system/TestPlc.py

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