qa/ganeti-qa.py

   1 #!/usr/bin/python -u
   2 #
   3
   4 # Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc.
   5 # All rights reserved.
   6 #
   7 # Redistribution and use in source and binary forms, with or without
   8 # modification, are permitted provided that the following conditions are
   9 # met:
  10 #
  11 # 1. Redistributions of source code must retain the above copyright notice,
  12 # this list of conditions and the following disclaimer.
  13 #
  14 # 2. Redistributions in binary form must reproduce the above copyright
  15 # notice, this list of conditions and the following disclaimer in the
  16 # documentation and/or other materials provided with the distribution.
  17 #
  18 # THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS
  19 # IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
  20 # TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
  21 # PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR
  22 # CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  23 # EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  24 # PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  25 # PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  26 # LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  27 # NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  28 # SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  29
  30
  31 """Script for doing QA on Ganeti.
  32
  33 """
  34
  35 # pylint: disable=C0103
  36 # due to invalid name
  37
  38 import copy
  39 import datetime
  40 import optparse
  41 import sys
  42
  43 import colors
  44 import qa_cluster
  45 import qa_config
  46 import qa_daemon
  47 import qa_env
  48 import qa_error
  49 import qa_filters
  50 import qa_group
  51 import qa_instance
  52 import qa_iptables
  53 import qa_monitoring
  54 import qa_network
  55 import qa_node
  56 import qa_os
  57 import qa_performance
  58 import qa_job
  59 import qa_rapi
  60 import qa_tags
  61 import qa_utils
  62
  63 from ganeti import utils
  64 from ganeti import rapi # pylint: disable=W0611
  65 from ganeti import constants
  66 from ganeti import netutils
  67
  68 import ganeti.rapi.client # pylint: disable=W0611
  69 from ganeti.rapi.client import UsesRapiClient
  70
  71
  72 def _FormatHeader(line, end=72, mark="-", color=None):
  73   """Fill a line up to the end column.
  74
  75   """
  76   line = (mark * 4) + " " + line + " "
  77   line += "-" * (end - len(line))
  78   line = line.rstrip()
  79   line = colors.colorize(line, color=color)
  80   return line
  81
  82
  83 def _DescriptionOf(fn):
  84   """Computes the description of an item.
  85
  86   """
  87   if fn.__doc__:
  88     desc = fn.__doc__.splitlines()[0].strip()
  89     desc = desc.rstrip(".")
  90     if fn.__name__:
  91       desc = "[" + fn.__name__ + "] " + desc
  92   else:
  93     desc = "%r" % fn
  94
  95   return desc
  96
  97
  98 def RunTest(fn, *args, **kwargs):
  99   """Runs a test after printing a header.
 100
 101   """
 102
 103   tstart = datetime.datetime.now()
 104
 105   desc = _DescriptionOf(fn)
 106
 107   print
 108   print _FormatHeader("%s start %s" % (tstart, desc),
 109                       color=colors.YELLOW, mark="<")
 110
 111   try:
 112     retval = fn(*args, **kwargs)
 113     print _FormatHeader("PASSED %s" % (desc, ), color=colors.GREEN)
 114     return retval
 115   except Exception, e:
 116     print _FormatHeader("FAILED %s: %s" % (desc, e), color=colors.RED)
 117     raise
 118   finally:
 119     tstop = datetime.datetime.now()
 120     tdelta = tstop - tstart
 121     print _FormatHeader("%s time=%s %s" % (tstop, tdelta, desc),
 122                         color=colors.MAGENTA, mark=">")
 123
 124
 125 def ReportTestSkip(desc, testnames):
 126   """Reports that tests have been skipped.
 127
 128   @type desc: string
 129   @param desc: string
 130   @type testnames: string or list of string
 131   @param testnames: either a single test name in the configuration
 132       file, or a list of testnames (which will be AND-ed together)
 133
 134   """
 135   tstart = datetime.datetime.now()
 136   # TODO: Formatting test names when non-string names are involved
 137   print _FormatHeader("%s skipping %s, test(s) %s disabled" %
 138                       (tstart, desc, testnames),
 139                       color=colors.BLUE, mark="*")
 140
 141
 142 def RunTestIf(testnames, fn, *args, **kwargs):
 143   """Runs a test conditionally.
 144
 145   @param testnames: either a single test name in the configuration
 146       file, or a list of testnames (which will be AND-ed together)
 147
 148   """
 149   if qa_config.TestEnabled(testnames):
 150     RunTest(fn, *args, **kwargs)
 151   else:
 152     desc = _DescriptionOf(fn)
 153     ReportTestSkip(desc, testnames)
 154
 155
 156 def RunTestBlock(fn, *args, **kwargs):
 157   """Runs a block of tests after printing a header.
 158
 159   """
 160   tstart = datetime.datetime.now()
 161
 162   desc = _DescriptionOf(fn)
 163
 164   print
 165   print _FormatHeader("BLOCK %s start %s" % (tstart, desc),
 166                       color=[colors.YELLOW, colors.BOLD], mark="v")
 167
 168   try:
 169     return fn(*args, **kwargs)
 170   except Exception, e:
 171     print _FormatHeader("BLOCK FAILED %s: %s" % (desc, e),
 172                         color=[colors.RED, colors.BOLD])
 173     raise
 174   finally:
 175     tstop = datetime.datetime.now()
 176     tdelta = tstop - tstart
 177     print _FormatHeader("BLOCK %s time=%s %s" % (tstop, tdelta, desc),
 178                         color=[colors.MAGENTA, colors.BOLD], mark="^")
 179
 180
 181 def RunEnvTests():
 182   """Run several environment tests.
 183
 184   """
 185   RunTestIf("env", qa_env.TestSshConnection)
 186   RunTestIf("env", qa_env.TestIcmpPing)
 187   RunTestIf("env", qa_env.TestGanetiCommands)
 188
 189
 190 def SetupCluster():
 191   """Initializes the cluster.
 192
 193   """
 194
 195   RunTestIf("create-cluster", qa_cluster.TestClusterInit)
 196   if not qa_config.TestEnabled("create-cluster"):
 197     # If the cluster is already in place, we assume that exclusive-storage is
 198     # already set according to the configuration
 199     qa_config.SetExclusiveStorage(qa_config.get("exclusive-storage", False))
 200
 201   qa_rapi.SetupRapi()
 202
 203   qa_group.ConfigureGroups()
 204
 205   # Test on empty cluster
 206   RunTestIf("node-list", qa_node.TestNodeList)
 207   RunTestIf("instance-list", qa_instance.TestInstanceList)
 208   RunTestIf("job-list", qa_job.TestJobList)
 209
 210   RunTestIf("create-cluster", qa_node.TestNodeAddAll)
 211   if not qa_config.TestEnabled("create-cluster"):
 212     # consider the nodes are already there
 213     qa_node.MarkNodeAddedAll()
 214
 215   RunTestIf("test-jobqueue", qa_cluster.TestJobqueue)
 216   RunTestIf("test-jobqueue", qa_job.TestJobCancellation)
 217
 218   # enable the watcher (unconditionally)
 219   RunTest(qa_daemon.TestResumeWatcher)
 220
 221   RunTestIf("node-list", qa_node.TestNodeList)
 222
 223   # Test listing fields
 224   RunTestIf("node-list", qa_node.TestNodeListFields)
 225   RunTestIf("instance-list", qa_instance.TestInstanceListFields)
 226   RunTestIf("job-list", qa_job.TestJobListFields)
 227   RunTestIf("instance-export", qa_instance.TestBackupListFields)
 228
 229   RunTestIf("node-info", qa_node.TestNodeInfo)
 230
 231
 232 def RunClusterTests():
 233   """Runs tests related to gnt-cluster.
 234
 235   """
 236   for test, fn in [
 237     ("create-cluster", qa_cluster.TestClusterInitDisk),
 238     ("cluster-renew-crypto", qa_cluster.TestClusterRenewCrypto)
 239     ]:
 240     RunTestIf(test, fn)
 241
 242   for test, fn in [
 243     ("cluster-verify", qa_cluster.TestClusterVerify),
 244     ("cluster-reserved-lvs", qa_cluster.TestClusterReservedLvs),
 245     # TODO: add more cluster modify tests
 246     ("cluster-modify", qa_cluster.TestClusterModifyEmpty),
 247     ("cluster-modify", qa_cluster.TestClusterModifyIPolicy),
 248     ("cluster-modify", qa_cluster.TestClusterModifyISpecs),
 249     ("cluster-modify", qa_cluster.TestClusterModifyBe),
 250     ("cluster-modify", qa_cluster.TestClusterModifyDisk),
 251     ("cluster-modify", qa_cluster.TestClusterModifyDiskTemplates),
 252     ("cluster-modify", qa_cluster.TestClusterModifyFileStorageDir),
 253     ("cluster-modify", qa_cluster.TestClusterModifySharedFileStorageDir),
 254     ("cluster-modify", qa_cluster.TestClusterModifyInstallImage),
 255     ("cluster-modify", qa_cluster.TestClusterModifyUserShutdown),
 256     ("cluster-rename", qa_cluster.TestClusterRename),
 257     ("cluster-info", qa_cluster.TestClusterVersion),
 258     ("cluster-info", qa_cluster.TestClusterInfo),
 259     ("cluster-info", qa_cluster.TestClusterGetmaster),
 260     ("cluster-redist-conf", qa_cluster.TestClusterRedistConf),
 261     (["cluster-copyfile", qa_config.NoVirtualCluster],
 262      qa_cluster.TestClusterCopyfile),
 263     ("cluster-command", qa_cluster.TestClusterCommand),
 264     ("cluster-burnin", qa_cluster.TestClusterBurnin),
 265     ("cluster-master-failover", qa_cluster.TestClusterMasterFailover),
 266     ("cluster-master-failover",
 267      qa_cluster.TestClusterMasterFailoverWithDrainedQueue),
 268     (["cluster-oob", qa_config.NoVirtualCluster],
 269      qa_cluster.TestClusterOob),
 270     ("cluster-instance-communication", qa_cluster.TestInstanceCommunication),
 271     (qa_rapi.Enabled, qa_rapi.TestVersion),
 272     (qa_rapi.Enabled, qa_rapi.TestEmptyCluster),
 273     (qa_rapi.Enabled, qa_rapi.TestRapiQuery),
 274     ]:
 275     RunTestIf(test, fn)
 276
 277
 278 def RunRepairDiskSizes():
 279   """Run the repair disk-sizes test.
 280
 281   """
 282   RunTestIf("cluster-repair-disk-sizes", qa_cluster.TestClusterRepairDiskSizes)
 283
 284
 285 def RunOsTests():
 286   """Runs all tests related to gnt-os.
 287
 288   """
 289   os_enabled = ["os", qa_config.NoVirtualCluster]
 290
 291   if qa_config.TestEnabled(qa_rapi.Enabled):
 292     rapi_getos = qa_rapi.GetOperatingSystems
 293   else:
 294     rapi_getos = None
 295
 296   for fn in [
 297     qa_os.TestOsList,
 298     qa_os.TestOsDiagnose,
 299     ]:
 300     RunTestIf(os_enabled, fn)
 301
 302   for fn in [
 303     qa_os.TestOsValid,
 304     qa_os.TestOsInvalid,
 305     qa_os.TestOsPartiallyValid,
 306     ]:
 307     RunTestIf(os_enabled, fn, rapi_getos)
 308
 309   for fn in [
 310     qa_os.TestOsModifyValid,
 311     qa_os.TestOsModifyInvalid,
 312     qa_os.TestOsStatesNonExisting,
 313     ]:
 314     RunTestIf(os_enabled, fn)
 315
 316
 317 def RunCommonInstanceTests(instance, inst_nodes):
 318   """Runs a few tests that are common to all disk types.
 319
 320   """
 321   RunTestIf("instance-shutdown", qa_instance.TestInstanceShutdown, instance)
 322   RunTestIf(["instance-shutdown", "instance-console", qa_rapi.Enabled],
 323             qa_rapi.TestRapiStoppedInstanceConsole, instance)
 324   RunTestIf(["instance-shutdown", "instance-modify"],
 325             qa_instance.TestInstanceStoppedModify, instance)
 326   RunTestIf("instance-shutdown", qa_instance.TestInstanceStartup, instance)
 327
 328   # Test shutdown/start via RAPI
 329   RunTestIf(["instance-shutdown", qa_rapi.Enabled],
 330             qa_rapi.TestRapiInstanceShutdown, instance)
 331   RunTestIf(["instance-shutdown", qa_rapi.Enabled],
 332             qa_rapi.TestRapiInstanceStartup, instance)
 333
 334   RunTestIf("instance-list", qa_instance.TestInstanceList)
 335
 336   RunTestIf("instance-info", qa_instance.TestInstanceInfo, instance)
 337
 338   RunTestIf("instance-modify", qa_instance.TestInstanceModify, instance)
 339   RunTestIf(["instance-modify", qa_rapi.Enabled],
 340             qa_rapi.TestRapiInstanceModify, instance)
 341
 342   RunTestIf("instance-console", qa_instance.TestInstanceConsole, instance)
 343   RunTestIf(["instance-console", qa_rapi.Enabled],
 344             qa_rapi.TestRapiInstanceConsole, instance)
 345
 346   RunTestIf("instance-device-names", qa_instance.TestInstanceDeviceNames,
 347             instance)
 348   DOWN_TESTS = qa_config.Either([
 349     "instance-reinstall",
 350     "instance-rename",
 351     "instance-grow-disk",
 352     ])
 353
 354   # shutdown instance for any 'down' tests
 355   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceShutdown, instance)
 356
 357   # now run the 'down' state tests
 358   RunTestIf("instance-reinstall", qa_instance.TestInstanceReinstall, instance)
 359   RunTestIf(["instance-reinstall", qa_rapi.Enabled],
 360             qa_rapi.TestRapiInstanceReinstall, instance)
 361
 362   if qa_config.TestEnabled("instance-rename"):
 363     tgt_instance = qa_config.AcquireInstance()
 364     try:
 365       rename_source = instance.name
 366       rename_target = tgt_instance.name
 367       # perform instance rename to the same name
 368       RunTest(qa_instance.TestInstanceRenameAndBack,
 369               rename_source, rename_source)
 370       RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceRenameAndBack,
 371                 rename_source, rename_source)
 372       if rename_target is not None:
 373         # perform instance rename to a different name, if we have one configured
 374         RunTest(qa_instance.TestInstanceRenameAndBack,
 375                 rename_source, rename_target)
 376         RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceRenameAndBack,
 377                   rename_source, rename_target)
 378     finally:
 379       tgt_instance.Release()
 380
 381   RunTestIf(["instance-grow-disk"], qa_instance.TestInstanceGrowDisk, instance)
 382
 383   # and now start the instance again
 384   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceStartup, instance)
 385
 386   RunTestIf("instance-reboot", qa_instance.TestInstanceReboot, instance)
 387
 388   RunTestIf("tags", qa_tags.TestInstanceTags, instance)
 389
 390   if instance.disk_template == constants.DT_DRBD8:
 391     RunTestIf("cluster-verify",
 392               qa_cluster.TestClusterVerifyDisksBrokenDRBD, instance, inst_nodes)
 393   RunTestIf("cluster-verify", qa_cluster.TestClusterVerify)
 394
 395   RunTestIf(qa_rapi.Enabled, qa_rapi.TestInstance, instance)
 396
 397   # Lists instances, too
 398   RunTestIf("node-list", qa_node.TestNodeList)
 399
 400   # Some jobs have been run, let's test listing them
 401   RunTestIf("job-list", qa_job.TestJobList)
 402
 403
 404 def RunCommonNodeTests():
 405   """Run a few common node tests.
 406
 407   """
 408   RunTestIf("node-volumes", qa_node.TestNodeVolumes)
 409   RunTestIf("node-storage", qa_node.TestNodeStorage)
 410   RunTestIf(["node-oob", qa_config.NoVirtualCluster], qa_node.TestOutOfBand)
 411
 412
 413 def RunGroupListTests():
 414   """Run tests for listing node groups.
 415
 416   """
 417   RunTestIf("group-list", qa_group.TestGroupList)
 418   RunTestIf("group-list", qa_group.TestGroupListFields)
 419
 420
 421 def RunNetworkTests():
 422   """Run tests for network management.
 423
 424   """
 425   RunTestIf("network", qa_network.TestNetworkAddRemove)
 426   RunTestIf("network", qa_network.TestNetworkConnect)
 427   RunTestIf(["network", "tags"], qa_network.TestNetworkTags)
 428
 429
 430 def RunFilterTests():
 431   """Run tests for job filter management.
 432
 433   """
 434   RunTestIf("filters", qa_filters.TestFilterList)
 435   RunTestIf("filters", qa_filters.TestFilterListFields)
 436   RunTestIf("filters", qa_filters.TestFilterAddRemove)
 437   RunTestIf("filters", qa_filters.TestFilterReject)
 438   RunTestIf("filters", qa_filters.TestFilterOpCode)
 439   RunTestIf("filters", qa_filters.TestFilterReasonChain)
 440   RunTestIf("filters", qa_filters.TestFilterContinue)
 441   RunTestIf("filters", qa_filters.TestFilterAcceptPause)
 442   RunTestIf("filters", qa_filters.TestFilterWatermark)
 443   RunTestIf("filters", qa_filters.TestFilterRateLimit)
 444   RunTestIf("filters", qa_filters.TestAdHocReasonRateLimit)
 445
 446
 447 def RunGroupRwTests():
 448   """Run tests for adding/removing/renaming groups.
 449
 450   """
 451   RunTestIf("group-rwops", qa_group.TestGroupAddRemoveRename)
 452   RunTestIf("group-rwops", qa_group.TestGroupAddWithOptions)
 453   RunTestIf("group-rwops", qa_group.TestGroupModify)
 454   RunTestIf(["group-rwops", qa_rapi.Enabled], qa_rapi.TestRapiNodeGroups)
 455   RunTestIf(["group-rwops", "tags"], qa_tags.TestGroupTags,
 456             qa_group.GetDefaultGroup())
 457
 458
 459 def RunExportImportTests(instance, inodes):
 460   """Tries to export and import the instance.
 461
 462   @type inodes: list of nodes
 463   @param inodes: current nodes of the instance
 464
 465   """
 466   # FIXME: export explicitly bails out on file based storage. other non-lvm
 467   # based storage types are untested, though. Also note that import could still
 468   # work, but is deeply embedded into the "export" case.
 469   if qa_config.TestEnabled("instance-export"):
 470     RunTest(qa_instance.TestInstanceExportNoTarget, instance)
 471
 472     pnode = inodes[0]
 473     expnode = qa_config.AcquireNode(exclude=pnode)
 474     try:
 475       name = RunTest(qa_instance.TestInstanceExport, instance, expnode)
 476
 477       RunTest(qa_instance.TestBackupList, expnode)
 478
 479       if qa_config.TestEnabled("instance-import"):
 480         newinst = qa_config.AcquireInstance()
 481         try:
 482           RunTest(qa_instance.TestInstanceImport, newinst, pnode,
 483                   expnode, name)
 484           # Check if starting the instance works
 485           RunTest(qa_instance.TestInstanceStartup, newinst)
 486           RunTest(qa_instance.TestInstanceRemove, newinst)
 487         finally:
 488           newinst.Release()
 489     finally:
 490       expnode.Release()
 491
 492   # FIXME: inter-cluster-instance-move crashes on file based instances :/
 493   # See Issue 414.
 494   if (qa_config.TestEnabled([qa_rapi.Enabled, "inter-cluster-instance-move"])):
 495     newinst = qa_config.AcquireInstance()
 496     try:
 497       tnode = qa_config.AcquireNode(exclude=inodes)
 498       try:
 499         RunTest(qa_rapi.TestInterClusterInstanceMove, instance, newinst,
 500                 inodes, tnode)
 501       finally:
 502         tnode.Release()
 503     finally:
 504       newinst.Release()
 505
 506
 507 def RunDaemonTests(instance):
 508   """Test the ganeti-watcher script.
 509
 510   """
 511   RunTest(qa_daemon.TestPauseWatcher)
 512
 513   RunTestIf("instance-automatic-restart",
 514             qa_daemon.TestInstanceAutomaticRestart, instance)
 515   RunTestIf("instance-consecutive-failures",
 516             qa_daemon.TestInstanceConsecutiveFailures, instance)
 517
 518   RunTest(qa_daemon.TestResumeWatcher)
 519
 520
 521 def RunHardwareFailureTests(instance, inodes):
 522   """Test cluster internal hardware failure recovery.
 523
 524   """
 525   RunTestIf("instance-failover", qa_instance.TestInstanceFailover, instance)
 526   RunTestIf(["instance-failover", qa_rapi.Enabled],
 527             qa_rapi.TestRapiInstanceFailover, instance)
 528
 529   RunTestIf("instance-migrate", qa_instance.TestInstanceMigrate, instance)
 530   RunTestIf(["instance-migrate", qa_rapi.Enabled],
 531             qa_rapi.TestRapiInstanceMigrate, instance)
 532
 533   if qa_config.TestEnabled("instance-replace-disks"):
 534     # We just need alternative secondary nodes, hence "- 1"
 535     othernodes = qa_config.AcquireManyNodes(len(inodes) - 1, exclude=inodes)
 536     try:
 537       RunTestIf(qa_rapi.Enabled, qa_rapi.TestRapiInstanceReplaceDisks, instance)
 538       RunTest(qa_instance.TestReplaceDisks,
 539               instance, inodes, othernodes)
 540     finally:
 541       qa_config.ReleaseManyNodes(othernodes)
 542     del othernodes
 543
 544   if qa_config.TestEnabled("instance-recreate-disks"):
 545     try:
 546       acquirednodes = qa_config.AcquireManyNodes(len(inodes), exclude=inodes)
 547       othernodes = acquirednodes
 548     except qa_error.OutOfNodesError:
 549       if len(inodes) > 1:
 550         # If the cluster is not big enough, let's reuse some of the nodes, but
 551         # with different roles. In this way, we can test a DRBD instance even on
 552         # a 3-node cluster.
 553         acquirednodes = [qa_config.AcquireNode(exclude=inodes)]
 554         othernodes = acquirednodes + inodes[:-1]
 555       else:
 556         raise
 557     try:
 558       RunTest(qa_instance.TestRecreateDisks,
 559               instance, inodes, othernodes)
 560     finally:
 561       qa_config.ReleaseManyNodes(acquirednodes)
 562
 563   if len(inodes) >= 2:
 564     RunTestIf("node-evacuate", qa_node.TestNodeEvacuate, inodes[0], inodes[1])
 565     RunTestIf("node-failover", qa_node.TestNodeFailover, inodes[0], inodes[1])
 566     RunTestIf("node-migrate", qa_node.TestNodeMigrate, inodes[0], inodes[1])
 567
 568
 569 def RunExclusiveStorageTests():
 570   """Test exclusive storage."""
 571   if not qa_config.TestEnabled("cluster-exclusive-storage"):
 572     return
 573
 574   node = qa_config.AcquireNode()
 575   try:
 576     old_es = qa_cluster.TestSetExclStorCluster(False)
 577     qa_node.TestExclStorSingleNode(node)
 578
 579     qa_cluster.TestSetExclStorCluster(True)
 580     qa_cluster.TestExclStorSharedPv(node)
 581
 582     if qa_config.TestEnabled("instance-add-plain-disk"):
 583       # Make sure that the cluster doesn't have any pre-existing problem
 584       qa_cluster.AssertClusterVerify()
 585
 586       # Create and allocate instances
 587       instance1 = qa_instance.TestInstanceAddWithPlainDisk([node])
 588       try:
 589         instance2 = qa_instance.TestInstanceAddWithPlainDisk([node])
 590         try:
 591           # cluster-verify checks that disks are allocated correctly
 592           qa_cluster.AssertClusterVerify()
 593
 594           # Remove instances
 595           qa_instance.TestInstanceRemove(instance2)
 596           qa_instance.TestInstanceRemove(instance1)
 597         finally:
 598           instance2.Release()
 599       finally:
 600         instance1.Release()
 601
 602     if qa_config.TestEnabled("instance-add-drbd-disk"):
 603       snode = qa_config.AcquireNode()
 604       try:
 605         qa_cluster.TestSetExclStorCluster(False)
 606         instance = qa_instance.TestInstanceAddWithDrbdDisk([node, snode])
 607         try:
 608           qa_cluster.TestSetExclStorCluster(True)
 609           exp_err = [constants.CV_EINSTANCEUNSUITABLENODE]
 610           qa_cluster.AssertClusterVerify(fail=True, errors=exp_err)
 611           qa_instance.TestInstanceRemove(instance)
 612         finally:
 613           instance.Release()
 614       finally:
 615         snode.Release()
 616     qa_cluster.TestSetExclStorCluster(old_es)
 617   finally:
 618     node.Release()
 619
 620
 621 def RunCustomSshPortTests():
 622   """Test accessing nodes with custom SSH ports.
 623
 624   This requires removing nodes, adding them to a new group, and then undoing
 625   the change.
 626   """
 627   if not qa_config.TestEnabled("group-custom-ssh-port"):
 628     return
 629
 630   std_port = netutils.GetDaemonPort(constants.SSH)
 631   port = 211
 632   master = qa_config.GetMasterNode()
 633   with qa_config.AcquireManyNodesCtx(1, exclude=master) as nodes:
 634     # Checks if the node(s) could be contacted through IPv6.
 635     # If yes, better skip the whole test.
 636
 637     for node in nodes:
 638       if qa_utils.UsesIPv6Connection(node.primary, std_port):
 639         print ("Node %s is likely to be reached using IPv6,"
 640                "skipping the test" % (node.primary, ))
 641         return
 642
 643     for node in nodes:
 644       qa_node.NodeRemove(node)
 645     with qa_iptables.RulesContext() as r:
 646       with qa_group.NewGroupCtx() as group:
 647         qa_group.ModifyGroupSshPort(r, group, nodes, port)
 648
 649         for node in nodes:
 650           qa_node.NodeAdd(node, group=group)
 651
 652         # Make sure that the cluster doesn't have any pre-existing problem
 653         qa_cluster.AssertClusterVerify()
 654
 655         # Create and allocate instances
 656         instance1 = qa_instance.TestInstanceAddWithPlainDisk(nodes)
 657         try:
 658           instance2 = qa_instance.TestInstanceAddWithPlainDisk(nodes)
 659           try:
 660             # cluster-verify checks that disks are allocated correctly
 661             qa_cluster.AssertClusterVerify()
 662
 663             # Remove instances
 664             qa_instance.TestInstanceRemove(instance2)
 665             qa_instance.TestInstanceRemove(instance1)
 666           finally:
 667             instance2.Release()
 668         finally:
 669           instance1.Release()
 670
 671         for node in nodes:
 672           qa_node.NodeRemove(node)
 673
 674     for node in nodes:
 675       qa_node.NodeAdd(node)
 676
 677     qa_cluster.AssertClusterVerify()
 678
 679
 680 def _BuildSpecDict(par, mn, st, mx):
 681   return {
 682     constants.ISPECS_MINMAX: [{
 683       constants.ISPECS_MIN: {par: mn},
 684       constants.ISPECS_MAX: {par: mx},
 685       }],
 686     constants.ISPECS_STD: {par: st},
 687     }
 688
 689
 690 def _BuildDoubleSpecDict(index, par, mn, st, mx):
 691   new_spec = {
 692     constants.ISPECS_MINMAX: [{}, {}],
 693     }
 694   if st is not None:
 695     new_spec[constants.ISPECS_STD] = {par: st}
 696   new_spec[constants.ISPECS_MINMAX][index] = {
 697     constants.ISPECS_MIN: {par: mn},
 698     constants.ISPECS_MAX: {par: mx},
 699     }
 700   return new_spec
 701
 702
 703 def TestIPolicyPlainInstance():
 704   """Test instance policy interaction with instances"""
 705   params = ["memory-size", "cpu-count", "disk-count", "disk-size", "nic-count"]
 706   if not qa_config.IsTemplateSupported(constants.DT_PLAIN):
 707     print "Template %s not supported" % constants.DT_PLAIN
 708     return
 709
 710   # This test assumes that the group policy is empty
 711   (_, old_specs) = qa_cluster.TestClusterSetISpecs()
 712   # We also assume to have only one min/max bound
 713   assert len(old_specs[constants.ISPECS_MINMAX]) == 1
 714   node = qa_config.AcquireNode()
 715   try:
 716     # Log of policy changes, list of tuples:
 717     # (full_change, incremental_change, policy_violated)
 718     history = []
 719     instance = qa_instance.TestInstanceAddWithPlainDisk([node])
 720     try:
 721       policyerror = [constants.CV_EINSTANCEPOLICY]
 722       for par in params:
 723         (iminval, imaxval) = qa_instance.GetInstanceSpec(instance.name, par)
 724         # Some specs must be multiple of 4
 725         new_spec = _BuildSpecDict(par, imaxval + 4, imaxval + 4, imaxval + 4)
 726         history.append((None, new_spec, True))
 727         if iminval > 0:
 728           # Some specs must be multiple of 4
 729           if iminval >= 4:
 730             upper = iminval - 4
 731           else:
 732             upper = iminval - 1
 733           new_spec = _BuildSpecDict(par, 0, upper, upper)
 734           history.append((None, new_spec, True))
 735         history.append((old_specs, None, False))
 736
 737       # Test with two instance specs
 738       double_specs = copy.deepcopy(old_specs)
 739       double_specs[constants.ISPECS_MINMAX] = \
 740           double_specs[constants.ISPECS_MINMAX] * 2
 741       (par1, par2) = params[0:2]
 742       (_, imaxval1) = qa_instance.GetInstanceSpec(instance.name, par1)
 743       (_, imaxval2) = qa_instance.GetInstanceSpec(instance.name, par2)
 744       old_minmax = old_specs[constants.ISPECS_MINMAX][0]
 745       history.extend([
 746         (double_specs, None, False),
 747         # The first min/max limit is being violated
 748         (None,
 749          _BuildDoubleSpecDict(0, par1, imaxval1 + 4, imaxval1 + 4,
 750                               imaxval1 + 4),
 751          False),
 752         # Both min/max limits are being violated
 753         (None,
 754          _BuildDoubleSpecDict(1, par2, imaxval2 + 4, None, imaxval2 + 4),
 755          True),
 756         # The second min/max limit is being violated
 757         (None,
 758          _BuildDoubleSpecDict(0, par1,
 759                               old_minmax[constants.ISPECS_MIN][par1],
 760                               old_specs[constants.ISPECS_STD][par1],
 761                               old_minmax[constants.ISPECS_MAX][par1]),
 762          False),
 763         (old_specs, None, False),
 764         ])
 765
 766       # Apply the changes, and check policy violations after each change
 767       qa_cluster.AssertClusterVerify()
 768       for (new_specs, diff_specs, failed) in history:
 769         qa_cluster.TestClusterSetISpecs(new_specs=new_specs,
 770                                         diff_specs=diff_specs)
 771         if failed:
 772           qa_cluster.AssertClusterVerify(warnings=policyerror)
 773         else:
 774           qa_cluster.AssertClusterVerify()
 775
 776       qa_instance.TestInstanceRemove(instance)
 777     finally:
 778       instance.Release()
 779
 780     # Now we replay the same policy changes, and we expect that the instance
 781     # cannot be created for the cases where we had a policy violation above
 782     for (new_specs, diff_specs, failed) in history:
 783       qa_cluster.TestClusterSetISpecs(new_specs=new_specs,
 784                                       diff_specs=diff_specs)
 785       if failed:
 786         qa_instance.TestInstanceAddWithPlainDisk([node], fail=True)
 787       # Instance creation with no policy violation has been tested already
 788   finally:
 789     node.Release()
 790
 791
 792 def IsExclusiveStorageInstanceTestEnabled():
 793   test_name = "exclusive-storage-instance-tests"
 794   if qa_config.TestEnabled(test_name):
 795     vgname = qa_config.get("vg-name", constants.DEFAULT_VG)
 796     vgscmd = utils.ShellQuoteArgs([
 797       "vgs", "--noheadings", "-o", "pv_count", vgname,
 798       ])
 799     nodes = qa_config.GetConfig()["nodes"]
 800     for node in nodes:
 801       try:
 802         pvnum = int(qa_utils.GetCommandOutput(node.primary, vgscmd))
 803       except Exception, e:
 804         msg = ("Cannot get the number of PVs on %s, needed by '%s': %s" %
 805                (node.primary, test_name, e))
 806         raise qa_error.Error(msg)
 807       if pvnum < 2:
 808         raise qa_error.Error("Node %s has not enough PVs (%s) to run '%s'" %
 809                              (node.primary, pvnum, test_name))
 810     res = True
 811   else:
 812     res = False
 813   return res
 814
 815
 816 def RunInstanceTests():
 817   """Create and exercise instances."""
 818
 819   requested_conversions = qa_config.get("convert-disk-templates", [])
 820   supported_conversions = \
 821       set(requested_conversions).difference(constants.DTS_NOT_CONVERTIBLE_TO)
 822   for (test_name, templ, create_fun, num_nodes) in \
 823       qa_instance.available_instance_tests:
 824     if (qa_config.TestEnabled(test_name) and
 825         qa_config.IsTemplateSupported(templ)):
 826       inodes = qa_config.AcquireManyNodes(num_nodes)
 827       try:
 828         instance = RunTest(create_fun, inodes)
 829         try:
 830           RunTestIf("instance-user-down", qa_instance.TestInstanceUserDown,
 831                     instance)
 832           RunTestIf("instance-communication",
 833                     qa_instance.TestInstanceCommunication,
 834                     instance,
 835                     qa_config.GetMasterNode())
 836           RunTestIf("cluster-epo", qa_cluster.TestClusterEpo)
 837           RunDaemonTests(instance)
 838           for node in inodes:
 839             RunTestIf("haskell-confd", qa_node.TestNodeListDrbd, node,
 840                       templ == constants.DT_DRBD8)
 841           if len(inodes) > 1:
 842             RunTestIf("group-rwops", qa_group.TestAssignNodesIncludingSplit,
 843                       constants.INITIAL_NODE_GROUP_NAME,
 844                       inodes[0].primary, inodes[1].primary)
 845           # This test will run once but it will cover all the supported
 846           # user-provided disk template conversions
 847           if qa_config.TestEnabled("instance-convert-disk"):
 848             if (len(supported_conversions) > 1 and
 849                 instance.disk_template in supported_conversions):
 850               RunTest(qa_instance.TestInstanceShutdown, instance)
 851               RunTest(qa_instance.TestInstanceConvertDiskTemplate, instance,
 852                       supported_conversions)
 853               RunTest(qa_instance.TestInstanceStartup, instance)
 854               # At this point we clear the set because the requested conversions
 855               # has been tested
 856               supported_conversions.clear()
 857             else:
 858               test_desc = "Converting instance of template %s" % templ
 859               ReportTestSkip(test_desc, "conversion feature")
 860           RunTestIf("instance-modify-disks",
 861                     qa_instance.TestInstanceModifyDisks, instance)
 862           RunCommonInstanceTests(instance, inodes)
 863           if qa_config.TestEnabled("instance-modify-primary"):
 864             othernode = qa_config.AcquireNode()
 865             RunTest(qa_instance.TestInstanceModifyPrimaryAndBack,
 866                     instance, inodes[0], othernode)
 867             othernode.Release()
 868           RunGroupListTests()
 869           RunExportImportTests(instance, inodes)
 870           RunHardwareFailureTests(instance, inodes)
 871           RunRepairDiskSizes()
 872           RunTestIf(["rapi", "instance-data-censorship"],
 873                     qa_rapi.TestInstanceDataCensorship, instance, inodes)
 874           RunTest(qa_instance.TestInstanceRemove, instance)
 875         finally:
 876           instance.Release()
 877         del instance
 878       finally:
 879         qa_config.ReleaseManyNodes(inodes)
 880       qa_cluster.AssertClusterVerify()
 881     else:
 882       test_desc = "Creating instances of template %s" % templ
 883       if not qa_config.TestEnabled(test_name):
 884         ReportTestSkip(test_desc, test_name)
 885       else:
 886         ReportTestSkip(test_desc, "disk template %s" % templ)
 887
 888
 889 def RunMonitoringTests():
 890   RunTestIf("mon-collector", qa_monitoring.TestInstStatusCollector)
 891
 892
 893 PARALLEL_TEST_DICT = {
 894   "parallel-failover": qa_performance.TestParallelInstanceFailover,
 895   "parallel-migration": qa_performance.TestParallelInstanceMigration,
 896   "parallel-replace-disks": qa_performance.TestParallelInstanceReplaceDisks,
 897   "parallel-reboot": qa_performance.TestParallelInstanceReboot,
 898   "parallel-reinstall": qa_performance.TestParallelInstanceReinstall,
 899   "parallel-rename": qa_performance.TestParallelInstanceRename,
 900   }
 901
 902
 903 def RunPerformanceTests():
 904   if not qa_config.TestEnabled("performance"):
 905     ReportTestSkip("performance related tests", "performance")
 906     return
 907
 908   # For reproducable performance, run performance tests with the watcher
 909   # paused.
 910   qa_utils.AssertCommand(["gnt-cluster", "watcher", "pause", "4h"])
 911
 912   if qa_config.TestEnabled("jobqueue-performance"):
 913     RunTest(qa_performance.TestParallelMaxInstanceCreationPerformance)
 914     RunTest(qa_performance.TestParallelNodeCountInstanceCreationPerformance)
 915
 916     instances = qa_performance.CreateAllInstances()
 917
 918     RunTest(qa_performance.TestParallelModify, instances)
 919     RunTest(qa_performance.TestParallelInstanceOSOperations, instances)
 920     RunTest(qa_performance.TestParallelInstanceQueries, instances)
 921
 922     qa_performance.RemoveAllInstances(instances)
 923
 924     RunTest(qa_performance.TestJobQueueSubmissionPerformance)
 925
 926   if qa_config.TestEnabled("parallel-performance"):
 927     if qa_config.IsTemplateSupported(constants.DT_DRBD8):
 928       RunTest(qa_performance.TestParallelDRBDInstanceCreationPerformance)
 929     if qa_config.IsTemplateSupported(constants.DT_PLAIN):
 930       RunTest(qa_performance.TestParallelPlainInstanceCreationPerformance)
 931
 932   # Preparations need to be made only if some of these tests are enabled
 933   if qa_config.IsTemplateSupported(constants.DT_DRBD8) and \
 934      qa_config.TestEnabled(qa_config.Either(PARALLEL_TEST_DICT.keys())):
 935     inodes = qa_config.AcquireManyNodes(2)
 936     try:
 937       instance = qa_instance.TestInstanceAddWithDrbdDisk(inodes)
 938       try:
 939         for (test_name, test_fn) in PARALLEL_TEST_DICT.items():
 940           RunTestIf(test_name, test_fn, instance)
 941       finally:
 942         instance.Release()
 943       qa_instance.TestInstanceRemove(instance)
 944     finally:
 945       qa_config.ReleaseManyNodes(inodes)
 946
 947   qa_utils.AssertCommand(["gnt-cluster", "watcher", "continue"])
 948
 949
 950 def RunQa():
 951   """Main QA body.
 952
 953   """
 954   RunTestBlock(RunEnvTests)
 955   SetupCluster()
 956
 957   RunTestBlock(RunClusterTests)
 958   RunTestBlock(RunOsTests)
 959
 960   RunTestIf("tags", qa_tags.TestClusterTags)
 961
 962   RunTestBlock(RunCommonNodeTests)
 963   RunTestBlock(RunGroupListTests)
 964   RunTestBlock(RunGroupRwTests)
 965   RunTestBlock(RunNetworkTests)
 966   RunTestBlock(RunFilterTests)
 967
 968   # The master shouldn't be readded or put offline; "delay" needs a non-master
 969   # node to test
 970   pnode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
 971   try:
 972     RunTestIf("node-readd", qa_node.TestNodeReadd, pnode)
 973     RunTestIf("node-modify", qa_node.TestNodeModify, pnode)
 974     RunTestIf("delay", qa_cluster.TestDelay, pnode)
 975   finally:
 976     pnode.Release()
 977
 978   # Make sure the cluster is clean before running instance tests
 979   qa_cluster.AssertClusterVerify()
 980
 981   pnode = qa_config.AcquireNode()
 982   try:
 983     RunTestIf("tags", qa_tags.TestNodeTags, pnode)
 984
 985     if qa_rapi.Enabled():
 986       RunTest(qa_rapi.TestNode, pnode)
 987
 988       if (qa_config.TestEnabled("instance-add-plain-disk")
 989           and qa_config.IsTemplateSupported(constants.DT_PLAIN)):
 990         # Normal instance allocation via RAPI
 991         for use_client in [True, False]:
 992           rapi_instance = RunTest(qa_rapi.TestRapiInstanceAdd, pnode,
 993                                   use_client)
 994           try:
 995             if qa_config.TestEnabled("instance-plain-rapi-common-tests"):
 996               RunCommonInstanceTests(rapi_instance, [pnode])
 997             RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance, use_client)
 998           finally:
 999             rapi_instance.Release()
1000           del rapi_instance
1001
1002         # Multi-instance allocation
1003         rapi_instance_one, rapi_instance_two = \
1004           RunTest(qa_rapi.TestRapiInstanceMultiAlloc, pnode)
1005
1006         try:
1007           RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance_one, True)
1008           RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance_two, True)
1009         finally:
1010           rapi_instance_one.Release()
1011           rapi_instance_two.Release()
1012   finally:
1013     pnode.Release()
1014
1015   config_list = [
1016     ("default-instance-tests", lambda: None, lambda _: None),
1017     (IsExclusiveStorageInstanceTestEnabled,
1018      lambda: qa_cluster.TestSetExclStorCluster(True),
1019      qa_cluster.TestSetExclStorCluster),
1020   ]
1021   for (conf_name, setup_conf_f, restore_conf_f) in config_list:
1022     if qa_config.TestEnabled(conf_name):
1023       oldconf = setup_conf_f()
1024       RunTestBlock(RunInstanceTests)
1025       restore_conf_f(oldconf)
1026
1027   pnode = qa_config.AcquireNode()
1028   try:
1029     if qa_config.TestEnabled(["instance-add-plain-disk", "instance-export"]):
1030       for shutdown in [False, True]:
1031         instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, [pnode])
1032         try:
1033           expnode = qa_config.AcquireNode(exclude=pnode)
1034           try:
1035             if shutdown:
1036               # Stop instance before exporting and removing it
1037               RunTest(qa_instance.TestInstanceShutdown, instance)
1038             RunTest(qa_instance.TestInstanceExportWithRemove, instance, expnode)
1039             RunTest(qa_instance.TestBackupList, expnode)
1040           finally:
1041             expnode.Release()
1042         finally:
1043           instance.Release()
1044         del expnode
1045         del instance
1046       qa_cluster.AssertClusterVerify()
1047
1048   finally:
1049     pnode.Release()
1050
1051   if qa_rapi.Enabled():
1052     RunTestIf("filters", qa_rapi.TestFilters)
1053
1054   RunTestIf("cluster-upgrade", qa_cluster.TestUpgrade)
1055
1056   RunTestBlock(RunExclusiveStorageTests)
1057   RunTestIf(["cluster-instance-policy", "instance-add-plain-disk"],
1058             TestIPolicyPlainInstance)
1059
1060   RunTestBlock(RunCustomSshPortTests)
1061
1062   RunTestIf(
1063     "instance-add-restricted-by-disktemplates",
1064     qa_instance.TestInstanceCreationRestrictedByDiskTemplates)
1065
1066   # Test removing instance with offline drbd secondary
1067   if qa_config.TestEnabled(["instance-remove-drbd-offline",
1068                             "instance-add-drbd-disk"]):
1069     # Make sure the master is not put offline
1070     snode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
1071     try:
1072       pnode = qa_config.AcquireNode(exclude=snode)
1073       try:
1074         instance = qa_instance.TestInstanceAddWithDrbdDisk([pnode, snode])
1075         set_offline = lambda node: qa_node.MakeNodeOffline(node, "yes")
1076         set_online = lambda node: qa_node.MakeNodeOffline(node, "no")
1077         RunTest(qa_instance.TestRemoveInstanceOfflineNode, instance, snode,
1078                 set_offline, set_online)
1079       finally:
1080         pnode.Release()
1081     finally:
1082       snode.Release()
1083     qa_cluster.AssertClusterVerify()
1084
1085   RunTestBlock(RunMonitoringTests)
1086
1087   RunPerformanceTests()
1088
1089   RunTestIf("cluster-destroy", qa_node.TestNodeRemoveAll)
1090
1091   RunTestIf("cluster-destroy", qa_cluster.TestClusterDestroy)
1092
1093
1094 @UsesRapiClient
1095 def main():
1096   """Main program.
1097
1098   """
1099   colors.check_for_colors()
1100
1101   parser = optparse.OptionParser(usage="%prog [options] <config-file>")
1102   parser.add_option("--yes-do-it", dest="yes_do_it",
1103                     action="store_true",
1104                     help="Really execute the tests")
1105   (opts, args) = parser.parse_args()
1106
1107   if len(args) == 1:
1108     (config_file, ) = args
1109   else:
1110     parser.error("Wrong number of arguments.")
1111
1112   if not opts.yes_do_it:
1113     print ("Executing this script irreversibly destroys any Ganeti\n"
1114            "configuration on all nodes involved. If you really want\n"
1115            "to start testing, supply the --yes-do-it option.")
1116     sys.exit(1)
1117
1118   qa_config.Load(config_file)
1119
1120   primary = qa_config.GetMasterNode().primary
1121   qa_utils.StartMultiplexer(primary)
1122   print ("SSH command for primary node: %s" %
1123          utils.ShellQuoteArgs(qa_utils.GetSSHCommand(primary, "")))
1124   print ("SSH command for other nodes: %s" %
1125          utils.ShellQuoteArgs(qa_utils.GetSSHCommand("NODE", "")))
1126   try:
1127     RunQa()
1128   finally:
1129     qa_utils.CloseMultiplexers()
1130
1131 if __name__ == "__main__":
1132   main()