check_nodes_ping()
1184
1185     def check_nodes_ssh(self, debug, timeout_minutes, silent_minutes, period_seconds=15):
1186         # various delays
1187         timeout  = timedelta(minutes=timeout_minutes)
1188         graceout = timedelta(minutes=silent_minutes)
1189         period   = timedelta(seconds=period_seconds)
1190         vservername = self.vservername
1191         if debug:
1192             message = "debug"
1193             completer_message = 'ssh_node_debug'
1194             local_key = "keys/{vservername}-debug.rsa".format(**locals())
1195         else:
1196             message = "boot"
1197             completer_message = 'ssh_node_boot'
1198             local_key = "keys/key_admin.rsa"
1199         utils.header("checking ssh access to nodes (expected in {} mode)".format(message))
1200         node_infos = self.all_node_infos()
1201         tasks = [ CompleterTaskNodeSsh(nodename, qemuname, local_key,
1202                                         boot_state=message, dry_run=self.options.dry_run) \
1203                       for (nodename, qemuname) in node_infos ]
1204         return Completer(tasks, message=completer_message).run(timeout, graceout, period)
1205
1206     def ssh_node_debug(self):
1207         "Tries to ssh into nodes in debug mode with the debug ssh key"
1208         return self.check_nodes_ssh(debug = True,
1209                                     timeout_minutes = self.ssh_node_debug_timeout,
1210                                     silent_minutes = self.ssh_node_debug_silent)
1211
1212     def ssh_node_boot(self):
1213         "Tries to ssh into nodes in production mode with the root ssh key"
1214         return self.check_nodes_ssh(debug = False,
1215                                     timeout_minutes = self.ssh_node_boot_timeout,
1216                                     silent_minutes = self.ssh_node_boot_silent)
1217
1218     def node_bmlogs(self):
1219         "Checks that there's a non-empty dir. /var/log/bm/raw"
1220         return utils.system(self.actual_command_in_guest("ls /var/log/bm/raw")) == 0
1221
1222     @node_mapper
1223     def qemu_local_init(self): pass
1224     @node_mapper
1225     def bootcd(self): pass
1226     @node_mapper
1227     def qemu_local_config(self): pass
1228     @node_mapper
1229     def qemu_export(self): pass
1230     @node_mapper
1231     def qemu_cleanlog(self): pass
1232     @node_mapper
1233     def nodestate_reinstall(self): pass
1234     @node_mapper
1235     def nodestate_upgrade(self): pass
1236     @node_mapper
1237     def nodestate_safeboot(self): pass
1238     @node_mapper
1239     def nodestate_boot(self): pass
1240     @node_mapper
1241     def nodestate_show(self): pass
1242     @node_mapper
1243     def nodedistro_f14(self): pass
1244     @node_mapper
1245     def nodedistro_f18(self): pass
1246     @node_mapper
1247     def nodedistro_f20(self): pass
1248     @node_mapper
1249     def nodedistro_f21(self): pass
1250     @node_mapper
1251     def nodedistro_f22(self): pass
1252     @node_mapper
1253     def nodedistro_show(self): pass
1254
1255     ### check hooks : invoke scripts from hooks/{node,slice}
1256     def check_hooks_node(self):
1257         return self.locate_first_node().check_hooks()
1258     def check_hooks_sliver(self) :
1259         return self.locate_first_sliver().check_hooks()
1260
1261     def check_hooks(self):
1262         "runs unit tests in the node and slice contexts - see hooks/{node,slice}"
1263         return self.check_hooks_node() and self.check_hooks_sliver()
1264
1265     ### initscripts
1266     def do_check_initscripts(self):
1267         class CompleterTaskInitscript(CompleterTask):
1268             def __init__(self, test_sliver, stamp):
1269                 self.test_sliver = test_sliver
1270                 self.stamp = stamp
1271             def actual_run(self):
1272                 return self.test_sliver.check_initscript_stamp(self.stamp)
1273             def message(self):
1274                 return "initscript checker for {}".format(self.test_sliver.name())
1275             def failure_epilogue(self):
1276                 print("initscript stamp {} not found in sliver {}"\
1277                     .format(self.stamp, self.test_sliver.name()))
1278
1279         tasks = []
1280         for slice_spec in self.plc_spec['slices']:
1281             if 'initscriptstamp' not in slice_spec:
1282                 continue
1283             stamp = slice_spec['initscriptstamp']
1284             slicename = slice_spec['slice_fields']['name']
1285             for nodename in slice_spec['nodenames']:
1286                 print('nodename', nodename, 'slicename', slicename, 'stamp', stamp)
1287                 site,node = self.locate_node(nodename)
1288                 # xxx - passing the wrong site - probably harmless
1289                 test_site = TestSite(self, site)
1290                 test_slice = TestSlice(self, test_site, slice_spec)
1291                 test_node = TestNode(self, test_site, node)
1292                 test_sliver = TestSliver(self, test_node, test_slice)
1293                 tasks.append(CompleterTaskInitscript(test_sliver, stamp))
1294         return Completer(tasks, message='check_initscripts').\
1295             run (timedelta(minutes=5), timedelta(minutes=4), timedelta(seconds=10))
1296
1297     def check_initscripts(self):
1298         "check that the initscripts have triggered"
1299         return self.do_check_initscripts()
1300
1301     def initscripts(self):
1302         "create initscripts with PLCAPI"
1303         for initscript in self.plc_spec['initscripts']:
1304             utils.pprint('Adding Initscript in plc {}'.format(self.plc_spec['name']), initscript)
1305             self.apiserver.AddInitScript(self.auth_root(), initscript['initscript_fields'])
1306         return True
1307
1308     def delete_initscripts(self):
1309         "delete initscripts with PLCAPI"
1310         for initscript in self.plc_spec['initscripts']:
1311             initscript_name = initscript['initscript_fields']['name']
1312             print(('Attempting to delete {} in plc {}'.format(initscript_name, self.plc_spec['name'])))
1313             try:
1314                 self.apiserver.DeleteInitScript(self.auth_root(), initscript_name)
1315                 print(initscript_name, 'deleted')
1316             except:
1317                 print('deletion went wrong - probably did not exist')
1318         return True
1319
1320     ### manage slices
1321     def slices(self):
1322         "create slices with PLCAPI"
1323         return self.do_slices(action="add")
1324
1325     def delete_slices(self):
1326         "delete slices with PLCAPI"
1327         return self.do_slices(action="delete")
1328
1329     def fill_slices(self):
1330         "add nodes in slices with PLCAPI"
1331         return self.do_slices(action="fill")
1332
1333     def empty_slices(self):
1334         "remove nodes from slices with PLCAPI"
1335         return self.do_slices(action="empty")
1336
1337     def do_slices(self,  action="add"):
1338         for slice in self.plc_spec['slices']:
1339             site_spec = self.locate_site(slice['sitename'])
1340             test_site = TestSite(self,site_spec)
1341             test_slice=TestSlice(self,test_site,slice)
1342             if action == "delete":
1343                 test_slice.delete_slice()
1344             elif action == "fill":
1345                 test_slice.add_nodes()
1346             elif action == "empty":
1347                 test_slice.delete_nodes()
1348             else:
1349                 test_slice.create_slice()
1350         return True
1351
1352     @slice_mapper__tasks(20, 10, 15)
1353     def ssh_slice(self): pass
1354     @slice_mapper__tasks(20, 19, 15)
1355     def ssh_slice_off(self): pass
1356     @slice_mapper__tasks(1, 1, 15)
1357     def slice_fs_present(self): pass
1358     @slice_mapper__tasks(1, 1, 15)
1359     def slice_fs_deleted(self): pass
1360
1361     # use another name so we can exclude/ignore it from the tests on the nightly command line
1362     def ssh_slice_again(self): return self.ssh_slice()
1363     # note that simply doing ssh_slice_again=ssh_slice would kind of work too
1364     # but for some reason the ignore-wrapping thing would not
1365
1366     @slice_mapper
1367     def ssh_slice_basics(self): pass
1368     @slice_mapper
1369     def check_vsys_defaults(self): pass
1370
1371     @node_mapper
1372     def keys_clear_known_hosts(self): pass
1373
1374     def plcapi_urls(self):
1375         """
1376         attempts to reach the PLCAPI with various forms for the URL
1377         """
1378         return PlcapiUrlScanner(self.auth_root(), ip=self.vserverip).scan()
1379
1380     def speed_up_slices(self):
1381         "tweak nodemanager cycle (wait time) to 30+/-10 s"
1382         return self._speed_up_slices (30, 10)
1383     def super_speed_up_slices(self):
1384         "dev mode: tweak nodemanager cycle (wait time) to 5+/-1 s"
1385         return self._speed_up_slices(5, 1)
1386
1387     def _speed_up_slices(self, p, r):
1388         # create the template on the server-side
1389         template = "{}.nodemanager".format(self.name())
1390         with open(template,"w") as template_file:
1391             template_file.write('OPTIONS="-p {} -r {} -d"\n'.format(p, r))
1392         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1393         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1394         self.test_ssh.copy_abs(template, remote)
1395         # Add a conf file
1396         if not self.apiserver.GetConfFiles(self.auth_root(),
1397                                            {'dest' : '/etc/sysconfig/nodemanager'}):
1398             self.apiserver.AddConfFile(self.auth_root(),
1399                                         {'dest' : '/etc/sysconfig/nodemanager',
1400                                          'source' : 'PlanetLabConf/nodemanager',
1401                                          'postinstall_cmd' : 'service nm restart',})
1402         return True
1403
1404     def debug_nodemanager(self):
1405         "sets verbose mode for nodemanager, and speeds up cycle even more (needs speed_up_slices first)"
1406         template = "{}.nodemanager".format(self.name())
1407         with open(template,"w") as template_file:
1408             template_file.write('OPTIONS="-p 10 -r 6 -v -d"\n')
1409         in_vm = "/var/www/html/PlanetLabConf/nodemanager"
1410         remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1411         self.test_ssh.copy_abs(template, remote)
1412         return True
1413
1414     @node_mapper
1415     def qemu_start(self) : pass
1416
1417     @node_mapper
1418     def qemu_timestamp(self) : pass
1419
1420     @node_mapper
1421     def qemu_nodefamily(self): pass
1422
1423     # when a spec refers to a node possibly on another plc
1424     def locate_sliver_obj_cross(self, nodename, slicename, other_plcs):
1425         for plc in [ self ] + other_plcs:
1426             try:
1427                 return plc.locate_sliver_obj(nodename, slicename)
1428             except:
1429                 pass
1430         raise Exception("Cannot locate sliver {}@{} among all PLCs".format(nodename, slicename))
1431
1432     # implement this one as a cross step so that we can take advantage of different nodes
1433     # in multi-plcs mode
1434     def cross_check_tcp(self, other_plcs):
1435         "check TCP connectivity between 2 slices (or in loopback if only one is defined)"
1436         if 'tcp_specs' not in self.plc_spec or not self.plc_spec['tcp_specs']:
1437             utils.header("check_tcp: no/empty config found")
1438             return True
1439         specs = self.plc_spec['tcp_specs']
1440         overall = True
1441
1442         # first wait for the network to be up and ready from the slices
1443         class CompleterTaskNetworkReadyInSliver(CompleterTask):
1444             def __init__(self, test_sliver):
1445                 self.test_sliver = test_sliver
1446             def actual_run(self):
1447                 return self.test_sliver.check_tcp_ready(port = 9999)
1448             def message(self):
1449                 return "network ready checker for {}".format(self.test_sliver.name())
1450             def failure_epilogue(self):
1451                 print("could not bind port from sliver {}".format(self.test_sliver.name()))
1452
1453         sliver_specs = {}
1454         tasks = []
1455         managed_sliver_names = set()
1456         for spec in specs:
1457             # locate the TestSliver instances involved, and cache them in the spec instance
1458             spec['s_sliver'] = self.locate_sliver_obj_cross(spec['server_node'], spec['server_slice'], other_plcs)
1459             spec['c_sliver'] = self.locate_sliver_obj_cross(spec['client_node'], spec['client_slice'], other_plcs)
1460             message = "Will check TCP between s={} and c={}"\
1461                       .format(spec['s_sliver'].name(), spec['c_sliver'].name())
1462             if 'client_connect' in spec:
1463                 message += " (using {})".format(spec['client_connect'])
1464             utils.header(message)
1465             # we need to check network presence in both slivers, but also
1466             # avoid to insert a sliver several times
1467             for sliver in [ spec['s_sliver'], spec['c_sliver'] ]:
1468                 if sliver.name() not in managed_sliver_names:
1469                     tasks.append(CompleterTaskNetworkReadyInSliver(sliver))
1470                     # add this sliver's name in the set
1471                     managed_sliver_names .update( {sliver.name()} )
1472
1473         # wait for the netork to be OK in all server sides
1474         if not Completer(tasks, message='check for network readiness in slivers').\
1475            run(timedelta(seconds=30), timedelta(seconds=24), period=timedelta(seconds=5)):
1476             return False
1477
1478         # run server and client
1479         for spec in specs:
1480             port = spec['port']
1481             # server side
1482             # the issue here is that we have the server run in background
1483             # and so we have no clue if it took off properly or not
1484             # looks like in some cases it does not
1485             address = spec['s_sliver'].test_node.name()
1486             if not spec['s_sliver'].run_tcp_server(address, port, timeout=20):
1487                 overall = False
1488                 break
1489
1490             # idem for the client side
1491             # use nodename from located sliver, unless 'client_connect' is set
1492             if 'client_connect' in spec:
1493                 destination = spec['client_connect']
1494             else:
1495                 destination = spec['s_sliver'].test_node.name()
1496             if not spec['c_sliver'].run_tcp_client(destination, port):
1497                 overall = False
1498         return overall
1499
1500     # painfully enough, we need to allow for some time as netflow might show up last
1501     def check_system_slice(self):
1502         "all nodes: check that a system slice is alive"
1503         # netflow currently not working in the lxc distro
1504         # drl not built at all in the wtx distro
1505         # if we find either of them we're happy
1506         return self.check_netflow() or self.check_drl()
1507
1508     # expose these
1509     def check_netflow(self): return self._check_system_slice('netflow')
1510     def check_drl(self): return self._check_system_slice('drl')
1511
1512     # we have the slices up already here, so it should not take too long
1513     def _check_system_slice(self, slicename, timeout_minutes=5, period_seconds=15):
1514         class CompleterTaskSystemSlice(CompleterTask):
1515             def __init__(self, test_node, dry_run):
1516                 self.test_node = test_node
1517                 self.dry_run = dry_run
1518             def actual_run(self):
1519                 return self.test_node._check_system_slice(slicename, dry_run=self.dry_run)
1520             def message(self):
1521                 return "System slice {} @ {}".format(slicename, self.test_node.name())
1522             def failure_epilogue(self):
1523                 print("COULD not find system slice {} @ {}".format(slicename, self.test_node.name()))
1524         timeout = timedelta(minutes=timeout_minutes)
1525         silent  = timedelta(0)
1526         period  = timedelta(seconds=period_seconds)
1527         tasks = [ CompleterTaskSystemSlice(test_node, self.options.dry_run) \
1528                       for test_node in self.all_nodes() ]
1529         return Completer(tasks, message='_check_system_slice').run(timeout, silent, period)
1530
1531     def plcsh_stress_test(self):
1532         "runs PLCAPI stress test, that checks Add/Update/Delete on all types - preserves contents"
1533         # install the stress-test in the plc image
1534         location = "/usr/share/plc_api/plcsh_stress_test.py"
1535         remote = "{}/{}".format(self.vm_root_in_host(), location)
1536         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1537         command = location
1538         command += " -- --check"
1539         if self.options.size == 1:
1540             command +=  " --tiny"
1541         return self.run_in_guest(command) == 0
1542
1543     # populate runs the same utility without slightly different options
1544     # in particular runs with --preserve (dont cleanup) and without --check
1545     # also it gets run twice, once with the --foreign option for creating fake foreign entries
1546
1547     def install_pip2(self):
1548
1549         replacements = [
1550             "http://mirror.onelab.eu/third-party/python2-pip-19.1.1-7.fc33.noarch.rpm",
1551         ]
1552
1553         return (
1554                self.run_in_guest("pip2 --version") == 0
1555             or self.run_in_guest("dnf install python2-pip") == 0
1556             or self.run_in_guest("dnf localinstall -y " + " ".join(replacements)) == 0)
1557
1558
1559     def install_m2crypto(self):
1560
1561         # installing m2crypto for python2 is increasingly difficult
1562         # f29 and f31: dnf install python2-m2crypto
1563         # f33: no longer available but the f31 repos below do the job just fine
1564         # note that using pip2 does not look like a viable option because it does
1565         # an install from sources and that's quite awkward
1566
1567         replacements = [
1568             # no longer on our mirror
1569             "https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/31/Everything/x86_64/os/Packages/p/python2-typing-3.6.2-5.fc31.noarch.rpm",
1570             "https://archives.fedoraproject.org/pub/archive/fedora/linux/releases/31/Everything/x86_64/os/Packages/p/python2-m2crypto-0.35.2-2.fc31.x86_64.rpm",
1571         ]
1572
1573         return (
1574                self.run_in_guest('python2 -c "import M2Crypto"', backslash=True) == 0
1575             or self.run_in_guest("pip2 install python2-m2crypto") == 0
1576             or self.run_in_guest("dnf localinstall -y " + " ".join(replacements)) == 0)
1577
1578         # about pip2: the logic goes like this
1579         # check for pip2 command
1580         # if not, try dnf install python2-pip
1581         # if still not, dnf localinstall the above
1582
1583
1584     def sfa_install_all(self):
1585         "yum install sfa sfa-plc sfa-sfatables sfa-client"
1586
1587         # the rpm/dnf packages named in python2-* are getting deprecated
1588         # we use pip2 instead
1589         # but that's not good for m2crypto
1590
1591         pip_dependencies = [
1592             'sqlalchemy-migrate',
1593             'lxml',
1594             'python-dateutil',
1595             'psycopg2-binary',
1596             'pyOpenSSL',
1597         ]
1598
1599         return (
1600                     self.install_pip2()
1601                 and self.install_m2crypto()
1602                 and all((self.run_in_guest(f"pip2 install {dep}") == 0)
1603                         for dep in pip_dependencies)
1604                 and self.dnf_install("sfa sfa-plc sfa-sfatables sfa-client")
1605                 and self.run_in_guest("systemctl enable sfa-registry")==0
1606                 and self.run_in_guest("systemctl enable sfa-aggregate")==0)
1607
1608     def sfa_install_core(self):
1609         "yum install sfa"
1610         return self.dnf_install("sfa")
1611
1612     def sfa_install_plc(self):
1613         "yum install sfa-plc"
1614         return self.dnf_install("sfa-plc")
1615
1616     def sfa_install_sfatables(self):
1617         "yum install sfa-sfatables"
1618         return self.dnf_install("sfa-sfatables")
1619
1620     # for some very odd reason, this sometimes fails with the following symptom
1621     # # yum install sfa-client
1622     # Setting up Install Process
1623     # ...
1624     # Downloading Packages:
1625     # Running rpm_check_debug
1626     # Running Transaction Test
1627     # Transaction Test Succeeded
1628     # Running Transaction
1629     # Transaction couldn't start:
1630     # installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem
1631     # [('installing package sfa-client-2.1-7.onelab.2012.05.23.i686 needs 68KB on the / filesystem', (9, '/', 69632L))]
1632     # even though in the same context I have
1633     # [2012.05.23--f14-32-sfastd1-1-vplc07] / # df -h
1634     # Filesystem            Size  Used Avail Use% Mounted on
1635     # /dev/hdv1             806G  264G  501G  35% /
1636     # none                   16M   36K   16M   1% /tmp
1637     #
1638     # so as a workaround, we first try yum install, and then invoke rpm on the cached rpm...
1639     def sfa_install_client(self):
1640         "yum install sfa-client"
1641         first_try = self.dnf_install("sfa-client")
1642         if first_try:
1643             return True
1644         utils.header("********** Regular yum failed - special workaround in place, 2nd chance")
1645         code, cached_rpm_path = \
1646                 utils.output_of(self.actual_command_in_guest('find /var/cache/yum -name sfa-client\*.rpm'))
1647         utils.header("rpm_path=<<{}>>".format(rpm_path))
1648         # just for checking
1649         self.run_in_guest("rpm -i {}".format(cached_rpm_path))
1650         return self.dnf_check_installed("sfa-client")
1651
1652     def sfa_dbclean(self):
1653         "thoroughly wipes off the SFA database"
1654         return self.run_in_guest("sfaadmin reg nuke") == 0 or \
1655             self.run_in_guest("sfa-nuke.py") == 0 or \
1656             self.run_in_guest("sfa-nuke-plc.py") == 0 or \
1657             self.run_in_guest("sfaadmin registry nuke") == 0
1658
1659     def sfa_fsclean(self):
1660         "cleanup /etc/sfa/trusted_roots and /var/lib/sfa"
1661         self.run_in_guest("rm -rf /etc/sfa/trusted_roots /var/lib/sfa/authorities")
1662         return True
1663
1664     def sfa_plcclean(self):
1665         "cleans the PLC entries that were created as a side effect of running the script"
1666         # ignore result
1667         sfa_spec = self.plc_spec['sfa']
1668
1669         for auth_sfa_spec in sfa_spec['auth_sfa_specs']:
1670             login_base = auth_sfa_spec['login_base']
1671             try:
1672                 self.apiserver.DeleteSite(self.auth_root(),login_base)
1673             except:
1674                 print("Site {} already absent from PLC db".format(login_base))
1675
1676             for spec_name in ['pi_spec', 'user_spec']:
1677                 user_spec = auth_sfa_spec[spec_name]
1678                 username = user_spec['email']
1679                 try:
1680                     self.apiserver.DeletePerson(self.auth_root(),username)
1681                 except:
1682                     # this in fact is expected as sites delete their members
1683                     #print "User {} already absent from PLC db".format(username)
1684                     pass
1685
1686         print("REMEMBER TO RUN sfa_import AGAIN")
1687         return True
1688
1689     def sfa_uninstall(self):
1690         "uses rpm to uninstall sfa - ignore result"
1691         self.run_in_guest("rpm -e sfa sfa-sfatables sfa-client sfa-plc")
1692         self.run_in_guest("rm -rf /var/lib/sfa")
1693         self.run_in_guest("rm -rf /etc/sfa")
1694         self.run_in_guest("rm -rf /var/log/sfa_access.log /var/log/sfa_import_plc.log /var/log/sfa.daemon")
1695         # xxx tmp
1696         self.run_in_guest("rpm -e --noscripts sfa-plc")
1697         return True
1698
1699     ### run unit tests for SFA
1700     # NOTE: for some reason on f14/i386, yum install sfa-tests fails for no reason
1701     # Running Transaction
1702     # Transaction couldn't start:
1703     # installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem
1704     # [('installing package sfa-tests-1.0-21.onelab.i686 needs 204KB on the / filesystem', (9, '/', 208896L))]
1705     # no matter how many Gbs are available on the testplc
1706     # could not figure out what's wrong, so...
1707     # if the yum install phase fails, consider the test is successful
1708     # other combinations will eventually run it hopefully
1709     def sfa_utest(self):
1710         "dnf install sfa-tests and run SFA unittests"
1711         self.run_in_guest("dnf -y install sfa-tests")
1712         # failed to install - forget it
1713         if self.run_in_guest("rpm -q sfa-tests") != 0:
1714             utils.header("WARNING: SFA unit tests failed to install, ignoring")
1715             return True
1716         return self.run_in_guest("/usr/share/sfa/tests/testAll.py") == 0
1717
1718     ###
1719     def confdir(self):
1720         dirname = "conf.{}".format(self.plc_spec['name'])
1721         if not os.path.isdir(dirname):
1722             utils.system("mkdir -p {}".format(dirname))
1723         if not os.path.isdir(dirname):
1724             raise Exception("Cannot create config dir for plc {}".format(self.name()))
1725         return dirname
1726
1727     def conffile(self, filename):
1728         return "{}/{}".format(self.confdir(), filename)
1729     def confsubdir(self, dirname, clean, dry_run=False):
1730         subdirname = "{}/{}".format(self.confdir(), dirname)
1731         if clean:
1732             utils.system("rm -rf {}".format(subdirname))
1733         if not os.path.isdir(subdirname):
1734             utils.system("mkdir -p {}".format(subdirname))
1735         if not dry_run and not os.path.isdir(subdirname):
1736             raise "Cannot create config subdir {} for plc {}".format(dirname, self.name())
1737         return subdirname
1738
1739     def conffile_clean(self, filename):
1740         filename=self.conffile(filename)
1741         return utils.system("rm -rf {}".format(filename))==0
1742
1743     ###
1744     def sfa_configure(self):
1745         "run sfa-config-tty"
1746         tmpname = self.conffile("sfa-config-tty")
1747         with open(tmpname,'w') as fileconf:
1748             for var, value in self.plc_spec['sfa']['settings'].items():
1749                 fileconf.write('e {}\n{}\n'.format(var, value))
1750             fileconf.write('w\n')
1751             fileconf.write('R\n')
1752             fileconf.write('q\n')
1753         utils.system('cat {}'.format(tmpname))
1754         self.run_in_guest_piped('cat {}'.format(tmpname), 'sfa-config-tty')
1755         return True
1756
1757     def aggregate_xml_line(self):
1758         port = self.plc_spec['sfa']['neighbours-port']
1759         return '<aggregate addr="{}" hrn="{}" port="{}"/>'\
1760             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'], port)
1761
1762     def registry_xml_line(self):
1763         return '<registry addr="{}" hrn="{}" port="12345"/>'\
1764             .format(self.vserverip, self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH'])
1765
1766
1767     # a cross step that takes all other plcs in argument
1768     def cross_sfa_configure(self, other_plcs):
1769         "writes aggregates.xml and registries.xml that point to all other PLCs in the test"
1770         # of course with a single plc, other_plcs is an empty list
1771         if not other_plcs:
1772             return True
1773         agg_fname = self.conffile("agg.xml")
1774         with open(agg_fname,"w") as out:
1775             out.write("<aggregates>{}</aggregates>\n"\
1776                       .format(" ".join([ plc.aggregate_xml_line() for plc in other_plcs ])))
1777         utils.header("(Over)wrote {}".format(agg_fname))
1778         reg_fname=self.conffile("reg.xml")
1779         with open(reg_fname,"w") as out:
1780             out.write("<registries>{}</registries>\n"\
1781                       .format(" ".join([ plc.registry_xml_line() for plc in other_plcs ])))
1782         utils.header("(Over)wrote {}".format(reg_fname))
1783         return self.test_ssh.copy_abs(agg_fname,
1784                                       '/{}/etc/sfa/aggregates.xml'.format(self.vm_root_in_host())) == 0 \
1785            and self.test_ssh.copy_abs(reg_fname,
1786                                       '/{}/etc/sfa/registries.xml'.format(self.vm_root_in_host())) == 0
1787
1788     def sfa_import(self):
1789         "use sfaadmin to import from plc"
1790         auth = self.plc_spec['sfa']['settings']['SFA_REGISTRY_ROOT_AUTH']
1791         return self.run_in_guest('sfaadmin reg import_registry') == 0
1792
1793     def sfa_start(self):
1794         "start SFA through systemctl - also install dependencies"
1795
1796         return (self.start_stop_systemd('sfa-registry', 'start')
1797             and self.start_stop_systemd('sfa-aggregate', 'start'))
1798
1799
1800     def sfi_configure(self):
1801         "Create /root/sfi on the plc side for sfi client configuration"
1802         if self.options.dry_run:
1803             utils.header("DRY RUN - skipping step")
1804             return True
1805         sfa_spec = self.plc_spec['sfa']
1806         # cannot use auth_sfa_mapper to pass dir_name
1807         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1808             test_slice = TestAuthSfa(self, slice_spec)
1809             dir_basename = os.path.basename(test_slice.sfi_path())
1810             dir_name = self.confsubdir("dot-sfi/{}".format(dir_basename),
1811                                        clean=True, dry_run=self.options.dry_run)
1812             test_slice.sfi_configure(dir_name)
1813             # push into the remote /root/sfi area
1814             location = test_slice.sfi_path()
1815             remote = "{}/{}".format(self.vm_root_in_host(), location)
1816             self.test_ssh.mkdir(remote, abs=True)
1817             # need to strip last level or remote otherwise we get an extra dir level
1818             self.test_ssh.copy_abs(dir_name, os.path.dirname(remote), recursive=True)
1819
1820         return True
1821
1822     def sfi_clean(self):
1823         "clean up /root/sfi on the plc side"
1824         self.run_in_guest("rm -rf /root/sfi")
1825         return True
1826
1827     def sfa_rspec_empty(self):
1828         "expose a static empty rspec (ships with the tests module) in the sfi directory"
1829         filename = "empty-rspec.xml"
1830         overall = True
1831         for slice_spec in self.plc_spec['sfa']['auth_sfa_specs']:
1832             test_slice = TestAuthSfa(self, slice_spec)
1833             in_vm = test_slice.sfi_path()
1834             remote = "{}/{}".format(self.vm_root_in_host(), in_vm)
1835             if self.test_ssh.copy_abs(filename, remote) !=0:
1836                 overall = False
1837         return overall
1838
1839     @auth_sfa_mapper
1840     def sfa_register_site(self): pass
1841     @auth_sfa_mapper
1842     def sfa_register_pi(self): pass
1843     @auth_sfa_mapper
1844     def sfa_register_user(self): pass
1845     @auth_sfa_mapper
1846     def sfa_update_user(self): pass
1847     @auth_sfa_mapper
1848     def sfa_register_slice(self): pass
1849     @auth_sfa_mapper
1850     def sfa_renew_slice(self): pass
1851     @auth_sfa_mapper
1852     def sfa_get_expires(self): pass
1853     @auth_sfa_mapper
1854     def sfa_discover(self): pass
1855     @auth_sfa_mapper
1856     def sfa_rspec(self): pass
1857     @auth_sfa_mapper
1858     def sfa_allocate(self): pass
1859     @auth_sfa_mapper
1860     def sfa_allocate_empty(self): pass
1861     @auth_sfa_mapper
1862     def sfa_provision(self): pass
1863     @auth_sfa_mapper
1864     def sfa_provision_empty(self): pass
1865     @auth_sfa_mapper
1866     def sfa_describe(self): pass
1867     @auth_sfa_mapper
1868     def sfa_check_slice_plc(self): pass
1869     @auth_sfa_mapper
1870     def sfa_check_slice_plc_empty(self): pass
1871     @auth_sfa_mapper
1872     def sfa_update_slice(self): pass
1873     @auth_sfa_mapper
1874     def sfa_remove_user_from_slice(self): pass
1875     @auth_sfa_mapper
1876     def sfa_insert_user_in_slice(self): pass
1877     @auth_sfa_mapper
1878     def sfi_list(self): pass
1879     @auth_sfa_mapper
1880     def sfi_show_site(self): pass
1881     @auth_sfa_mapper
1882     def sfi_show_slice(self): pass
1883     @auth_sfa_mapper
1884     def sfi_show_slice_researchers(self): pass
1885     @auth_sfa_mapper
1886     def ssh_slice_sfa(self): pass
1887     @auth_sfa_mapper
1888     def sfa_delete_user(self): pass
1889     @auth_sfa_mapper
1890     def sfa_delete_slice(self): pass
1891
1892     def sfa_stop(self):
1893         "stop sfa through systemclt"
1894         return (self.start_stop_systemd('sfa-aggregate', 'stop') and
1895                 self.start_stop_systemd('sfa-registry', 'stop'))
1896
1897     def populate(self):
1898         "creates random entries in the PLCAPI"
1899         # install the stress-test in the plc image
1900         location = "/usr/share/plc_api/plcsh_stress_test.py"
1901         remote = "{}/{}".format(self.vm_root_in_host(), location)
1902         self.test_ssh.copy_abs("plcsh_stress_test.py", remote)
1903         command = location
1904         command += " -- --preserve --short-names"
1905         local = (self.run_in_guest(command) == 0);
1906         # second run with --foreign
1907         command += ' --foreign'
1908         remote = (self.run_in_guest(command) == 0);
1909         return local and remote
1910
1911
1912     ####################
1913     @bonding_redirector
1914     def bonding_init_partial(self): pass
1915
1916     @bonding_redirector
1917     def bonding_add_yum(self): pass
1918
1919     @bonding_redirector
1920     def bonding_install_rpms(self): pass
1921
1922     ####################
1923
1924     def gather_logs(self):
1925         "gets all possible logs from plc's/qemu node's/slice's for future reference"
1926         # (1.a) get the plc's /var/log/ and store it locally in logs/myplc.var-log.<plcname>/*
1927         # (1.b) get the plc's  /var/lib/pgsql/data/pg_log/ -> logs/myplc.pgsql-log.<plcname>/*
1928         # (1.c) get the plc's /root/sfi -> logs/sfi.<plcname>/
1929         # (2) get all the nodes qemu log and store it as logs/node.qemu.<node>.log
1930         # (3) get the nodes /var/log and store is as logs/node.var-log.<node>/*
1931         # (4) as far as possible get the slice's /var/log as logs/sliver.var-log.<sliver>/*
1932         # (1.a)
1933         print("-------------------- TestPlc.gather_logs : PLC's /var/log")
1934         self.gather_var_logs()
1935         # (1.b)
1936         print("-------------------- TestPlc.gather_logs : PLC's /var/lib/psql/data/pg_log/")
1937         self.gather_pgsql_logs()
1938         # (1.c)
1939         print("-------------------- TestPlc.gather_logs : PLC's /root/sfi/")
1940         self.gather_root_sfi()
1941         # (2)
1942         print("-------------------- TestPlc.gather_logs : nodes's QEMU logs")
1943         for site_spec in self.plc_spec['sites']:
1944             test_site = TestSite(self,site_spec)
1945             for node_spec in site_spec['nodes']:
1946                 test_node = TestNode(self, test_site, node_spec)
1947                 test_node.gather_qemu_logs()
1948         # (3)
1949         print("-------------------- TestPlc.gather_logs : nodes's /var/log")
1950         self.gather_nodes_var_logs()
1951         # (4)
1952         print("-------------------- TestPlc.gather_logs : sample sliver's /var/log")
1953         self.gather_slivers_var_logs()
1954         return True
1955
1956     def gather_slivers_var_logs(self):
1957         for test_sliver in self.all_sliver_objs():
1958             remote = test_sliver.tar_var_logs()
1959             utils.system("mkdir -p logs/sliver.var-log.{}".format(test_sliver.name()))
1960             command = remote + " | tar -C logs/sliver.var-log.{} -xf -".format(test_sliver.name())
1961             utils.system(command)
1962         return True
1963
1964     def gather_var_logs(self):
1965         utils.system("mkdir -p logs/myplc.var-log.{}".format(self.name()))
1966         to_plc = self.actual_command_in_guest("tar -C /var/log/ -cf - .")
1967         command = to_plc + "| tar -C logs/myplc.var-log.{} -xf -".format(self.name())
1968         utils.system(command)
1969         command = "chmod a+r,a+x logs/myplc.var-log.{}/httpd".format(self.name())
1970         utils.system(command)
1971
1972     def gather_pgsql_logs(self):
1973         utils.system("mkdir -p logs/myplc.pgsql-log.{}".format(self.name()))
1974         to_plc = self.actual_command_in_guest("tar -C /var/lib/pgsql/data/pg_log/ -cf - .")
1975         command = to_plc + "| tar -C logs/myplc.pgsql-log.{} -xf -".format(self.name())
1976         utils.system(command)
1977
1978     def gather_root_sfi(self):
1979         utils.system("mkdir -p logs/sfi.{}".format(self.name()))
1980         to_plc = self.actual_command_in_guest("tar -C /root/sfi/ -cf - .")
1981         command = to_plc + "| tar -C logs/sfi.{} -xf -".format(self.name())
1982         utils.system(command)
1983
1984     def gather_nodes_var_logs(self):
1985         for site_spec in self.plc_spec['sites']:
1986             test_site = TestSite(self, site_spec)
1987             for node_spec in site_spec['nodes']:
1988                 test_node = TestNode(self, test_site, node_spec)
1989                 test_ssh = TestSsh(test_node.name(), key="keys/key_admin.rsa")
1990                 command = test_ssh.actual_command("tar -C /var/log -cf - .")
1991                 command = command + "| tar -C logs/node.var-log.{} -xf -".format(test_node.name())
1992                 utils.system("mkdir -p logs/node.var-log.{}".format(test_node.name()))
1993                 utils.system(command)
1994
1995
1996     # returns the filename to use for sql dump/restore, using options.dbname if set
1997     def dbfile(self, database):
1998         # uses options.dbname if it is found
1999         try:
2000             name = self.options.dbname
2001             if not isinstance(name, str):
2002                 raise Exception
2003         except:
2004             t = datetime.now()
2005             d = t.date()
2006             name = str(d)
2007         return "/root/{}-{}.sql".format(database, name)
2008
2009     def plc_db_dump(self):
2010         'dump the planetlab5 DB in /root in the PLC - filename has time'
2011         dump=self.dbfile("planetab5")
2012         self.run_in_guest('pg_dump -U pgsqluser planetlab5 -f '+ dump)
2013         utils.header('Dumped planetlab5 database in {}'.format(dump))
2014         return True
2015
2016     def plc_db_restore(self):
2017         'restore the planetlab5 DB - looks broken, but run -n might help'
2018         dump = self.dbfile("planetab5")
2019         self.run_in_guest('systemctl stop httpd')
2020         # xxx - need another wrapper
2021         self.run_in_guest_piped('echo drop database planetlab5', 'psql --user=pgsqluser template1')
2022         self.run_in_guest('createdb -U postgres --encoding=UNICODE --owner=pgsqluser planetlab5')
2023         self.run_in_guest('psql -U pgsqluser planetlab5 -f ' + dump)
2024         ##starting httpd service
2025         self.run_in_guest('systemctl start httpd')
2026
2027         utils.header('Database restored from ' + dump)
2028
2029     @staticmethod
2030     def create_ignore_steps():
2031         for step in TestPlc.default_steps + TestPlc.other_steps:
2032             # default step can have a plc qualifier
2033             if '@' in step:
2034                 step, qualifier = step.split('@')
2035             # or be defined as forced or ignored by default
2036             for keyword in ['_ignore', '_force']:
2037                 if step.endswith(keyword):
2038                     step=step.replace(keyword,'')
2039             if step == SEP or step == SEPSFA :
2040                 continue
2041             method = getattr(TestPlc,step)
2042             name = step + '_ignore'
2043             wrapped = ignore_result(method)
2044 #            wrapped.__doc__ = method.__doc__ + " (run in ignore-result mode)"
2045             setattr(TestPlc, name, wrapped)
2046
2047 #    @ignore_result
2048 #    def ssh_slice_again_ignore (self): pass
2049 #    @ignore_result
2050 #    def check_initscripts_ignore (self): pass
2051
2052     def standby_1_through_20(self):
2053         """convenience function to wait for a specified number of minutes"""
2054         pass
2055     @standby_generic
2056     def standby_1(): pass
2057     @standby_generic
2058     def standby_2(): pass
2059     @standby_generic
2060     def standby_3(): pass
2061     @standby_generic
2062     def standby_4(): pass
2063     @standby_generic
2064     def standby_5(): pass
2065     @standby_generic
2066     def standby_6(): pass
2067     @standby_generic
2068     def standby_7(): pass
2069     @standby_generic
2070     def standby_8(): pass
2071     @standby_generic
2072     def standby_9(): pass
2073     @standby_generic
2074     def standby_10(): pass
2075     @standby_generic
2076     def standby_11(): pass
2077     @standby_generic
2078     def standby_12(): pass
2079     @standby_generic
2080     def standby_13(): pass
2081     @standby_generic
2082     def standby_14(): pass
2083     @standby_generic
2084     def standby_15(): pass
2085     @standby_generic
2086     def standby_16(): pass
2087     @standby_generic
2088     def standby_17(): pass
2089     @standby_generic
2090     def standby_18(): pass
2091     @standby_generic
2092     def standby_19(): pass
2093     @standby_generic
2094     def standby_20(): pass
2095
2096     # convenience for debugging the test logic
2097     def yes(self): return True
2098     def no(self): return False
2099     def fail(self): return False