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 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 sys
  30 import datetime
  31 import optparse
  32
  33 import qa_cluster
  34 import qa_config
  35 import qa_daemon
  36 import qa_env
  37 import qa_error
  38 import qa_group
  39 import qa_instance
  40 import qa_node
  41 import qa_os
  42 import qa_job
  43 import qa_rapi
  44 import qa_tags
  45 import qa_utils
  46
  47 from ganeti import utils
  48 from ganeti import rapi
  49 from ganeti import constants
  50
  51 import ganeti.rapi.client # pylint: disable=W0611
  52
  53
  54 def _FormatHeader(line, end=72):
  55   """Fill a line up to the end column.
  56
  57   """
  58   line = "---- " + line + " "
  59   line += "-" * (end - len(line))
  60   line = line.rstrip()
  61   return line
  62
  63
  64 def _DescriptionOf(fn):
  65   """Computes the description of an item.
  66
  67   """
  68   if fn.__doc__:
  69     desc = fn.__doc__.splitlines()[0].strip()
  70   else:
  71     desc = "%r" % fn
  72
  73   return desc.rstrip(".")
  74
  75
  76 def RunTest(fn, *args, **kwargs):
  77   """Runs a test after printing a header.
  78
  79   """
  80
  81   tstart = datetime.datetime.now()
  82
  83   desc = _DescriptionOf(fn)
  84
  85   print
  86   print _FormatHeader("%s start %s" % (tstart, desc))
  87
  88   try:
  89     retval = fn(*args, **kwargs)
  90     return retval
  91   finally:
  92     tstop = datetime.datetime.now()
  93     tdelta = tstop - tstart
  94     print _FormatHeader("%s time=%s %s" % (tstop, tdelta, desc))
  95
  96
  97 def RunTestIf(testnames, fn, *args, **kwargs):
  98   """Runs a test conditionally.
  99
 100   @param testnames: either a single test name in the configuration
 101       file, or a list of testnames (which will be AND-ed together)
 102
 103   """
 104   if qa_config.TestEnabled(testnames):
 105     RunTest(fn, *args, **kwargs)
 106   else:
 107     tstart = datetime.datetime.now()
 108     desc = _DescriptionOf(fn)
 109     print _FormatHeader("%s skipping %s, test(s) %s disabled" %
 110                         (tstart, desc, testnames))
 111
 112
 113 def RunEnvTests():
 114   """Run several environment tests.
 115
 116   """
 117   RunTestIf("env", qa_env.TestSshConnection)
 118   RunTestIf("env", qa_env.TestIcmpPing)
 119   RunTestIf("env", qa_env.TestGanetiCommands)
 120
 121
 122 def SetupCluster(rapi_user, rapi_secret):
 123   """Initializes the cluster.
 124
 125   @param rapi_user: Login user for RAPI
 126   @param rapi_secret: Login secret for RAPI
 127
 128   """
 129   RunTestIf("create-cluster", qa_cluster.TestClusterInit,
 130             rapi_user, rapi_secret)
 131
 132   # Test on empty cluster
 133   RunTestIf("node-list", qa_node.TestNodeList)
 134   RunTestIf("instance-list", qa_instance.TestInstanceList)
 135   RunTestIf("job-list", qa_job.TestJobList)
 136
 137   RunTestIf("create-cluster", qa_node.TestNodeAddAll)
 138   if not qa_config.TestEnabled("create-cluster"):
 139     # consider the nodes are already there
 140     qa_node.MarkNodeAddedAll()
 141
 142   RunTestIf("test-jobqueue", qa_cluster.TestJobqueue)
 143
 144   # enable the watcher (unconditionally)
 145   RunTest(qa_daemon.TestResumeWatcher)
 146
 147   RunTestIf("node-list", qa_node.TestNodeList)
 148
 149   # Test listing fields
 150   RunTestIf("node-list", qa_node.TestNodeListFields)
 151   RunTestIf("instance-list", qa_instance.TestInstanceListFields)
 152   RunTestIf("job-list", qa_job.TestJobListFields)
 153   RunTestIf("instance-export", qa_instance.TestBackupListFields)
 154
 155   RunTestIf("node-info", qa_node.TestNodeInfo)
 156
 157
 158 def RunClusterTests():
 159   """Runs tests related to gnt-cluster.
 160
 161   """
 162   for test, fn in [
 163     ("create-cluster", qa_cluster.TestClusterInitDisk),
 164     ("cluster-renew-crypto", qa_cluster.TestClusterRenewCrypto),
 165     ("cluster-verify", qa_cluster.TestClusterVerify),
 166     ("cluster-reserved-lvs", qa_cluster.TestClusterReservedLvs),
 167     # TODO: add more cluster modify tests
 168     ("cluster-modify", qa_cluster.TestClusterModifyEmpty),
 169     ("cluster-modify", qa_cluster.TestClusterModifyBe),
 170     ("cluster-modify", qa_cluster.TestClusterModifyDisk),
 171     ("cluster-rename", qa_cluster.TestClusterRename),
 172     ("cluster-info", qa_cluster.TestClusterVersion),
 173     ("cluster-info", qa_cluster.TestClusterInfo),
 174     ("cluster-info", qa_cluster.TestClusterGetmaster),
 175     ("cluster-redist-conf", qa_cluster.TestClusterRedistConf),
 176     ("cluster-copyfile", qa_cluster.TestClusterCopyfile),
 177     ("cluster-command", qa_cluster.TestClusterCommand),
 178     ("cluster-burnin", qa_cluster.TestClusterBurnin),
 179     ("cluster-master-failover", qa_cluster.TestClusterMasterFailover),
 180     ("cluster-master-failover",
 181      qa_cluster.TestClusterMasterFailoverWithDrainedQueue),
 182     ("cluster-oob", qa_cluster.TestClusterOob),
 183     ("rapi", qa_rapi.TestVersion),
 184     ("rapi", qa_rapi.TestEmptyCluster),
 185     ("rapi", qa_rapi.TestRapiQuery),
 186     ]:
 187     RunTestIf(test, fn)
 188
 189
 190 def RunRepairDiskSizes():
 191   """Run the repair disk-sizes test.
 192
 193   """
 194   RunTestIf("cluster-repair-disk-sizes", qa_cluster.TestClusterRepairDiskSizes)
 195
 196
 197 def RunOsTests():
 198   """Runs all tests related to gnt-os.
 199
 200   """
 201   if qa_config.TestEnabled("rapi"):
 202     rapi_getos = qa_rapi.GetOperatingSystems
 203   else:
 204     rapi_getos = None
 205
 206   for fn in [
 207     qa_os.TestOsList,
 208     qa_os.TestOsDiagnose,
 209     ]:
 210     RunTestIf("os", fn)
 211
 212   for fn in [
 213     qa_os.TestOsValid,
 214     qa_os.TestOsInvalid,
 215     qa_os.TestOsPartiallyValid,
 216     ]:
 217     RunTestIf("os", fn, rapi_getos)
 218
 219   for fn in [
 220     qa_os.TestOsModifyValid,
 221     qa_os.TestOsModifyInvalid,
 222     qa_os.TestOsStatesNonExisting,
 223     ]:
 224     RunTestIf("os", fn)
 225
 226
 227 def RunCommonInstanceTests(instance):
 228   """Runs a few tests that are common to all disk types.
 229
 230   """
 231   RunTestIf("instance-shutdown", qa_instance.TestInstanceShutdown, instance)
 232   RunTestIf(["instance-shutdown", "instance-console", "rapi"],
 233             qa_rapi.TestRapiStoppedInstanceConsole, instance)
 234   RunTestIf(["instance-shutdown", "instance-modify"],
 235             qa_instance.TestInstanceStoppedModify, instance)
 236   RunTestIf("instance-shutdown", qa_instance.TestInstanceStartup, instance)
 237
 238   # Test shutdown/start via RAPI
 239   RunTestIf(["instance-shutdown", "rapi"],
 240             qa_rapi.TestRapiInstanceShutdown, instance)
 241   RunTestIf(["instance-shutdown", "rapi"],
 242             qa_rapi.TestRapiInstanceStartup, instance)
 243
 244   RunTestIf("instance-list", qa_instance.TestInstanceList)
 245
 246   RunTestIf("instance-info", qa_instance.TestInstanceInfo, instance)
 247
 248   RunTestIf("instance-modify", qa_instance.TestInstanceModify, instance)
 249   RunTestIf(["instance-modify", "rapi"],
 250             qa_rapi.TestRapiInstanceModify, instance)
 251
 252   RunTestIf("instance-console", qa_instance.TestInstanceConsole, instance)
 253   RunTestIf(["instance-console", "rapi"],
 254             qa_rapi.TestRapiInstanceConsole, instance)
 255
 256   DOWN_TESTS = qa_config.Either([
 257     "instance-reinstall",
 258     "instance-rename",
 259     "instance-grow-disk",
 260     ])
 261
 262   # shutdown instance for any 'down' tests
 263   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceShutdown, instance)
 264
 265   # now run the 'down' state tests
 266   RunTestIf("instance-reinstall", qa_instance.TestInstanceReinstall, instance)
 267   RunTestIf(["instance-reinstall", "rapi"],
 268             qa_rapi.TestRapiInstanceReinstall, instance)
 269
 270   if qa_config.TestEnabled("instance-rename"):
 271     rename_source = instance["name"]
 272     rename_target = qa_config.get("rename", None)
 273     # perform instance rename to the same name
 274     RunTest(qa_instance.TestInstanceRenameAndBack,
 275             rename_source, rename_source)
 276     RunTestIf("rapi", qa_rapi.TestRapiInstanceRenameAndBack,
 277               rename_source, rename_source)
 278     if rename_target is not None:
 279       # perform instance rename to a different name, if we have one configured
 280       RunTest(qa_instance.TestInstanceRenameAndBack,
 281               rename_source, rename_target)
 282       RunTestIf("rapi", qa_rapi.TestRapiInstanceRenameAndBack,
 283                 rename_source, rename_target)
 284
 285   RunTestIf(["instance-grow-disk"], qa_instance.TestInstanceGrowDisk, instance)
 286
 287   # and now start the instance again
 288   RunTestIf(DOWN_TESTS, qa_instance.TestInstanceStartup, instance)
 289
 290   RunTestIf("instance-reboot", qa_instance.TestInstanceReboot, instance)
 291
 292   RunTestIf("tags", qa_tags.TestInstanceTags, instance)
 293
 294   RunTestIf("cluster-verify", qa_cluster.TestClusterVerify)
 295
 296   RunTestIf("rapi", qa_rapi.TestInstance, instance)
 297
 298   # Lists instances, too
 299   RunTestIf("node-list", qa_node.TestNodeList)
 300
 301   # Some jobs have been run, let's test listing them
 302   RunTestIf("job-list", qa_job.TestJobList)
 303
 304
 305 def RunCommonNodeTests():
 306   """Run a few common node tests.
 307
 308   """
 309   RunTestIf("node-volumes", qa_node.TestNodeVolumes)
 310   RunTestIf("node-storage", qa_node.TestNodeStorage)
 311   RunTestIf("node-oob", qa_node.TestOutOfBand)
 312
 313
 314 def RunGroupListTests():
 315   """Run tests for listing node groups.
 316
 317   """
 318   RunTestIf("group-list", qa_group.TestGroupList)
 319   RunTestIf("group-list", qa_group.TestGroupListFields)
 320
 321
 322 def RunGroupRwTests():
 323   """Run tests for adding/removing/renaming groups.
 324
 325   """
 326   RunTestIf("group-rwops", qa_group.TestGroupAddRemoveRename)
 327   RunTestIf("group-rwops", qa_group.TestGroupAddWithOptions)
 328   RunTestIf("group-rwops", qa_group.TestGroupModify)
 329   RunTestIf(["group-rwops", "rapi"], qa_rapi.TestRapiNodeGroups)
 330   RunTestIf(["group-rwops", "tags"], qa_tags.TestGroupTags,
 331             qa_group.GetDefaultGroup())
 332
 333
 334 def RunExportImportTests(instance, pnode, snode):
 335   """Tries to export and import the instance.
 336
 337   @param pnode: current primary node of the instance
 338   @param snode: current secondary node of the instance, if any,
 339       otherwise None
 340
 341   """
 342   if qa_config.TestEnabled("instance-export"):
 343     RunTest(qa_instance.TestInstanceExportNoTarget, instance)
 344
 345     expnode = qa_config.AcquireNode(exclude=pnode)
 346     try:
 347       name = RunTest(qa_instance.TestInstanceExport, instance, expnode)
 348
 349       RunTest(qa_instance.TestBackupList, expnode)
 350
 351       if qa_config.TestEnabled("instance-import"):
 352         newinst = qa_config.AcquireInstance()
 353         try:
 354           RunTest(qa_instance.TestInstanceImport, newinst, pnode,
 355                   expnode, name)
 356           # Check if starting the instance works
 357           RunTest(qa_instance.TestInstanceStartup, newinst)
 358           RunTest(qa_instance.TestInstanceRemove, newinst)
 359         finally:
 360           qa_config.ReleaseInstance(newinst)
 361     finally:
 362       qa_config.ReleaseNode(expnode)
 363
 364   if qa_config.TestEnabled(["rapi", "inter-cluster-instance-move"]):
 365     newinst = qa_config.AcquireInstance()
 366     try:
 367       if snode is None:
 368         excl = [pnode]
 369       else:
 370         excl = [pnode, snode]
 371       tnode = qa_config.AcquireNode(exclude=excl)
 372       try:
 373         RunTest(qa_rapi.TestInterClusterInstanceMove, instance, newinst,
 374                 pnode, snode, tnode)
 375       finally:
 376         qa_config.ReleaseNode(tnode)
 377     finally:
 378       qa_config.ReleaseInstance(newinst)
 379
 380
 381 def RunDaemonTests(instance):
 382   """Test the ganeti-watcher script.
 383
 384   """
 385   RunTest(qa_daemon.TestPauseWatcher)
 386
 387   RunTestIf("instance-automatic-restart",
 388             qa_daemon.TestInstanceAutomaticRestart, instance)
 389   RunTestIf("instance-consecutive-failures",
 390             qa_daemon.TestInstanceConsecutiveFailures, instance)
 391
 392   RunTest(qa_daemon.TestResumeWatcher)
 393
 394
 395 def RunSingleHomedHardwareFailureTests(instance, pnode):
 396   """Test hardware failure recovery for single-homed instances.
 397
 398   """
 399   if qa_config.TestEnabled("instance-recreate-disks"):
 400     othernode = qa_config.AcquireNode(exclude=[pnode])
 401     try:
 402       RunTest(qa_instance.TestRecreateDisks,
 403               instance, pnode, None, [othernode])
 404     finally:
 405       qa_config.ReleaseNode(othernode)
 406
 407
 408 def RunHardwareFailureTests(instance, pnode, snode):
 409   """Test cluster internal hardware failure recovery.
 410
 411   """
 412   RunTestIf("instance-failover", qa_instance.TestInstanceFailover, instance)
 413   RunTestIf(["instance-failover", "rapi"],
 414             qa_rapi.TestRapiInstanceFailover, instance)
 415
 416   RunTestIf("instance-migrate", qa_instance.TestInstanceMigrate, instance)
 417   RunTestIf(["instance-migrate", "rapi"],
 418             qa_rapi.TestRapiInstanceMigrate, instance)
 419
 420   if qa_config.TestEnabled("instance-replace-disks"):
 421     othernode = qa_config.AcquireNode(exclude=[pnode, snode])
 422     try:
 423       RunTestIf("rapi", qa_rapi.TestRapiInstanceReplaceDisks, instance)
 424       RunTest(qa_instance.TestReplaceDisks,
 425               instance, pnode, snode, othernode)
 426     finally:
 427       qa_config.ReleaseNode(othernode)
 428
 429   if qa_config.TestEnabled("instance-recreate-disks"):
 430     othernode1 = qa_config.AcquireNode(exclude=[pnode, snode])
 431     try:
 432       othernode2 = qa_config.AcquireNode(exclude=[pnode, snode, othernode1])
 433     except qa_error.OutOfNodesError:
 434       # Let's reuse one of the nodes if the cluster is not big enough
 435       othernode2 = pnode
 436     try:
 437       RunTest(qa_instance.TestRecreateDisks,
 438               instance, pnode, snode, [othernode1, othernode2])
 439     finally:
 440       qa_config.ReleaseNode(othernode1)
 441       if othernode2 != pnode:
 442         qa_config.ReleaseNode(othernode2)
 443
 444   RunTestIf("node-evacuate", qa_node.TestNodeEvacuate, pnode, snode)
 445
 446   RunTestIf("node-failover", qa_node.TestNodeFailover, pnode, snode)
 447
 448   RunTestIf("instance-disk-failure", qa_instance.TestInstanceMasterDiskFailure,
 449             instance, pnode, snode)
 450   RunTestIf("instance-disk-failure",
 451             qa_instance.TestInstanceSecondaryDiskFailure, instance,
 452             pnode, snode)
 453
 454
 455 def RunQa():
 456   """Main QA body.
 457
 458   """
 459   rapi_user = "ganeti-qa"
 460   rapi_secret = utils.GenerateSecret()
 461
 462   RunEnvTests()
 463   SetupCluster(rapi_user, rapi_secret)
 464
 465   # Load RAPI certificate
 466   qa_rapi.Setup(rapi_user, rapi_secret)
 467
 468   RunClusterTests()
 469   RunOsTests()
 470
 471   RunTestIf("tags", qa_tags.TestClusterTags)
 472
 473   RunCommonNodeTests()
 474   RunGroupListTests()
 475   RunGroupRwTests()
 476
 477   pnode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
 478   try:
 479     RunTestIf("node-readd", qa_node.TestNodeReadd, pnode)
 480     RunTestIf("node-modify", qa_node.TestNodeModify, pnode)
 481     RunTestIf("delay", qa_cluster.TestDelay, pnode)
 482   finally:
 483     qa_config.ReleaseNode(pnode)
 484
 485   pnode = qa_config.AcquireNode()
 486   try:
 487     RunTestIf("tags", qa_tags.TestNodeTags, pnode)
 488
 489     if qa_rapi.Enabled():
 490       RunTest(qa_rapi.TestNode, pnode)
 491
 492       if qa_config.TestEnabled("instance-add-plain-disk"):
 493         for use_client in [True, False]:
 494           rapi_instance = RunTest(qa_rapi.TestRapiInstanceAdd, pnode,
 495                                   use_client)
 496           if qa_config.TestEnabled("instance-plain-rapi-common-tests"):
 497             RunCommonInstanceTests(rapi_instance)
 498           RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance, use_client)
 499           del rapi_instance
 500
 501     if qa_config.TestEnabled("instance-add-plain-disk"):
 502       instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, pnode)
 503       RunCommonInstanceTests(instance)
 504       RunGroupListTests()
 505       RunTestIf("cluster-epo", qa_cluster.TestClusterEpo)
 506       RunExportImportTests(instance, pnode, None)
 507       RunDaemonTests(instance)
 508       RunRepairDiskSizes()
 509       RunSingleHomedHardwareFailureTests(instance, pnode)
 510       RunTest(qa_instance.TestInstanceRemove, instance)
 511       del instance
 512
 513     multinode_tests = [
 514       ("instance-add-drbd-disk",
 515        qa_instance.TestInstanceAddWithDrbdDisk),
 516     ]
 517
 518     for name, func in multinode_tests:
 519       if qa_config.TestEnabled(name):
 520         snode = qa_config.AcquireNode(exclude=pnode)
 521         try:
 522           instance = RunTest(func, pnode, snode)
 523           RunTestIf("haskell-confd", qa_node.TestNodeListDrbd, pnode)
 524           RunTestIf("haskell-confd", qa_node.TestNodeListDrbd, snode)
 525           RunCommonInstanceTests(instance)
 526           RunGroupListTests()
 527           RunTest(qa_group.TestAssignNodesIncludingSplit,
 528                   constants.INITIAL_NODE_GROUP_NAME,
 529                   pnode["primary"], snode["primary"])
 530           if qa_config.TestEnabled("instance-convert-disk"):
 531             RunTest(qa_instance.TestInstanceShutdown, instance)
 532             RunTest(qa_instance.TestInstanceConvertDisk, instance, snode)
 533             RunTest(qa_instance.TestInstanceStartup, instance)
 534           RunExportImportTests(instance, pnode, snode)
 535           RunHardwareFailureTests(instance, pnode, snode)
 536           RunRepairDiskSizes()
 537           RunTest(qa_instance.TestInstanceRemove, instance)
 538           del instance
 539         finally:
 540           qa_config.ReleaseNode(snode)
 541
 542     # Test removing instance with offline drbd secondary
 543     if qa_config.TestEnabled("instance-remove-drbd-offline"):
 544       snode = qa_config.AcquireNode(exclude=pnode)
 545       instance = \
 546         qa_instance.TestInstanceAddWithDrbdDisk(pnode, snode)
 547       try:
 548         qa_node.MakeNodeOffline(snode, "yes")
 549         RunTest(qa_instance.TestInstanceRemove, instance)
 550       finally:
 551         qa_node.MakeNodeOffline(snode, "no")
 552         qa_config.ReleaseNode(snode)
 553
 554     if qa_config.TestEnabled(["instance-add-plain-disk", "instance-export"]):
 555       for shutdown in [False, True]:
 556         instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, pnode)
 557         expnode = qa_config.AcquireNode(exclude=pnode)
 558         try:
 559           if shutdown:
 560             # Stop instance before exporting and removing it
 561             RunTest(qa_instance.TestInstanceShutdown, instance)
 562           RunTest(qa_instance.TestInstanceExportWithRemove, instance, expnode)
 563           RunTest(qa_instance.TestBackupList, expnode)
 564         finally:
 565           qa_config.ReleaseNode(expnode)
 566         del expnode
 567         del instance
 568
 569   finally:
 570     qa_config.ReleaseNode(pnode)
 571
 572   RunTestIf("create-cluster", qa_node.TestNodeRemoveAll)
 573
 574   RunTestIf("cluster-destroy", qa_cluster.TestClusterDestroy)
 575
 576
 577 @rapi.client.UsesRapiClient
 578 def main():
 579   """Main program.
 580
 581   """
 582   parser = optparse.OptionParser(usage="%prog [options] <config-file>")
 583   parser.add_option("--yes-do-it", dest="yes_do_it",
 584                     action="store_true",
 585                     help="Really execute the tests")
 586   (qa_config.options, args) = parser.parse_args()
 587
 588   if len(args) == 1:
 589     (config_file, ) = args
 590   else:
 591     parser.error("Wrong number of arguments.")
 592
 593   if not qa_config.options.yes_do_it:
 594     print ("Executing this script irreversibly destroys any Ganeti\n"
 595            "configuration on all nodes involved. If you really want\n"
 596            "to start testing, supply the --yes-do-it option.")
 597     sys.exit(1)
 598
 599   qa_config.Load(config_file)
 600
 601   primary = qa_config.GetMasterNode()["primary"]
 602   qa_utils.StartMultiplexer(primary)
 603   print ("SSH command for primary node: %s" %
 604          utils.ShellQuoteArgs(qa_utils.GetSSHCommand(primary, "")))
 605   print ("SSH command for other nodes: %s" %
 606          utils.ShellQuoteArgs(qa_utils.GetSSHCommand("NODE", "")))
 607   try:
 608     RunQa()
 609   finally:
 610     qa_utils.CloseMultiplexers()
 611
 612 if __name__ == "__main__":
 613   main()