#!/usr/bin/python -u
#
-# Copyright (C) 2007, 2008, 2009, 2010, 2011 Google Inc.
+# Copyright (C) 2007, 2008, 2009, 2010, 2011, 2012, 2013 Google Inc.
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
import qa_config
import qa_daemon
import qa_env
+import qa_error
import qa_group
import qa_instance
+import qa_network
import qa_node
import qa_os
+import qa_job
import qa_rapi
import qa_tags
import qa_utils
from ganeti import utils
-from ganeti import rapi
+from ganeti import rapi # pylint: disable=W0611
from ganeti import constants
import ganeti.rapi.client # pylint: disable=W0611
+from ganeti.rapi.client import UsesRapiClient
def _FormatHeader(line, end=72):
return desc.rstrip(".")
-def RunTest(fn, *args):
+def RunTest(fn, *args, **kwargs):
"""Runs a test after printing a header.
"""
print _FormatHeader("%s start %s" % (tstart, desc))
try:
- retval = fn(*args)
+ retval = fn(*args, **kwargs)
return retval
finally:
tstop = datetime.datetime.now()
print _FormatHeader("%s time=%s %s" % (tstop, tdelta, desc))
-def RunTestIf(testnames, fn, *args):
+def RunTestIf(testnames, fn, *args, **kwargs):
"""Runs a test conditionally.
@param testnames: either a single test name in the configuration
"""
if qa_config.TestEnabled(testnames):
- RunTest(fn, *args)
+ RunTest(fn, *args, **kwargs)
else:
tstart = datetime.datetime.now()
desc = _DescriptionOf(fn)
+ # TODO: Formatting test names when non-string names are involved
print _FormatHeader("%s skipping %s, test(s) %s disabled" %
(tstart, desc, testnames))
"""
RunTestIf("create-cluster", qa_cluster.TestClusterInit,
rapi_user, rapi_secret)
+ if not qa_config.TestEnabled("create-cluster"):
+ # If the cluster is already in place, we assume that exclusive-storage is
+ # already set according to the configuration
+ qa_config.SetExclusiveStorage(qa_config.get("exclusive-storage", False))
# Test on empty cluster
RunTestIf("node-list", qa_node.TestNodeList)
RunTestIf("instance-list", qa_instance.TestInstanceList)
+ RunTestIf("job-list", qa_job.TestJobList)
RunTestIf("create-cluster", qa_node.TestNodeAddAll)
if not qa_config.TestEnabled("create-cluster"):
# Test listing fields
RunTestIf("node-list", qa_node.TestNodeListFields)
RunTestIf("instance-list", qa_instance.TestInstanceListFields)
+ RunTestIf("job-list", qa_job.TestJobListFields)
+ RunTestIf("instance-export", qa_instance.TestBackupListFields)
RunTestIf("node-info", qa_node.TestNodeInfo)
("cluster-verify", qa_cluster.TestClusterVerify),
("cluster-reserved-lvs", qa_cluster.TestClusterReservedLvs),
# TODO: add more cluster modify tests
+ ("cluster-modify", qa_cluster.TestClusterModifyEmpty),
("cluster-modify", qa_cluster.TestClusterModifyBe),
("cluster-modify", qa_cluster.TestClusterModifyDisk),
("cluster-rename", qa_cluster.TestClusterRename),
RunTestIf("instance-shutdown", qa_instance.TestInstanceShutdown, instance)
RunTestIf(["instance-shutdown", "instance-console", "rapi"],
qa_rapi.TestRapiStoppedInstanceConsole, instance)
+ RunTestIf(["instance-shutdown", "instance-modify"],
+ qa_instance.TestInstanceStoppedModify, instance)
RunTestIf("instance-shutdown", qa_instance.TestInstanceStartup, instance)
# Test shutdown/start via RAPI
RunTestIf(["instance-console", "rapi"],
qa_rapi.TestRapiInstanceConsole, instance)
- RunTestIf("instance-reinstall", qa_instance.TestInstanceShutdown, instance)
+ DOWN_TESTS = qa_config.Either([
+ "instance-reinstall",
+ "instance-rename",
+ "instance-grow-disk",
+ ])
+
+ # shutdown instance for any 'down' tests
+ RunTestIf(DOWN_TESTS, qa_instance.TestInstanceShutdown, instance)
+
+ # now run the 'down' state tests
RunTestIf("instance-reinstall", qa_instance.TestInstanceReinstall, instance)
RunTestIf(["instance-reinstall", "rapi"],
qa_rapi.TestRapiInstanceReinstall, instance)
- RunTestIf("instance-reinstall", qa_instance.TestInstanceStartup, instance)
-
- RunTestIf("instance-reboot", qa_instance.TestInstanceReboot, instance)
if qa_config.TestEnabled("instance-rename"):
- rename_source = instance["name"]
- rename_target = qa_config.get("rename", None)
- RunTest(qa_instance.TestInstanceShutdown, instance)
- # perform instance rename to the same name
- RunTest(qa_instance.TestInstanceRename, rename_source, rename_source)
- RunTestIf("rapi", qa_rapi.TestRapiInstanceRename,
+ tgt_instance = qa_config.AcquireInstance()
+ try:
+ rename_source = instance.name
+ rename_target = tgt_instance.name
+ # perform instance rename to the same name
+ RunTest(qa_instance.TestInstanceRenameAndBack,
rename_source, rename_source)
- if rename_target is not None:
- # perform instance rename to a different name, if we have one configured
- RunTest(qa_instance.TestInstanceRename, rename_source, rename_target)
- RunTest(qa_instance.TestInstanceRename, rename_target, rename_source)
- RunTestIf("rapi", qa_rapi.TestRapiInstanceRename,
+ RunTestIf("rapi", qa_rapi.TestRapiInstanceRenameAndBack,
+ rename_source, rename_source)
+ if rename_target is not None:
+ # perform instance rename to a different name, if we have one configured
+ RunTest(qa_instance.TestInstanceRenameAndBack,
rename_source, rename_target)
- RunTestIf("rapi", qa_rapi.TestRapiInstanceRename,
- rename_target, rename_source)
- RunTest(qa_instance.TestInstanceStartup, instance)
+ RunTestIf("rapi", qa_rapi.TestRapiInstanceRenameAndBack,
+ rename_source, rename_target)
+ finally:
+ tgt_instance.Release()
+
+ RunTestIf(["instance-grow-disk"], qa_instance.TestInstanceGrowDisk, instance)
+
+ # and now start the instance again
+ RunTestIf(DOWN_TESTS, qa_instance.TestInstanceStartup, instance)
+
+ RunTestIf("instance-reboot", qa_instance.TestInstanceReboot, instance)
RunTestIf("tags", qa_tags.TestInstanceTags, instance)
# Lists instances, too
RunTestIf("node-list", qa_node.TestNodeList)
+ # Some jobs have been run, let's test listing them
+ RunTestIf("job-list", qa_job.TestJobList)
+
def RunCommonNodeTests():
"""Run a few common node tests.
RunTestIf("group-list", qa_group.TestGroupListFields)
+def RunNetworkTests():
+ """Run tests for network management.
+
+ """
+ RunTestIf("network", qa_network.TestNetworkAddRemove)
+ RunTestIf("network", qa_network.TestNetworkConnect)
+
+
def RunGroupRwTests():
"""Run tests for adding/removing/renaming groups.
qa_group.GetDefaultGroup())
-def RunExportImportTests(instance, pnode, snode):
+def RunExportImportTests(instance, inodes):
"""Tries to export and import the instance.
- @param pnode: current primary node of the instance
- @param snode: current secondary node of the instance, if any,
- otherwise None
+ @type inodes: list of nodes
+ @param inodes: current nodes of the instance
"""
if qa_config.TestEnabled("instance-export"):
RunTest(qa_instance.TestInstanceExportNoTarget, instance)
+ pnode = inodes[0]
expnode = qa_config.AcquireNode(exclude=pnode)
try:
name = RunTest(qa_instance.TestInstanceExport, instance, expnode)
if qa_config.TestEnabled("instance-import"):
newinst = qa_config.AcquireInstance()
try:
- RunTest(qa_instance.TestInstanceImport, pnode, newinst,
+ RunTest(qa_instance.TestInstanceImport, newinst, pnode,
expnode, name)
+ # Check if starting the instance works
+ RunTest(qa_instance.TestInstanceStartup, newinst)
RunTest(qa_instance.TestInstanceRemove, newinst)
finally:
- qa_config.ReleaseInstance(newinst)
+ newinst.Release()
finally:
- qa_config.ReleaseNode(expnode)
+ expnode.Release()
if qa_config.TestEnabled(["rapi", "inter-cluster-instance-move"]):
newinst = qa_config.AcquireInstance()
try:
- if snode is None:
- excl = [pnode]
- else:
- excl = [pnode, snode]
- tnode = qa_config.AcquireNode(exclude=excl)
+ tnode = qa_config.AcquireNode(exclude=inodes)
try:
RunTest(qa_rapi.TestInterClusterInstanceMove, instance, newinst,
- pnode, snode, tnode)
+ inodes, tnode)
finally:
- qa_config.ReleaseNode(tnode)
+ tnode.Release()
finally:
- qa_config.ReleaseInstance(newinst)
+ newinst.Release()
def RunDaemonTests(instance):
RunTest(qa_daemon.TestResumeWatcher)
-def RunHardwareFailureTests(instance, pnode, snode):
+def RunHardwareFailureTests(instance, inodes):
"""Test cluster internal hardware failure recovery.
"""
qa_rapi.TestRapiInstanceMigrate, instance)
if qa_config.TestEnabled("instance-replace-disks"):
- othernode = qa_config.AcquireNode(exclude=[pnode, snode])
+ # We just need alternative secondary nodes, hence "- 1"
+ othernodes = qa_config.AcquireManyNodes(len(inodes) - 1, exclude=inodes)
try:
RunTestIf("rapi", qa_rapi.TestRapiInstanceReplaceDisks, instance)
RunTest(qa_instance.TestReplaceDisks,
- instance, pnode, snode, othernode)
+ instance, inodes, othernodes)
+ finally:
+ qa_config.ReleaseManyNodes(othernodes)
+ del othernodes
+
+ if qa_config.TestEnabled("instance-recreate-disks"):
+ try:
+ acquirednodes = qa_config.AcquireManyNodes(len(inodes), exclude=inodes)
+ othernodes = acquirednodes
+ except qa_error.OutOfNodesError:
+ if len(inodes) > 1:
+ # If the cluster is not big enough, let's reuse some of the nodes, but
+ # with different roles. In this way, we can test a DRBD instance even on
+ # a 3-node cluster.
+ acquirednodes = [qa_config.AcquireNode(exclude=inodes)]
+ othernodes = acquirednodes + inodes[:-1]
+ else:
+ raise
+ try:
+ RunTest(qa_instance.TestRecreateDisks,
+ instance, inodes, othernodes)
finally:
- qa_config.ReleaseNode(othernode)
+ qa_config.ReleaseManyNodes(acquirednodes)
+
+ if len(inodes) >= 2:
+ RunTestIf("node-evacuate", qa_node.TestNodeEvacuate, inodes[0], inodes[1])
+ RunTestIf("node-failover", qa_node.TestNodeFailover, inodes[0], inodes[1])
+
+
+def RunExclusiveStorageTests():
+ """Test exclusive storage."""
+ if not qa_config.TestEnabled("cluster-exclusive-storage"):
+ return
+
+ node = qa_config.AcquireNode()
+ try:
+ old_es = qa_cluster.TestSetExclStorCluster(False)
+ qa_node.TestExclStorSingleNode(node)
- RunTestIf("node-evacuate", qa_node.TestNodeEvacuate, pnode, snode)
+ qa_cluster.TestSetExclStorCluster(True)
+ qa_cluster.TestExclStorSharedPv(node)
- RunTestIf("node-failover", qa_node.TestNodeFailover, pnode, snode)
+ if qa_config.TestEnabled("instance-add-plain-disk"):
+ # Make sure that the cluster doesn't have any pre-existing problem
+ qa_cluster.AssertClusterVerify()
- RunTestIf("instance-disk-failure", qa_instance.TestInstanceMasterDiskFailure,
- instance, pnode, snode)
- RunTestIf("instance-disk-failure",
- qa_instance.TestInstanceSecondaryDiskFailure, instance,
- pnode, snode)
+ # Create and allocate instances
+ instance1 = qa_instance.TestInstanceAddWithPlainDisk([node])
+ try:
+ instance2 = qa_instance.TestInstanceAddWithPlainDisk([node])
+ try:
+ # cluster-verify checks that disks are allocated correctly
+ qa_cluster.AssertClusterVerify()
+
+ # Remove instances
+ qa_instance.TestInstanceRemove(instance2)
+ qa_instance.TestInstanceRemove(instance1)
+ finally:
+ instance2.Release()
+ finally:
+ instance1.Release()
+
+ if qa_config.TestEnabled("instance-add-drbd-disk"):
+ snode = qa_config.AcquireNode()
+ try:
+ qa_cluster.TestSetExclStorCluster(False)
+ instance = qa_instance.TestInstanceAddWithDrbdDisk([node, snode])
+ try:
+ qa_cluster.TestSetExclStorCluster(True)
+ exp_err = [constants.CV_EINSTANCEUNSUITABLENODE]
+ qa_cluster.AssertClusterVerify(fail=True, errors=exp_err)
+ qa_instance.TestInstanceRemove(instance)
+ finally:
+ instance.Release()
+ finally:
+ snode.Release()
+ qa_cluster.TestSetExclStorCluster(old_es)
+ finally:
+ node.Release()
+
+
+def RunInstanceTests():
+ """Create and exercise instances."""
+ instance_tests = [
+ ("instance-add-plain-disk", constants.DT_PLAIN,
+ qa_instance.TestInstanceAddWithPlainDisk, 1),
+ ("instance-add-drbd-disk", constants.DT_DRBD8,
+ qa_instance.TestInstanceAddWithDrbdDisk, 2),
+ ]
+
+ for (test_name, templ, create_fun, num_nodes) in instance_tests:
+ if (qa_config.TestEnabled(test_name) and
+ qa_config.IsTemplateSupported(templ)):
+ inodes = qa_config.AcquireManyNodes(num_nodes)
+ try:
+ instance = RunTest(create_fun, inodes)
+ try:
+ RunTestIf("cluster-epo", qa_cluster.TestClusterEpo)
+ RunDaemonTests(instance)
+ for node in inodes:
+ RunTestIf("haskell-confd", qa_node.TestNodeListDrbd, node)
+ if len(inodes) > 1:
+ RunTestIf("group-rwops", qa_group.TestAssignNodesIncludingSplit,
+ constants.INITIAL_NODE_GROUP_NAME,
+ inodes[0].primary, inodes[1].primary)
+ if qa_config.TestEnabled("instance-convert-disk"):
+ RunTest(qa_instance.TestInstanceShutdown, instance)
+ RunTest(qa_instance.TestInstanceConvertDiskToPlain,
+ instance, inodes)
+ RunTest(qa_instance.TestInstanceStartup, instance)
+ RunCommonInstanceTests(instance)
+ RunGroupListTests()
+ RunExportImportTests(instance, inodes)
+ RunHardwareFailureTests(instance, inodes)
+ RunRepairDiskSizes()
+ RunTest(qa_instance.TestInstanceRemove, instance)
+ finally:
+ instance.Release()
+ del instance
+ finally:
+ qa_config.ReleaseManyNodes(inodes)
+ qa_cluster.AssertClusterVerify()
def RunQa():
RunCommonNodeTests()
RunGroupListTests()
RunGroupRwTests()
+ RunNetworkTests()
+ # The master shouldn't be readded or put offline; "delay" needs a non-master
+ # node to test
pnode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
try:
RunTestIf("node-readd", qa_node.TestNodeReadd, pnode)
RunTestIf("node-modify", qa_node.TestNodeModify, pnode)
RunTestIf("delay", qa_cluster.TestDelay, pnode)
finally:
- qa_config.ReleaseNode(pnode)
+ pnode.Release()
+
+ # Make sure the cluster is clean before running instance tests
+ qa_cluster.AssertClusterVerify()
pnode = qa_config.AcquireNode()
try:
for use_client in [True, False]:
rapi_instance = RunTest(qa_rapi.TestRapiInstanceAdd, pnode,
use_client)
- RunCommonInstanceTests(rapi_instance)
- RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance, use_client)
+ try:
+ if qa_config.TestEnabled("instance-plain-rapi-common-tests"):
+ RunCommonInstanceTests(rapi_instance)
+ RunTest(qa_rapi.TestRapiInstanceRemove, rapi_instance, use_client)
+ finally:
+ rapi_instance.Release()
del rapi_instance
- if qa_config.TestEnabled("instance-add-plain-disk"):
- instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, pnode)
- RunCommonInstanceTests(instance)
- RunGroupListTests()
- RunTestIf("cluster-epo", qa_cluster.TestClusterEpo)
- RunExportImportTests(instance, pnode, None)
- RunDaemonTests(instance)
- RunRepairDiskSizes()
- RunTest(qa_instance.TestInstanceRemove, instance)
- del instance
-
- multinode_tests = [
- ("instance-add-drbd-disk",
- qa_instance.TestInstanceAddWithDrbdDisk),
- ]
-
- for name, func in multinode_tests:
- if qa_config.TestEnabled(name):
- snode = qa_config.AcquireNode(exclude=pnode)
- try:
- instance = RunTest(func, pnode, snode)
- RunCommonInstanceTests(instance)
- RunGroupListTests()
- RunTest(qa_group.TestAssignNodesIncludingSplit,
- constants.INITIAL_NODE_GROUP_NAME,
- pnode["primary"], snode["primary"])
- if qa_config.TestEnabled("instance-convert-disk"):
- RunTest(qa_instance.TestInstanceShutdown, instance)
- RunTest(qa_instance.TestInstanceConvertDisk, instance, snode)
- RunTest(qa_instance.TestInstanceStartup, instance)
- RunExportImportTests(instance, pnode, snode)
- RunHardwareFailureTests(instance, pnode, snode)
- RunRepairDiskSizes()
- RunTest(qa_instance.TestInstanceRemove, instance)
- del instance
- finally:
- qa_config.ReleaseNode(snode)
+ finally:
+ pnode.Release()
+
+ config_list = [
+ ("default-instance-tests", lambda: None, lambda _: None),
+ ("exclusive-storage-instance-tests",
+ lambda: qa_cluster.TestSetExclStorCluster(True),
+ qa_cluster.TestSetExclStorCluster),
+ ]
+ for (conf_name, setup_conf_f, restore_conf_f) in config_list:
+ if qa_config.TestEnabled(conf_name):
+ oldconf = setup_conf_f()
+ RunInstanceTests()
+ restore_conf_f(oldconf)
+ pnode = qa_config.AcquireNode()
+ try:
if qa_config.TestEnabled(["instance-add-plain-disk", "instance-export"]):
for shutdown in [False, True]:
- instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, pnode)
- expnode = qa_config.AcquireNode(exclude=pnode)
+ instance = RunTest(qa_instance.TestInstanceAddWithPlainDisk, [pnode])
try:
- if shutdown:
- # Stop instance before exporting and removing it
- RunTest(qa_instance.TestInstanceShutdown, instance)
- RunTest(qa_instance.TestInstanceExportWithRemove, instance, expnode)
- RunTest(qa_instance.TestBackupList, expnode)
+ expnode = qa_config.AcquireNode(exclude=pnode)
+ try:
+ if shutdown:
+ # Stop instance before exporting and removing it
+ RunTest(qa_instance.TestInstanceShutdown, instance)
+ RunTest(qa_instance.TestInstanceExportWithRemove, instance, expnode)
+ RunTest(qa_instance.TestBackupList, expnode)
+ finally:
+ expnode.Release()
finally:
- qa_config.ReleaseNode(expnode)
+ instance.Release()
del expnode
del instance
+ qa_cluster.AssertClusterVerify()
finally:
- qa_config.ReleaseNode(pnode)
+ pnode.Release()
+
+ RunExclusiveStorageTests()
+
+ # Test removing instance with offline drbd secondary
+ if qa_config.TestEnabled("instance-remove-drbd-offline"):
+ # Make sure the master is not put offline
+ snode = qa_config.AcquireNode(exclude=qa_config.GetMasterNode())
+ try:
+ pnode = qa_config.AcquireNode(exclude=snode)
+ try:
+ instance = qa_instance.TestInstanceAddWithDrbdDisk([pnode, snode])
+ set_offline = lambda node: qa_node.MakeNodeOffline(node, "yes")
+ set_online = lambda node: qa_node.MakeNodeOffline(node, "no")
+ RunTest(qa_instance.TestRemoveInstanceOfflineNode, instance, snode,
+ set_offline, set_online)
+ finally:
+ pnode.Release()
+ finally:
+ snode.Release()
+ qa_cluster.AssertClusterVerify()
RunTestIf("create-cluster", qa_node.TestNodeRemoveAll)
RunTestIf("cluster-destroy", qa_cluster.TestClusterDestroy)
-@rapi.client.UsesRapiClient
+@UsesRapiClient
def main():
"""Main program.
"""
parser = optparse.OptionParser(usage="%prog [options] <config-file>")
parser.add_option("--yes-do-it", dest="yes_do_it",
- action="store_true",
- help="Really execute the tests")
- (qa_config.options, args) = parser.parse_args()
+ action="store_true",
+ help="Really execute the tests")
+ (opts, args) = parser.parse_args()
if len(args) == 1:
(config_file, ) = args
else:
parser.error("Wrong number of arguments.")
- if not qa_config.options.yes_do_it:
+ if not opts.yes_do_it:
print ("Executing this script irreversibly destroys any Ganeti\n"
"configuration on all nodes involved. If you really want\n"
"to start testing, supply the --yes-do-it option.")
qa_config.Load(config_file)
- qa_utils.StartMultiplexer(qa_config.GetMasterNode()["primary"])
+ primary = qa_config.GetMasterNode().primary
+ qa_utils.StartMultiplexer(primary)
+ print ("SSH command for primary node: %s" %
+ utils.ShellQuoteArgs(qa_utils.GetSSHCommand(primary, "")))
+ print ("SSH command for other nodes: %s" %
+ utils.ShellQuoteArgs(qa_utils.GetSSHCommand("NODE", "")))
try:
RunQa()
finally: