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