code.grnet.gr Git - ganeti-local/blob - qa/ganeti-qa.py

   1 #!/usr/bin/python -u
   2 #
   3
   4 # Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc.
   5 #
   6 # This program is free software; you can redistribute it and/or modify
   7 # it under the terms of the GNU General Public License as published by
   8 # the Free Software Foundation; either version 2 of the License, or
   9 # (at your option) any later version.
  10 #
  11 # This program is distributed in the hope that it will be useful, but
  12 # WITHOUT ANY WARRANTY; without even the implied warranty of
  13 # MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14 # General Public License for more details.
  15 #
  16 # You should have received a copy of the GNU General Public License
  17 # along with this program; if not, write to the Free Software
  18 # Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  19 # 02110-1301, USA.
  20
  21
  22 """Script for doing QA on Ganeti.
  23
  24 """
  25
  26 # pylint: disable=C0103
  27 # due to invalid name
  28
  29 import copy
  30 import datetime
  31 import optparse
  32 import sys
  33
  34 import qa_cluster
  35 import qa_config
  36 import qa_daemon
  37 import qa_env
  38 import qa_error
  39 import qa_group
  40 import qa_instance
  41 import qa_monitoring
  42 import qa_network
  43 import qa_node
  44 import qa_os
  45 import qa_job
  46 import qa_rapi
  47 import qa_tags
  48 import qa_utils
  49
  50 from ganeti import utils
  51 from ganeti import rapi # pylint: disable=W0611
  52 from ganeti import constants
  53 from ganeti import pathutils
  54
  55 from ganeti.http.auth import ParsePasswordFile
  56 import ganeti.rapi.client # pylint: disable=W0611
  57 from ganeti.rapi.client import UsesRapiClient
  58
  59
  60 def _FormatHeader(line, end=72):
  61   """Fill a line up to the end column.
  62
  63   """
  64   line = "---- " + line + " "
  65   line += "-" * (end - len(line))
  66   line = line.rstrip()
  67   return line
  68
  69
  70 def _DescriptionOf(fn):
  71   """Computes the description of an item.
  72
  73   """
  74   if fn.__doc__:
  75     desc = fn.__doc__.splitlines()[0].strip()
  76   else:
  77     desc = "%r" % fn
  78
  79   return desc.rstrip(".")
  80
  81
  82 def RunTest(fn, *args, **kwargs):
  83   """Runs a test after printing a header.
  84
  85   """
  86
  87   tstart = datetime.datetime.now()
  88
  89   desc = _DescriptionOf(fn)
  90
  91   print
  92   print _FormatHeader("%s start %s" % (tstart, desc))
  93
  94   try:
  95     retval = fn(*args, **kwargs)
  96     return retval
  97   finally:
  98     tstop = datetime.datetime.now()
  99     tdelta = tstop - tstart
 100     print _FormatHeader("%s time=%s %s" % (tstop, tdelta, desc))
 101
 102
 103 def RunTestIf(testnames, fn, *args, **kwargs):
 104   """Runs a test conditionally.
 105
 106   @param testnames: either a single test name in the configuration
 107       file, or a list of testnames (which will be AND-ed together)
 108
 109   """
 110   if qa_config.TestEnabled(testnames):
 111     RunTest(fn, *args, **kwargs)
 112   else:
 113     tstart = datetime.datetime.now()
 114     desc = _DescriptionOf(fn)
 115     # TODO: Formatting test names when non-string names are involved
 116     print _FormatHeader("%s skipping %s, test(s) %s disabled" %
 117                         (tstart, desc, testnames))
 118
 119
 120 def RunEnvTests():
 121   """Run several environment tests.
 122
 123   """
 124   RunTestIf("env", qa_env.TestSshConnection)
 125   RunTestIf("env", qa_env.TestIcmpPing)
 126   RunTestIf("env", qa_env.TestGanetiCommands)
 127
 128
 129 def _LookupRapiSecret(rapi_user):
 130   """Find the RAPI secret for the given user.
 131
 132   @param rapi_user: Login user
 133   @return: Login secret for the user
 134
 135   """
 136   CTEXT = "{CLEARTEXT}"
 137   master = qa_config.GetMasterNode()
 138   cmd = ["cat", qa_utils.MakeNodePath(master, pathutils.RAPI_USERS_FILE)]
 139   file_content = qa_utils.GetCommandOutput(master.primary,
 140                                            utils.ShellQuoteArgs(cmd))
 141   users = ParsePasswordFile(file_content)
 142   entry = users.get(rapi_user)
 143   if not entry:
 144     raise qa_error.Error("User %s not found in RAPI users file" % rapi_user)
 145   secret = entry.password
 146   if secret.upper().startswith(CTEXT):
 147     secret = secret[len(CTEXT):]
 148   elif secret.startswith("{"):
 149     raise qa_error.Error("Unsupported password schema for RAPI user %s:"
 150                          " not a clear text password" % rapi_user)
 151   return secret
 152
 153
 154 def SetupCluster(rapi_user):
 155   """Initializes the cluster.
 156
 157   @param rapi_user: Login user for RAPI
 158   @return: Login secret for RAPI
 159
 160   """
 161   rapi_secret = utils.GenerateSecret()
 162   RunTestIf("create-cluster", qa_cluster.TestClusterInit,
 163             rapi_user, rapi_secret)
 164   if not qa_config.TestEnabled("create-cluster"):
 165     # If the cluster is already in place, we assume that exclusive-storage is
 166     # already set according to the configuration
 167     qa_config.SetExclusiveStorage(qa_config.get("exclusive-storage", False))
 168     if qa_rapi.Enabled():
 169       # To support RAPI on an existing cluster we have to find out the secret
 170       rapi_secret = _LookupRapiSecret(rapi_user)
 171
 172   # Test on empty cluster
 173   RunTestIf("node-list", qa_node.TestNodeList)
 174   RunTestIf("instance-list", qa_instance.TestInstanceList)
 175   RunTestIf("job-list", qa_job.TestJobList)
 176
 177   RunTestIf("create-cluster", qa_node.TestNodeAddAll)
 178   if not qa_config.TestEnabled("create-cluster"):
 179     # consider the nodes are already there
 180     qa_node.MarkNodeAddedAll()
 181
 182   RunTestIf("test-jobqueue", qa_cluster.TestJobqueue)
 183
 184   # enable the watcher (unconditionally)
 185   RunTest(qa_daemon.TestResumeWatcher)
 186
 187   RunTestIf("node-list", qa_node.TestNodeList)
 188
 189   # Test listing fields
 190   RunTestIf("node-list", qa_node.TestNodeListFields)
 191   RunTestIf("instance-list", qa_instance.TestInstanceListFields)
 192   RunTestIf("job-list", qa_job.TestJobListFields)
 193   RunTestIf("instance-export", qa_instance.TestBackupListFields)
 194
 195   RunTestIf("node-info", qa_node.TestNodeInfo)
 196
 197   return rapi_secret
 198
 199
 200 def RunClusterTests():
 201   """Runs tests related to gnt-cluster.
 202
 203   """
 204   for test, fn in [
 205     ("create-cluster", qa_cluster.TestClusterInitDisk),
 206     ("cluster-renew-crypto", qa_cluster.TestClusterRenewCrypto),
 207     ("cluster-verify", qa_cluster.TestClusterVerify),
 208     ("cluster-reserved-lvs", qa_cluster.TestClusterReservedLvs),
 209     # TODO: add more cluster modify tests
 210     ("cluster-modify", qa_cluster.TestClusterModifyEmpty),
 211     ("cluster-modify", qa_cluster.TestClusterModifyIPolicy),
 212     ("cluster-modify", qa_cluster.TestClusterModifyISpecs),
 213     ("cluster-modify", qa_cluster.TestClusterModifyBe),
 214     ("cluster-modify", qa_cluster.TestClusterModifyDisk),
 215     ("cluster-modify", qa_cluster.TestClusterModifyDiskTemplates),
 216     ("cluster-rename", qa_cluster.TestClusterRename),
 217     ("cluster-info", qa_cluster.TestClusterVersion),
 218     ("cluster-info", qa_cluster.TestClusterInfo),
 219     ("cluster-info", qa_cluster.TestClusterGetmaster),
 220     ("cluster-redist-conf", qa_cluster.TestClusterRedistConf),
 221     (["cluster-copyfile", qa_config.NoVirtualCluster],
 222      qa_cluster.TestClusterCopyfile),
 223     ("cluster-command", qa_cluster.TestClusterCommand),
 224     ("cluster-burnin", qa_cluster.TestClusterBurnin),
 225     ("cluster-master-failover", qa_cluster.TestClusterMasterFailover),
 226     ("cluster-master-failover",
 227      qa_cluster.TestClusterMasterFailoverWithDrainedQueue),
 228     (["cluster-oob", qa_config.NoVirtualCluster],
 229      qa_cluster.TestClusterOob),
 230     (qa_rapi.Enabled, qa_rapi.TestVersion),
 231     (qa_rapi.Enabled, qa_rapi.TestEmptyCluster),
 232     (qa_rapi.Enabled, qa_rapi.TestRapiQuery),
 233     ]:
 234     RunTestIf(test, fn)
 235
 236
 237 def RunRepairDiskSizes():
 238   """Run the repair disk-sizes test.
 239
 240   """
 241   RunTestIf("cluster-repair-disk-sizes", qa_cluster.TestClusterRepairDiskSizes)
 242
 243
 244 def RunOsTests():
 245   """Runs all tests related to gnt-os.
 246
 247   """
 248   os_enabled = ["os", qa_config.NoVirtualCluster]
 249
 250   if qa_config.TestEnabled(qa_rapi.Enabled):
 251     rapi_getos = qa_rapi.GetOperatingSystems
 252   else:
 253     rapi_getos = None
 254
 255   for fn in [
 256     qa_os.TestOsList,
 257     qa_os.TestOsDiagnose,
 258     ]:
 259     RunTestIf(os_enabled, fn)
 260
 261   for fn in [
 262     qa_os.TestOsValid,
 263     qa_os.TestOsInvalid,
 264     qa_os.TestOsPartiallyValid,
 265     ]:
 266     RunTestIf(os_enabled, fn, rapi_getos)
 267
 268   for fn in [
 269     qa_os.TestOsModifyValid,
 270     qa_os.TestOsModifyInvalid,
 271     qa_os.TestOsStatesNonExisting,
 272     ]:
 273     RunTestIf(os_enabled, fn)
 274
 275
 276 def RunCommonInstanceTests(instance, inst_nodes):
 277   """Runs a few tests that are common to all disk types.
 278
 279   """
 280   RunTestIf("instance-shutdown", qa_instance.TestInstanceShutdown, instance)
 281   RunTestIf(["instance-shutdown", "instance-console", qa_rapi.Enabled],
 282             qa_rapi.TestRapiStoppedInstanceConsole, instance)
 283   RunTestIf(["instance-shutdown", "instance-modify"],
 284             qa_instance.TestInstanceStoppedModify, instance)
 285   RunTestIf("instance-shutdown", qa_instance.TestInstanceStartup, instance)
 286
 287   # Test shutdown/start via RAPI
 288   RunTestIf(["instance-shutdown", qa_rapi.Enabled],
 289             qa_rapi.TestRapiInstanceShutdown, instance)
 290   RunTestIf(["instance-shutdown", qa_rapi.Enabled],
 291             qa_rapi.TestRapiInstanceStartup, instance)
 292
 293   RunTestIf("instance-list", qa_instance.TestInstanceList)
 294
 295   RunTestIf("instance-info", qa_instance.TestInstanceInfo, instance)
 296
 297   RunTestIf("instance-modify", qa_instance.TestInstanceModify, instance)
 298   RunTestIf(["instance-modify", qa_rapi.Enabled],
 299             qa_rapi.TestRapiInstanceModify, instance)
 300
 301   RunTestIf("instance-console", qa_instance.TestInstanceConsole, instance)
 302   RunTestIf(["instance-console", qa_rapi.Enabled],
 303             qa_rapi.TestRapiInstanceConsole, instance)
 304
 305   RunTestIf("instance-device-names", qa_instance.TestInstanceDeviceNames,
 306             instance)
 307   DOWN_TESTS = qa_config.Either([
 308     "instance-reinstall",
 309     "instance-rename",
 310     "instance-grow-disk",
 311     ])
 312
 313   # shutdown instance for any 'down' tests
 314   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceShutdown, instance)
 315
 316   # now run the 'down' state tests
 317   RunTestIf("instance-reinstall", qa_instance.TestInstanceReinstall, instance)
 318   RunTestIf(["instance-reinstall", qa_rapi.Enabled],
 319             qa_rapi.TestRapiInstanceReinstall, instance)
 320
 321   if qa_config.TestEnabled("instance-rename"):
 322     tgt_instance = qa_config.AcquireInstance()
 323     try:
 324       rename_source = instance.name
 325       rename_target = tgt_instance.name
 326       # perform instance rename to the same name
 327       RunTest(qa_instance.TestInstanceRenameAndBack,
 328               rename_source, rename_source)
 329       RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceRenameAndBack,
 330                 rename_source, rename_source)
 331       if rename_target is not None:
 332         # perform instance rename to a different name, if we have one configured
 333         RunTest(qa_instance.TestInstanceRenameAndBack,
 334                 rename_source, rename_target)
 335         RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceRenameAndBack,
 336                   rename_source, rename_target)
 337     finally:
 338       tgt_instance.Release()
 339
 340   RunTestIf(["instance-grow-disk"], qa_instance.TestInstanceGrowDisk, instance)
 341
 342   # and now start the instance again
 343   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceStartup, instance)
 344
 345   RunTestIf("instance-reboot", qa_instance.TestInstanceReboot, instance)
 346
 347   RunTestIf("tags", qa_tags.TestInstanceTags, instance)
 348
 349   if instance.disk_template == constants.DT_DRBD8:
 350     RunTestIf("cluster-verify",
 351               qa_cluster.TestClusterVerifyDisksBrokenDRBD, instance, inst_nodes)
 352   RunTestIf("cluster-verify", qa_cluster.TestClusterVerify)
 353
 354   RunTestIf(qa_rapi.Enabled, qa_rapi.TestInstance, instance)
 355
 356   # Lists instances, too
 357   RunTestIf("node-list", qa_node.TestNodeList)
 358
 359   # Some jobs have been run, let's test listing them
 360   RunTestIf("job-list", qa_job.TestJobList)
 361
 362
 363 def RunCommonNodeTests():
 364   """Run a few common node tests.
 365
 366   """
 367   RunTestIf("node-volumes", qa_node.TestNodeVolumes)
 368   RunTestIf("node-storage", qa_node.TestNodeStorage)
 369   RunTestIf(["node-oob", qa_config.NoVirtualCluster], qa_node.TestOutOfBand)
 370
 371
 372 def RunGroupListTests():
 373   """Run tests for listing node groups.
 374
 375   """
 376   RunTestIf("group-list", qa_group.TestGroupList)
 377   RunTestIf("group-list", qa_group.TestGroupListFields)
 378
 379
 380 def RunNetworkTests():
 381   """Run tests for network management.
 382
 383   """
 384   RunTestIf("network", qa_network.TestNetworkAddRemove)
 385   RunTestIf("network", qa_network.TestNetworkConnect)
 386
 387
 388 def RunGroupRwTests():
 389   """Run tests for adding/removing/renaming groups.
 390
 391   """
 392   RunTestIf("group-rwops", qa_group.TestGroupAddRemoveRename)
 393   RunTestIf("group-rwops", qa_group.TestGroupAddWithOptions)
 394   RunTestIf("group-rwops", qa_group.TestGroupModify)
 395   RunTestIf(["group-rwops", qa_rapi.Enabled], qa_rapi.TestRapiNodeGroups)
 396   RunTestIf(["group-rwops", "tags"], qa_tags.TestGroupTags,
 397             qa_group.GetDefaultGroup())
 398
 399
 400 def RunExportImportTests(instance, inodes):
 401   """Tries to export and import the instance.
 402
 403   @type inodes: list of nodes
 404   @param inodes: current nodes of the instance
 405
 406   """
 407   # FIXME: export explicitly bails out on file based storage. other non-lvm
 408   # based storage types are untested, though. Also note that import could still
 409   # work, but is deeply embedded into the "export" case.
 410   if (qa_config.TestEnabled("instance-export") and
 411       instance.disk_template != constants.DT_FILE):
 412     RunTest(qa_instance.TestInstanceExportNoTarget, instance)
 413
 414     pnode = inodes[0]
 415     expnode = qa_config.AcquireNode(exclude=pnode)
 416     try:
 417       name = RunTest(qa_instance.TestInstanceExport, instance, expnode)
 418
 419       RunTest(qa_instance.TestBackupList, expnode)
 420
 421       if qa_config.TestEnabled("instance-import"):
 422         newinst = qa_config.AcquireInstance()
 423         try:
 424           RunTest(qa_instance.TestInstanceImport, newinst, pnode,
 425                   expnode, name)
 426           # Check if starting the instance works
 427           RunTest(qa_instance.TestInstanceStartup, newinst)
 428           RunTest(qa_instance.TestInstanceRemove, newinst)
 429         finally:
 430           newinst.Release()
 431     finally:
 432       expnode.Release()
 433
 434   # FIXME: inter-cluster-instance-move crashes on file based instances :/
 435   # See Issue 414.
 436   if (qa_config.TestEnabled([qa_rapi.Enabled, "inter-cluster-instance-move"])
 437       and instance.disk_template != constants.DT_FILE):
 438     newinst = qa_config.AcquireInstance()
 439     try:
 440       tnode = qa_config.AcquireNode(exclude=inodes)
 441       try:
 442         RunTest(qa_rapi.TestInterClusterInstanceMove, instance, newinst,
 443                 inodes, tnode)
 444       finally:
 445         tnode.Release()
 446     finally:
 447       newinst.Release()
 448
 449
 450 def RunDaemonTests(instance):
 451   """Test the ganeti-watcher script.
 452
 453   """
 454   RunTest(qa_daemon.TestPauseWatcher)
 455
 456   RunTestIf("instance-automatic-restart",
 457             qa_daemon.TestInstanceAutomaticRestart, instance)
 458   RunTestIf("instance-consecutive-failures",
 459             qa_daemon.TestInstanceConsecutiveFailures, instance)
 460
 461   RunTest(qa_daemon.TestResumeWatcher)
 462
 463
 464 def RunHardwareFailureTests(instance, inodes):
 465   """Test cluster internal hardware failure recovery.
 466
 467   """
 468   RunTestIf("instance-failover", qa_instance.TestInstanceFailover, instance)
 469   RunTestIf(["instance-failover", qa_rapi.Enabled],
 470             qa_rapi.TestRapiInstanceFailover, instance)
 471
 472   RunTestIf("instance-migrate", qa_instance.TestInstanceMigrate, instance)
 473   RunTestIf(["instance-migrate", qa_rapi.Enabled],
 474             qa_rapi.TestRapiInstanceMigrate, instance)
 475
 476   if qa_config.TestEnabled("instance-replace-disks"):
 477     # We just need alternative secondary nodes, hence "- 1"
 478     othernodes = qa_config.AcquireManyNodes(len(inodes) - 1, exclude=inodes)
 479     try:
 480       RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceReplaceDisks, instance)
 481       RunTest(qa_instance.TestReplaceDisks,
 482               instance, inodes, othernodes)
 483     finally:
 484       qa_config.ReleaseManyNodes(othernodes)
 485     del othernodes
 486
 487   if qa_config.TestEnabled("instance-recreate-disks"):
 488     try:
 489       acquirednodes = qa_config.AcquireManyNodes(len(inodes), exclude=inodes)
 490       othernodes = acquirednodes
 491     except qa_error.OutOfNodesError:
 492       if len(inodes) > 1:
 493         # If the cluster is not big enough, let's reuse some of the nodes, but
 494         # with different roles. In this way, we can test a DRBD instance even on
 495         # a 3-node cluster.
 496         acquirednodes = [qa_config.AcquireNode(exclude=inodes)]
 497         othernodes = acquirednodes + inodes[:-1]
 498       else:
 499         raise
 500     try:
 501       RunTest(qa_instance.TestRecreateDisks,
 502               instance, inodes, othernodes)
 503     finally:
 504       qa_config.ReleaseManyNodes(acquirednodes)
 505
 506   if len(inodes) >= 2:
 507     RunTestIf("node-evacuate", qa_node.TestNodeEvacuate, inodes[0], inodes[1])
 508     RunTestIf("node-failover", qa_node.TestNodeFailover, inodes[0], inodes[1])
 509     RunTestIf("node-migrate", qa_node.TestNodeMigrate, inodes[0], inodes[1])
 510
 511
 512 def RunExclusiveStorageTests():
 513   """Test exclusive storage."""
 514   if not qa_config.TestEnabled("cluster-exclusive-storage"):
 515     return
 516
 517   node = qa_config.AcquireNode()
 518   try:
 519     old_es = qa_cluster.TestSetExclStorCluster(False)
 520     qa_node.TestExclStorSingleNode(node)
 521
 522     qa_cluster.TestSetExclStorCluster(True)
 523     qa_cluster.TestExclStorSharedPv(node)
 524
 525     if qa_config.TestEnabled("instance-add-plain-disk"):
 526       # Make sure that the cluster doesn't have any pre-existing problem
 527       qa_cluster.AssertClusterVerify()
 528
 529       # Create and allocate instances
 530       instance1 = qa_instance.TestInstanceAddWithPlainDisk([node])
 531       try:
 532         instance2 = qa_instance.TestInstanceAddWithPlainDisk([node])
 533         try:
 534           # cluster-verify checks that disks are allocated correctly
 535           qa_cluster.AssertClusterVerify()
 536
 537           # Remove instances
 538           qa_instance.TestInstanceRemove(instance2)
 539           qa_instance.TestInstanceRemove(instance1)
 540         finally:
 541           instance2.Release()
 542       finally:
 543         instance1.Release()
 544
 545     if qa_config.TestEnabled("instance-add-drbd-disk"):
 546       snode = qa_config.AcquireNode()
 547       try:
 548         qa_cluster.TestSetExclStorCluster(False)
 549         instance = qa_instance.TestInstanceAddWithDrbdDisk([node, snode])
 550         try:
 551           qa_cluster.TestSetExclStorCluster(True)
 552           exp_err = [constants.CV_EINSTANCEUNSUITABLENODE]
 553           qa_cluster.AssertClusterVerify(fail=True, errors=exp_err)
 554           qa_instance.TestInstanceRemove(instance)
 555         finally:
 556           instance.Release()
 557       finally:
 558         snode.Release()
 559     qa_cluster.TestSetExclStorCluster(old_es)
 560   finally:
 561     node.Release()
 562
 563
 564 def _BuildSpecDict(par, mn, st, mx):
 565   return {
 566     constants.ISPECS_MINMAX: [{
 567       constants.ISPECS_MIN: {par: mn},
 568       constants.ISPECS_MAX: {par: mx},
 569       }],
 570     constants.ISPECS_STD: {par: st},
 571     }
 572
 573
 574 def _BuildDoubleSpecDict(index, par, mn, st, mx):
 575   new_spec = {
 576     constants.ISPECS_MINMAX: [{}, {}],
 577     }
 578   if st is not None:
 579     new_spec[constants.ISPECS_STD] = {par: st}
 580   new_spec[constants.ISPECS_MINMAX][index] = {
 581     constants.ISPECS_MIN: {par: mn},
 582     constants.ISPECS_MAX: {par: mx},
 583     }
 584   return new_spec
 585
 586
 587 def TestIPolicyPlainInstance():
 588   """Test instance policy interaction with instances"""
 589   params = ["memory-size", "cpu-count", "disk-count", "disk-size", "nic-count"]
 590   if not qa_config.IsTemplateSupported(constants.DT_PLAIN):
 591     print "Template %s not supported" % constants.DT_PLAIN
 592     return
 593
 594   # This test assumes that the group policy is empty
 595   (_, old_specs) = qa_cluster.TestClusterSetISpecs()
 596   # We also assume to have only one min/max bound
 597   assert len(old_specs[constants.ISPECS_MINMAX]) == 1
 598   node = qa_config.AcquireNode()
 599   try:
 600     # Log of policy changes, list of tuples:
 601     # (full_change, incremental_change, policy_violated)
 602     history = []
 603     instance = qa_instance.TestInstanceAddWithPlainDisk([node])
 604     try:
 605       policyerror = [constants.CV_EINSTANCEPOLICY]
 606       for par in params:
 607         (iminval, imaxval) = qa_instance.GetInstanceSpec(instance.name, par)
 608         # Some specs must be multiple of 4
 609         new_spec = _BuildSpecDict(par, imaxval + 4, imaxval + 4, imaxval + 4)
 610         history.append((None, new_spec, True))
 611         if iminval > 0:
 612           # Some specs must be multiple of 4
 613           if iminval >= 4:
 614             upper = iminval - 4
 615           else:
 616             upper = iminval - 1
 617           new_spec = _BuildSpecDict(par, 0, upper, upper)
 618           history.append((None, new_spec, True))
 619         history.append((old_specs, None, False))
 620
 621       # Test with two instance specs
 622       double_specs = copy.deepcopy(old_specs)
 623       double_specs[constants.ISPECS_MINMAX] = \
 624           double_specs[constants.ISPECS_MINMAX] * 2
 625       (par1, par2) = params[0:2]
 626       (_, imaxval1) = qa_instance.GetInstanceSpec(instance.name, par1)
 627       (_, imaxval2) = qa_instance.GetInstanceSpec(instance.name, par2)
 628       old_minmax = old_specs[constants.ISPECS_MINMAX][0]
 629       history.extend([
 630         (double_specs, None, False),
 631         # The first min/max limit is being violated
 632         (None,
 633          _BuildDoubleSpecDict(0, par1, imaxval1 + 4, imaxval1 + 4,
 634                               imaxval1 + 4),
 635          False),
 636         # Both min/max limits are being violated
 637         (None,
 638          _BuildDoubleSpecDict(1, par2, imaxval2 + 4, None, imaxval2 + 4),
 639          True),
 640         # The second min/max limit is being violated
 641         (None,
 642          _BuildDoubleSpecDict(0, par1,
 643                               old_minmax[constants.ISPECS_MIN][par1],
 644                               old_specs[constants.ISPECS_STD][par1],
 645                               old_minmax[constants.ISPECS_MAX][par1]),
 646          False),
 647         (old_specs, None, False),
 648         ])
 649
 650       # Apply the changes, and check policy violations after each change
 651       qa_cluster.AssertClusterVerify()
 652       for (new_specs, diff_specs, failed) in history:
 653         qa_cluster.TestClusterSetISpecs(new_specs=new_specs,
 654                                         diff_specs=diff_specs)
 655         if failed:
 656           qa_cluster.AssertClusterVerify(warnings=policyerror)
 657         else:
 658           qa_cluster.AssertClusterVerify()
 659
 660       qa_instance.TestInstanceRemove(instance)
 661     finally:
 662       instance.Release()
 663
 664     # Now we replay the same policy changes, and we expect that the instance
 665     # cannot be created for the cases where we had a policy violation above
 666     for (new_specs, diff_specs, failed) in history:
 667       qa_cluster.TestClusterSetISpecs(new_specs=new_specs,
 668                                       diff_specs=diff_specs)
 669       if failed:
 670         qa_instance.TestInstanceAddWithPlainDisk([node], fail=True)
 671       # Instance creation with no policy violation has been tested already
 672   finally:
 673     node.Release()
 674
 675
 676 def IsExclusiveStorageInstanceTestEnabled():
 677   test_name = "exclusive-storage-instance-tests"
 678   if qa_config.TestEnabled(test_name):
 679     vgname = qa_config.get("vg-name", constants.DEFAULT_VG)
 680     vgscmd = utils.ShellQuoteArgs([
 681       "vgs", "--noheadings", "-o", "pv_count", vgname,
 682       ])
 683     nodes = qa_config.GetConfig()["nodes"]
 684     for node in nodes:
 685       try:
 686         pvnum = int(qa_utils.GetCommandOutput(node.primary, vgscmd))
 687       except Exception, e:
 688         msg = ("Cannot get the number of PVs on %s, needed by '%s': %s" %
 689                (node.primary, test_name, e))
 690         raise qa_error.Error(msg)
 691       if pvnum < 2:
 692         raise qa_error.Error("Node %s has not enough PVs (%s) to run '%s'" %
 693                              (node.primary, pvnum, test_name))
 694     res = True
 695   else:
 696     res = False
 697   return res
 698
 699
 700 def RunInstanceTests():
 701   """Create and exercise instances."""
 702   instance_tests = [
 703     ("instance-add-plain-disk", constants.DT_PLAIN,
 704      qa_instance.TestInstanceAddWithPlainDisk, 1),
 705     ("instance-add-drbd-disk", constants.DT_DRBD8,
 706      qa_instance.TestInstanceAddWithDrbdDisk, 2),
 707     ("instance-add-diskless", constants.DT_DISKLESS,
 708      qa_instance.TestInstanceAddDiskless, 1),
 709     ("instance-add-file", constants.DT_FILE,
 710      qa_instance.TestInstanceAddFile, 1)
 711     ]
 712
 713   for (test_name, templ, create_fun, num_nodes) in instance_tests:
 714     if (qa_config.TestEnabled(test_name) and
 715         qa_config.IsTemplateSupported(templ)):
 716       inodes = qa_config.AcquireManyNodes(num_nodes)
 717       try:
 718         instance = RunTest(create_fun, inodes)
 719         try:
 720           RunTestIf("cluster-epo", qa_cluster.TestClusterEpo)
 721           RunDaemonTests(instance)
 722           for node in inodes:
 723             RunTestIf("haskell-confd", qa_node.TestNodeListDrbd, node)
 724           if len(inodes) > 1:
 725             RunTestIf("group-rwops", qa_group.TestAssignNodesIncludingSplit,
 726                       constants.INITIAL_NODE_GROUP_NAME,
 727                       inodes[0].primary, inodes[1].primary)
 728           if qa_config.TestEnabled("instance-convert-disk"):
 729             RunTest(qa_instance.TestInstanceShutdown, instance)
 730             RunTest(qa_instance.TestInstanceConvertDiskToPlain,
 731                     instance, inodes)
 732             RunTest(qa_instance.TestInstanceStartup, instance)
 733           RunTestIf("instance-modify-disks",
 734                     qa_instance.TestInstanceModifyDisks, instance)
 735           RunCommonInstanceTests(instance, inodes)
 736           if qa_config.TestEnabled("instance-modify-primary"):
 737             othernode = qa_config.AcquireNode()
 738             RunTest(qa_instance.TestInstanceModifyPrimaryAndBack,
 739                     instance, inodes[0], othernode)
 740             othernode.Release()
 741           RunGroupListTests()
 742           RunExportImportTests(instance, inodes)
 743           RunHardwareFailureTests(instance, inodes)
 744           RunRepairDiskSizes()
 745           RunTest(qa_instance.TestInstanceRemove, instance)
 746         finally:
 747           instance.Release()
 748         del instance
 749       finally:
 750         qa_config.ReleaseManyNodes(inodes)
 751       qa_cluster.AssertClusterVerify()
 752
 753
 754 def RunMonitoringTests():
 755   if qa_config.TestEnabled("mon-collector"):
 756     RunTest(qa_monitoring.TestInstStatusCollector)
 757
 758
 759 def RunQa():
 760   """Main QA body.
 761
 762   """
 763   rapi_user = "ganeti-qa"
 764
 765   RunEnvTests()
 766   rapi_secret = SetupCluster(rapi_user)
 767
 768   if qa_rapi.Enabled():
 769     # Load RAPI certificate
 770     qa_rapi.Setup(rapi_user, rapi_secret)
 771
 772   RunClusterTests()
 773   RunOsTests()
 774
 775   RunTestIf("tags", qa_tags.TestClusterTags)
 776
 777   RunCommonNodeTests()
 778   RunGroupListTests()
 779   RunGroupRwTests()
 780   RunNetworkTests()
 781
 782   # The master shouldn't be readded or put offline; "delay" needs a non-master
 783   # node to test
 784   pnode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
 785   try:
 786     RunTestIf("node-readd", qa_node.TestNodeReadd, pnode)
 787     RunTestIf("node-modify", qa_node.TestNodeModify, pnode)
 788     RunTestIf("delay", qa_cluster.TestDelay, pnode)
 789   finally:
 790     pnode.Release()
 791
 792   # Make sure the cluster is clean before running instance tests
 793   qa_cluster.AssertClusterVerify()
 794
 795   pnode = qa_config.AcquireNode()
 796   try:
 797     RunTestIf("tags", qa_tags.TestNodeTags, pnode)
 798
 799     if qa_rapi.Enabled():
 800       RunTest(qa_rapi.TestNode, pnode)
 801
 802       if qa_config.TestEnabled("instance-add-plain-disk"):
 803         for use_client in [True, False]:
 804           rapi_instance = RunTest(qa_rapi.TestRapiInstanceAdd, pnode,
 805                                   use_client)
 806           try:
 807             if qa_config.TestEnabled("instance-plain-rapi-common-tests"):
 808               RunCommonInstanceTests(rapi_instance, [pnode])
 809             RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance, use_client)
 810           finally:
 811             rapi_instance.Release()
 812           del rapi_instance
 813
 814   finally:
 815     pnode.Release()
 816
 817   config_list = [
 818     ("default-instance-tests", lambda: None, lambda _: None),
 819     (IsExclusiveStorageInstanceTestEnabled,
 820      lambda: qa_cluster.TestSetExclStorCluster(True),
 821      qa_cluster.TestSetExclStorCluster),
 822   ]
 823   for (conf_name, setup_conf_f, restore_conf_f) in config_list:
 824     if qa_config.TestEnabled(conf_name):
 825       oldconf = setup_conf_f()
 826       RunInstanceTests()
 827       restore_conf_f(oldconf)
 828
 829   pnode = qa_config.AcquireNode()
 830   try:
 831     if qa_config.TestEnabled(["instance-add-plain-disk", "instance-export"]):
 832       for shutdown in [False, True]:
 833         instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, [pnode])
 834         try:
 835           expnode = qa_config.AcquireNode(exclude=pnode)
 836           try:
 837             if shutdown:
 838               # Stop instance before exporting and removing it
 839               RunTest(qa_instance.TestInstanceShutdown, instance)
 840             RunTest(qa_instance.TestInstanceExportWithRemove, instance, expnode)
 841             RunTest(qa_instance.TestBackupList, expnode)
 842           finally:
 843             expnode.Release()
 844         finally:
 845           instance.Release()
 846         del expnode
 847         del instance
 848       qa_cluster.AssertClusterVerify()
 849
 850   finally:
 851     pnode.Release()
 852
 853   RunExclusiveStorageTests()
 854   RunTestIf(["cluster-instance-policy", "instance-add-plain-disk"],
 855             TestIPolicyPlainInstance)
 856
 857   RunTestIf(
 858     "instance-add-restricted-by-disktemplates",
 859     qa_instance.TestInstanceCreationRestrictedByDiskTemplates)
 860
 861   # Test removing instance with offline drbd secondary
 862   if qa_config.TestEnabled(["instance-remove-drbd-offline",
 863                             "instance-add-drbd-disk"]):
 864     # Make sure the master is not put offline
 865     snode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
 866     try:
 867       pnode = qa_config.AcquireNode(exclude=snode)
 868       try:
 869         instance = qa_instance.TestInstanceAddWithDrbdDisk([pnode, snode])
 870         set_offline = lambda node: qa_node.MakeNodeOffline(node, "yes")
 871         set_online = lambda node: qa_node.MakeNodeOffline(node, "no")
 872         RunTest(qa_instance.TestRemoveInstanceOfflineNode, instance, snode,
 873                 set_offline, set_online)
 874       finally:
 875         pnode.Release()
 876     finally:
 877       snode.Release()
 878     qa_cluster.AssertClusterVerify()
 879
 880   RunMonitoringTests()
 881
 882   RunTestIf("create-cluster", qa_node.TestNodeRemoveAll)
 883
 884   RunTestIf("cluster-destroy", qa_cluster.TestClusterDestroy)
 885
 886
 887 @UsesRapiClient
 888 def main():
 889   """Main program.
 890
 891   """
 892   parser = optparse.OptionParser(usage="%prog [options] <config-file>")
 893   parser.add_option("--yes-do-it", dest="yes_do_it",
 894                     action="store_true",
 895                     help="Really execute the tests")
 896   (opts, args) = parser.parse_args()
 897
 898   if len(args) == 1:
 899     (config_file, ) = args
 900   else:
 901     parser.error("Wrong number of arguments.")
 902
 903   if not opts.yes_do_it:
 904     print ("Executing this script irreversibly destroys any Ganeti\n"
 905            "configuration on all nodes involved. If you really want\n"
 906            "to start testing, supply the --yes-do-it option.")
 907     sys.exit(1)
 908
 909   qa_config.Load(config_file)
 910
 911   primary = qa_config.GetMasterNode().primary
 912   qa_utils.StartMultiplexer(primary)
 913   print ("SSH command for primary node: %s" %
 914          utils.ShellQuoteArgs(qa_utils.GetSSHCommand(primary, "")))
 915   print ("SSH command for other nodes: %s" %
 916          utils.ShellQuoteArgs(qa_utils.GetSSHCommand("NODE", "")))
 917   try:
 918     RunQa()
 919   finally:
 920     qa_utils.CloseMultiplexers()
 921
 922 if __name__ == "__main__":
 923   main()