X-Git-Url: https://code.grnet.gr/git/ganeti-local/blobdiff_plain/af2ae1c02d63e6bf96981eaf04d3ae54e4eac576..6d96ede4d916fc6aad4c61f0dc382693e5f0e00e:/qa/qa_cluster.py

diff --git a/qa/qa_cluster.py b/qa/qa_cluster.py
index 202d382..ecdf369 100644
--- a/qa/qa_cluster.py
+++ b/qa/qa_cluster.py
@@ -1,7 +1,7 @@
 #
 #
 
-# Copyright (C) 2007 Google Inc.
+# Copyright (C) 2007, 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
@@ -23,26 +23,37 @@
 
 """
 
+import re
 import tempfile
+import os.path
 
 from ganeti import constants
+from ganeti import compat
 from ganeti import utils
+from ganeti import pathutils
 
 import qa_config
-import qa_utils
 import qa_error
+import qa_instance
+import qa_logging
+import qa_utils
+
+from qa_utils import AssertEqual, AssertCommand, GetCommandOutput
+
+
+# Prefix for LVM volumes created by QA code during tests
+_QA_LV_PREFIX = "qa-"
 
-from qa_utils import AssertEqual, AssertNotEqual, StartSSH
+#: cluster verify command
+_CLUSTER_VERIFY = ["gnt-cluster", "verify"]
 
 
 def _RemoveFileFromAllNodes(filename):
   """Removes a file from all nodes.
 
   """
-  for node in qa_config.get('nodes'):
-    cmd = ['rm', '-f', filename]
-    AssertEqual(StartSSH(node['primary'],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+  for node in qa_config.get("nodes"):
+    AssertCommand(["rm", "-f", filename], node=node)
 
 
 def _CheckFileOnAllNodes(filename, content):
@@ -50,106 +61,651 @@ def _CheckFileOnAllNodes(filename, content):
 
   """
   cmd = utils.ShellQuoteArgs(["cat", filename])
-  for node in qa_config.get('nodes'):
-    AssertEqual(qa_utils.GetCommandOutput(node['primary'], cmd),
-                content)
+  for node in qa_config.get("nodes"):
+    AssertEqual(qa_utils.GetCommandOutput(node.primary, cmd), content)
+
+
+def _GetClusterField(field_path):
+  """Get the value of a cluster field.
+
+  @type field_path: list of strings
+  @param field_path: Names of the groups/fields to navigate to get the desired
+      value, e.g. C{["Default node parameters", "oob_program"]}
+  @return: The effective value of the field (the actual type depends on the
+      chosen field)
+
+  """
+  assert isinstance(field_path, list)
+  assert field_path
+  ret = qa_utils.GetObjectInfo(["gnt-cluster", "info"])
+  for key in field_path:
+    ret = ret[key]
+  return ret
+
+
+# Cluster-verify errors (date, "ERROR", then error code)
+_CVERROR_RE = re.compile(r"^[\w\s:]+\s+- (ERROR|WARNING):([A-Z0-9_-]+):")
+
+
+def _GetCVErrorCodes(cvout):
+  errs = set()
+  warns = set()
+  for l in cvout.splitlines():
+    m = _CVERROR_RE.match(l)
+    if m:
+      etype = m.group(1)
+      ecode = m.group(2)
+      if etype == "ERROR":
+        errs.add(ecode)
+      elif etype == "WARNING":
+        warns.add(ecode)
+  return (errs, warns)
+
+
+def _CheckVerifyErrors(actual, expected, etype):
+  exp_codes = compat.UniqueFrozenset(e for (_, e, _) in expected)
+  if not actual.issuperset(exp_codes):
+    missing = exp_codes.difference(actual)
+    raise qa_error.Error("Cluster-verify didn't return these expected"
+                         " %ss: %s" % (etype, utils.CommaJoin(missing)))
+
+
+def AssertClusterVerify(fail=False, errors=None, warnings=None):
+  """Run cluster-verify and check the result
+
+  @type fail: bool
+  @param fail: if cluster-verify is expected to fail instead of succeeding
+  @type errors: list of tuples
+  @param errors: List of CV_XXX errors that are expected; if specified, all the
+      errors listed must appear in cluster-verify output. A non-empty value
+      implies C{fail=True}.
+  @type warnings: list of tuples
+  @param warnings: Same as C{errors} but for warnings.
+
+  """
+  cvcmd = "gnt-cluster verify"
+  mnode = qa_config.GetMasterNode()
+  if errors or warnings:
+    cvout = GetCommandOutput(mnode.primary, cvcmd + " --error-codes",
+                             fail=(fail or errors))
+    (act_errs, act_warns) = _GetCVErrorCodes(cvout)
+    if errors:
+      _CheckVerifyErrors(act_errs, errors, "error")
+    if warnings:
+      _CheckVerifyErrors(act_warns, warnings, "warning")
+  else:
+    AssertCommand(cvcmd, fail=fail, node=mnode)
+
+
+# data for testing failures due to bad keys/values for disk parameters
+_FAIL_PARAMS = ["nonexistent:resync-rate=1",
+                "drbd:nonexistent=1",
+                "drbd:resync-rate=invalid",
+                ]
+
+
+def TestClusterInitDisk():
+  """gnt-cluster init -D"""
+  name = qa_config.get("name")
+  for param in _FAIL_PARAMS:
+    AssertCommand(["gnt-cluster", "init", "-D", param, name], fail=True)
 
 
 def TestClusterInit(rapi_user, rapi_secret):
   """gnt-cluster init"""
   master = qa_config.GetMasterNode()
 
+  rapi_users_path = qa_utils.MakeNodePath(master, pathutils.RAPI_USERS_FILE)
+  rapi_dir = os.path.dirname(rapi_users_path)
+
   # First create the RAPI credentials
-  cred_string = "%s %s write" % (rapi_user, rapi_secret)
-  cmd = ("echo %s > %s" %
-         (utils.ShellQuote(cred_string),
-          utils.ShellQuote(constants.RAPI_USERS_FILE)))
-  AssertEqual(StartSSH(master['primary'], cmd).wait(), 0)
+  fh = tempfile.NamedTemporaryFile()
+  try:
+    fh.write("%s %s write\n" % (rapi_user, rapi_secret))
+    fh.flush()
 
-  cmd = ['gnt-cluster', 'init']
+    tmpru = qa_utils.UploadFile(master.primary, fh.name)
+    try:
+      AssertCommand(["mkdir", "-p", rapi_dir])
+      AssertCommand(["mv", tmpru, rapi_users_path])
+    finally:
+      AssertCommand(["rm", "-f", tmpru])
+  finally:
+    fh.close()
+
+  # Initialize cluster
+  cmd = [
+    "gnt-cluster", "init",
+    "--primary-ip-version=%d" % qa_config.get("primary_ip_version", 4),
+    "--enabled-hypervisors=%s" % ",".join(qa_config.GetEnabledHypervisors()),
+    "--enabled-disk-templates=%s" %
+      ",".join(qa_config.GetEnabledDiskTemplates())
+    ]
+
+  for spec_type in ("mem-size", "disk-size", "disk-count", "cpu-count",
+                    "nic-count"):
+    for spec_val in ("min", "max", "std"):
+      spec = qa_config.get("ispec_%s_%s" %
+                           (spec_type.replace("-", "_"), spec_val), None)
+      if spec is not None:
+        cmd.append("--specs-%s=%s=%d" % (spec_type, spec_val, spec))
+
+  if master.secondary:
+    cmd.append("--secondary-ip=%s" % master.secondary)
+
+  vgname = qa_config.get("vg-name", None)
+  if vgname:
+    cmd.append("--vg-name=%s" % vgname)
+
+  master_netdev = qa_config.get("master-netdev", None)
+  if master_netdev:
+    cmd.append("--master-netdev=%s" % master_netdev)
+
+  nicparams = qa_config.get("default-nicparams", None)
+  if nicparams:
+    cmd.append("--nic-parameters=%s" %
+               ",".join(utils.FormatKeyValue(nicparams)))
+
+  # Cluster value of the exclusive-storage node parameter
+  e_s = qa_config.get("exclusive-storage")
+  if e_s is not None:
+    cmd.extend(["--node-parameters", "exclusive_storage=%s" % e_s])
+  else:
+    e_s = False
+  qa_config.SetExclusiveStorage(e_s)
 
-  if master.get('secondary', None):
-    cmd.append('--secondary-ip=%s' % master['secondary'])
+  extra_args = qa_config.get("cluster-init-args")
+  if extra_args:
+    cmd.extend(extra_args)
 
-  bridge = qa_config.get('bridge', None)
-  if bridge:
-    cmd.append('--bridge=%s' % bridge)
-    cmd.append('--master-netdev=%s' % bridge)
+  cmd.append(qa_config.get("name"))
 
-  htype = qa_config.get('enabled-hypervisors', None)
-  if htype:
-    cmd.append('--enabled-hypervisors=%s' % htype)
+  AssertCommand(cmd)
 
-  cmd.append(qa_config.get('name'))
+  cmd = ["gnt-cluster", "modify"]
 
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+  # hypervisor parameter modifications
+  hvp = qa_config.get("hypervisor-parameters", {})
+  for k, v in hvp.items():
+    cmd.extend(["-H", "%s:%s" % (k, v)])
+  # backend parameter modifications
+  bep = qa_config.get("backend-parameters", "")
+  if bep:
+    cmd.extend(["-B", bep])
+
+  if len(cmd) > 2:
+    AssertCommand(cmd)
+
+  # OS parameters
+  osp = qa_config.get("os-parameters", {})
+  for k, v in osp.items():
+    AssertCommand(["gnt-os", "modify", "-O", v, k])
+
+  # OS hypervisor parameters
+  os_hvp = qa_config.get("os-hvp", {})
+  for os_name in os_hvp:
+    for hv, hvp in os_hvp[os_name].items():
+      AssertCommand(["gnt-os", "modify", "-H", "%s:%s" % (hv, hvp), os_name])
 
 
 def TestClusterRename():
   """gnt-cluster rename"""
-  master = qa_config.GetMasterNode()
+  cmd = ["gnt-cluster", "rename", "-f"]
 
-  cmd = ['gnt-cluster', 'rename', '-f']
-
-  original_name = qa_config.get('name')
-  rename_target = qa_config.get('rename', None)
+  original_name = qa_config.get("name")
+  rename_target = qa_config.get("rename", None)
   if rename_target is None:
-    print qa_utils.FormatError('"rename" entry is missing')
+    print qa_logging.FormatError('"rename" entry is missing')
     return
 
-  cmd_1 = cmd + [rename_target]
-  cmd_2 = cmd + [original_name]
+  for data in [
+    cmd + [rename_target],
+    _CLUSTER_VERIFY,
+    cmd + [original_name],
+    _CLUSTER_VERIFY,
+    ]:
+    AssertCommand(data)
+
 
-  cmd_verify = ['gnt-cluster', 'verify']
+def TestClusterOob():
+  """out-of-band framework"""
+  oob_path_exists = "/tmp/ganeti-qa-oob-does-exist-%s" % utils.NewUUID()
 
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd_1)).wait(), 0)
+  AssertCommand(_CLUSTER_VERIFY)
+  AssertCommand(["gnt-cluster", "modify", "--node-parameters",
+                 "oob_program=/tmp/ganeti-qa-oob-does-not-exist-%s" %
+                 utils.NewUUID()])
 
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd_verify)).wait(), 0)
+  AssertCommand(_CLUSTER_VERIFY, fail=True)
 
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd_2)).wait(), 0)
+  AssertCommand(["touch", oob_path_exists])
+  AssertCommand(["chmod", "0400", oob_path_exists])
+  AssertCommand(["gnt-cluster", "copyfile", oob_path_exists])
+
+  try:
+    AssertCommand(["gnt-cluster", "modify", "--node-parameters",
+                   "oob_program=%s" % oob_path_exists])
 
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd_verify)).wait(), 0)
+    AssertCommand(_CLUSTER_VERIFY, fail=True)
+
+    AssertCommand(["chmod", "0500", oob_path_exists])
+    AssertCommand(["gnt-cluster", "copyfile", oob_path_exists])
+
+    AssertCommand(_CLUSTER_VERIFY)
+  finally:
+    AssertCommand(["gnt-cluster", "command", "rm", oob_path_exists])
+
+  AssertCommand(["gnt-cluster", "modify", "--node-parameters",
+                 "oob_program="])
+
+
+def TestClusterEpo():
+  """gnt-cluster epo"""
+  master = qa_config.GetMasterNode()
+
+  # Assert that OOB is unavailable for all nodes
+  result_output = GetCommandOutput(master.primary,
+                                   "gnt-node list --verbose --no-headers -o"
+                                   " powered")
+  AssertEqual(compat.all(powered == "(unavail)"
+                         for powered in result_output.splitlines()), True)
+
+  # Conflicting
+  AssertCommand(["gnt-cluster", "epo", "--groups", "--all"], fail=True)
+  # --all doesn't expect arguments
+  AssertCommand(["gnt-cluster", "epo", "--all", "some_arg"], fail=True)
+
+  # Unless --all is given master is not allowed to be in the list
+  AssertCommand(["gnt-cluster", "epo", "-f", master.primary], fail=True)
+
+  # This shouldn't fail
+  AssertCommand(["gnt-cluster", "epo", "-f", "--all"])
+
+  # All instances should have been stopped now
+  result_output = GetCommandOutput(master.primary,
+                                   "gnt-instance list --no-headers -o status")
+  # ERROR_down because the instance is stopped but not recorded as such
+  AssertEqual(compat.all(status == "ERROR_down"
+                         for status in result_output.splitlines()), True)
+
+  # Now start everything again
+  AssertCommand(["gnt-cluster", "epo", "--on", "-f", "--all"])
+
+  # All instances should have been started now
+  result_output = GetCommandOutput(master.primary,
+                                   "gnt-instance list --no-headers -o status")
+  AssertEqual(compat.all(status == "running"
+                         for status in result_output.splitlines()), True)
 
 
 def TestClusterVerify():
   """gnt-cluster verify"""
-  master = qa_config.GetMasterNode()
+  AssertCommand(_CLUSTER_VERIFY)
+  AssertCommand(["gnt-cluster", "verify-disks"])
+
+
+def TestJobqueue():
+  """gnt-debug test-jobqueue"""
+  AssertCommand(["gnt-debug", "test-jobqueue"])
+
+
+def TestDelay(node):
+  """gnt-debug delay"""
+  AssertCommand(["gnt-debug", "delay", "1"])
+  AssertCommand(["gnt-debug", "delay", "--no-master", "1"])
+  AssertCommand(["gnt-debug", "delay", "--no-master",
+                 "-n", node.primary, "1"])
+
+
+def TestClusterReservedLvs():
+  """gnt-cluster reserved lvs"""
+  vgname = qa_config.get("vg-name", constants.DEFAULT_VG)
+  lvname = _QA_LV_PREFIX + "test"
+  lvfullname = "/".join([vgname, lvname])
+  for fail, cmd in [
+    (False, _CLUSTER_VERIFY),
+    (False, ["gnt-cluster", "modify", "--reserved-lvs", ""]),
+    (False, ["lvcreate", "-L1G", "-n", lvname, vgname]),
+    (True, _CLUSTER_VERIFY),
+    (False, ["gnt-cluster", "modify", "--reserved-lvs",
+             "%s,.*/other-test" % lvfullname]),
+    (False, _CLUSTER_VERIFY),
+    (False, ["gnt-cluster", "modify", "--reserved-lvs",
+             ".*/%s.*" % _QA_LV_PREFIX]),
+    (False, _CLUSTER_VERIFY),
+    (False, ["gnt-cluster", "modify", "--reserved-lvs", ""]),
+    (True, _CLUSTER_VERIFY),
+    (False, ["lvremove", "-f", lvfullname]),
+    (False, _CLUSTER_VERIFY),
+    ]:
+    AssertCommand(cmd, fail=fail)
+
+
+def TestClusterModifyEmpty():
+  """gnt-cluster modify"""
+  AssertCommand(["gnt-cluster", "modify"], fail=True)
+
+
+def TestClusterModifyDisk():
+  """gnt-cluster modify -D"""
+  for param in _FAIL_PARAMS:
+    AssertCommand(["gnt-cluster", "modify", "-D", param], fail=True)
+
+
+def TestClusterModifyDiskTemplates():
+  """gnt-cluster modify --enabled-disk-templates=..."""
+  enabled_disk_templates = qa_config.GetEnabledDiskTemplates()
+  default_disk_template = qa_config.GetDefaultDiskTemplate()
+
+  _TestClusterModifyDiskTemplatesArguments(default_disk_template,
+                                           enabled_disk_templates)
+
+  _RestoreEnabledDiskTemplates()
+  nodes = qa_config.AcquireManyNodes(2)
+
+  instance_template = enabled_disk_templates[0]
+  instance = qa_instance.CreateInstanceByDiskTemplate(nodes, instance_template)
+
+  _TestClusterModifyUnusedDiskTemplate(instance_template)
+  _TestClusterModifyUsedDiskTemplate(instance_template,
+                                     enabled_disk_templates)
+
+  qa_instance.TestInstanceRemove(instance)
+  _RestoreEnabledDiskTemplates()
 
-  cmd = ['gnt-cluster', 'verify']
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+
+def _RestoreEnabledDiskTemplates():
+  """Sets the list of enabled disk templates back to the list of enabled disk
+     templates from the QA configuration. This can be used to make sure that
+     the tests that modify the list of disk templates do not interfere with
+     other tests.
+
+  """
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-template=%s" %
+       ",".join(qa_config.GetEnabledDiskTemplates())],
+    fail=False)
+
+
+def _TestClusterModifyDiskTemplatesArguments(default_disk_template,
+                                             enabled_disk_templates):
+  """Tests argument handling of 'gnt-cluster modify' with respect to
+     the parameter '--enabled-disk-templates'. This test is independent
+     of instances.
+
+  """
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-template=%s" %
+       ",".join(enabled_disk_templates)],
+    fail=False)
+
+  # bogus templates
+  AssertCommand(["gnt-cluster", "modify",
+                 "--enabled-disk-templates=pinkbunny"],
+                fail=True)
+
+  # duplicate entries do no harm
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-templates=%s,%s" %
+      (default_disk_template, default_disk_template)],
+    fail=False)
+
+
+def _TestClusterModifyUsedDiskTemplate(instance_template,
+                                       enabled_disk_templates):
+  """Tests that disk templates that are currently in use by instances cannot
+     be disabled on the cluster.
+
+  """
+  # If the list of enabled disk templates contains only one template
+  # we need to add some other templates, because the list of enabled disk
+  # templates can only be set to a non-empty list.
+  new_disk_templates = list(set(enabled_disk_templates)
+                              - set([instance_template]))
+  if not new_disk_templates:
+    new_disk_templates = list(set(constants.DISK_TEMPLATES)
+                                - set([instance_template]))
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-templates=%s" %
+       ",".join(new_disk_templates)],
+    fail=True)
+
+
+def _TestClusterModifyUnusedDiskTemplate(instance_template):
+  """Tests that unused disk templates can be disabled safely."""
+  all_disk_templates = constants.DISK_TEMPLATES
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-templates=%s" %
+       ",".join(all_disk_templates)],
+    fail=False)
+  new_disk_templates = [instance_template]
+  AssertCommand(
+    ["gnt-cluster", "modify",
+     "--enabled-disk-templates=%s" %
+       ",".join(new_disk_templates)],
+    fail=False)
+
+
+def TestClusterModifyBe():
+  """gnt-cluster modify -B"""
+  for fail, cmd in [
+    # max/min mem
+    (False, ["gnt-cluster", "modify", "-B", "maxmem=256"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *maxmem: 256$'"]),
+    (False, ["gnt-cluster", "modify", "-B", "minmem=256"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *minmem: 256$'"]),
+    (True, ["gnt-cluster", "modify", "-B", "maxmem=a"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *maxmem: 256$'"]),
+    (True, ["gnt-cluster", "modify", "-B", "minmem=a"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *minmem: 256$'"]),
+    (False, ["gnt-cluster", "modify", "-B", "maxmem=128,minmem=128"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *maxmem: 128$'"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *minmem: 128$'"]),
+    # vcpus
+    (False, ["gnt-cluster", "modify", "-B", "vcpus=4"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *vcpus: 4$'"]),
+    (True, ["gnt-cluster", "modify", "-B", "vcpus=a"]),
+    (False, ["gnt-cluster", "modify", "-B", "vcpus=1"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *vcpus: 1$'"]),
+    # auto_balance
+    (False, ["gnt-cluster", "modify", "-B", "auto_balance=False"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *auto_balance: False$'"]),
+    (True, ["gnt-cluster", "modify", "-B", "auto_balance=1"]),
+    (False, ["gnt-cluster", "modify", "-B", "auto_balance=True"]),
+    (False, ["sh", "-c", "gnt-cluster info|grep '^ *auto_balance: True$'"]),
+    ]:
+    AssertCommand(cmd, fail=fail)
+
+  # redo the original-requested BE parameters, if any
+  bep = qa_config.get("backend-parameters", "")
+  if bep:
+    AssertCommand(["gnt-cluster", "modify", "-B", bep])
+
+
+def _GetClusterIPolicy():
+  """Return the run-time values of the cluster-level instance policy.
+
+  @rtype: tuple
+  @return: (policy, specs), where:
+      - policy is a dictionary of the policy values, instance specs excluded
+      - specs is a dictionary containing only the specs, using the internal
+        format (see L{constants.IPOLICY_DEFAULTS} for an example)
+
+  """
+  info = qa_utils.GetObjectInfo(["gnt-cluster", "info"])
+  policy = info["Instance policy - limits for instances"]
+  (ret_policy, ret_specs) = qa_utils.ParseIPolicy(policy)
+
+  # Sanity checks
+  assert "minmax" in ret_specs and "std" in ret_specs
+  assert len(ret_specs["minmax"]) > 0
+  assert len(ret_policy) > 0
+  return (ret_policy, ret_specs)
+
+
+def TestClusterModifyIPolicy():
+  """gnt-cluster modify --ipolicy-*"""
+  basecmd = ["gnt-cluster", "modify"]
+  (old_policy, old_specs) = _GetClusterIPolicy()
+  for par in ["vcpu-ratio", "spindle-ratio"]:
+    curr_val = float(old_policy[par])
+    test_values = [
+      (True, 1.0),
+      (True, 1.5),
+      (True, 2),
+      (False, "a"),
+      # Restore the old value
+      (True, curr_val),
+      ]
+    for (good, val) in test_values:
+      cmd = basecmd + ["--ipolicy-%s=%s" % (par, val)]
+      AssertCommand(cmd, fail=not good)
+      if good:
+        curr_val = val
+      # Check the affected parameter
+      (eff_policy, eff_specs) = _GetClusterIPolicy()
+      AssertEqual(float(eff_policy[par]), curr_val)
+      # Check everything else
+      AssertEqual(eff_specs, old_specs)
+      for p in eff_policy.keys():
+        if p == par:
+          continue
+        AssertEqual(eff_policy[p], old_policy[p])
+
+  # Disk templates are treated slightly differently
+  par = "disk-templates"
+  disp_str = "allowed disk templates"
+  curr_val = old_policy[disp_str]
+  test_values = [
+    (True, constants.DT_PLAIN),
+    (True, "%s,%s" % (constants.DT_PLAIN, constants.DT_DRBD8)),
+    (False, "thisisnotadisktemplate"),
+    (False, ""),
+    # Restore the old value
+    (True, curr_val.replace(" ", "")),
+    ]
+  for (good, val) in test_values:
+    cmd = basecmd + ["--ipolicy-%s=%s" % (par, val)]
+    AssertCommand(cmd, fail=not good)
+    if good:
+      curr_val = val
+    # Check the affected parameter
+    (eff_policy, eff_specs) = _GetClusterIPolicy()
+    AssertEqual(eff_policy[disp_str].replace(" ", ""), curr_val)
+    # Check everything else
+    AssertEqual(eff_specs, old_specs)
+    for p in eff_policy.keys():
+      if p == disp_str:
+        continue
+      AssertEqual(eff_policy[p], old_policy[p])
+
+
+def TestClusterSetISpecs(new_specs=None, diff_specs=None, fail=False,
+                         old_values=None):
+  """Change instance specs.
+
+  At most one of new_specs or diff_specs can be specified.
+
+  @type new_specs: dict
+  @param new_specs: new complete specs, in the same format returned by
+      L{_GetClusterIPolicy}
+  @type diff_specs: dict
+  @param diff_specs: partial specs, it can be an incomplete specifications, but
+      if min/max specs are specified, their number must match the number of the
+      existing specs
+  @type fail: bool
+  @param fail: if the change is expected to fail
+  @type old_values: tuple
+  @param old_values: (old_policy, old_specs), as returned by
+      L{_GetClusterIPolicy}
+  @return: same as L{_GetClusterIPolicy}
+
+  """
+  build_cmd = lambda opts: ["gnt-cluster", "modify"] + opts
+  return qa_utils.TestSetISpecs(
+    new_specs=new_specs, diff_specs=diff_specs,
+    get_policy_fn=_GetClusterIPolicy, build_cmd_fn=build_cmd,
+    fail=fail, old_values=old_values)
+
+
+def TestClusterModifyISpecs():
+  """gnt-cluster modify --specs-*"""
+  params = ["memory-size", "disk-size", "disk-count", "cpu-count", "nic-count"]
+  (cur_policy, cur_specs) = _GetClusterIPolicy()
+  # This test assumes that there is only one min/max bound
+  assert len(cur_specs[constants.ISPECS_MINMAX]) == 1
+  for par in params:
+    test_values = [
+      (True, 0, 4, 12),
+      (True, 4, 4, 12),
+      (True, 4, 12, 12),
+      (True, 4, 4, 4),
+      (False, 4, 0, 12),
+      (False, 4, 16, 12),
+      (False, 4, 4, 0),
+      (False, 12, 4, 4),
+      (False, 12, 4, 0),
+      (False, "a", 4, 12),
+      (False, 0, "a", 12),
+      (False, 0, 4, "a"),
+      # This is to restore the old values
+      (True,
+       cur_specs[constants.ISPECS_MINMAX][0][constants.ISPECS_MIN][par],
+       cur_specs[constants.ISPECS_STD][par],
+       cur_specs[constants.ISPECS_MINMAX][0][constants.ISPECS_MAX][par])
+      ]
+    for (good, mn, st, mx) in test_values:
+      new_vals = {
+        constants.ISPECS_MINMAX: [{
+          constants.ISPECS_MIN: {par: mn},
+          constants.ISPECS_MAX: {par: mx}
+          }],
+        constants.ISPECS_STD: {par: st}
+        }
+      cur_state = (cur_policy, cur_specs)
+      # We update cur_specs, as we've copied the values to restore already
+      (cur_policy, cur_specs) = TestClusterSetISpecs(
+        diff_specs=new_vals, fail=not good, old_values=cur_state)
+
+    # Get the ipolicy command
+    mnode = qa_config.GetMasterNode()
+    initcmd = GetCommandOutput(mnode.primary, "gnt-cluster show-ispecs-cmd")
+    modcmd = ["gnt-cluster", "modify"]
+    opts = initcmd.split()
+    assert opts[0:2] == ["gnt-cluster", "init"]
+    for k in range(2, len(opts) - 1):
+      if opts[k].startswith("--ipolicy-"):
+        assert k + 2 <= len(opts)
+        modcmd.extend(opts[k:k + 2])
+    # Re-apply the ipolicy (this should be a no-op)
+    AssertCommand(modcmd)
+    new_initcmd = GetCommandOutput(mnode.primary, "gnt-cluster show-ispecs-cmd")
+    AssertEqual(initcmd, new_initcmd)
 
 
 def TestClusterInfo():
   """gnt-cluster info"""
-  master = qa_config.GetMasterNode()
+  AssertCommand(["gnt-cluster", "info"])
 
-  cmd = ['gnt-cluster', 'info']
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+
+def TestClusterRedistConf():
+  """gnt-cluster redist-conf"""
+  AssertCommand(["gnt-cluster", "redist-conf"])
 
 
 def TestClusterGetmaster():
   """gnt-cluster getmaster"""
-  master = qa_config.GetMasterNode()
-
-  cmd = ['gnt-cluster', 'getmaster']
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+  AssertCommand(["gnt-cluster", "getmaster"])
 
 
 def TestClusterVersion():
   """gnt-cluster version"""
-  master = qa_config.GetMasterNode()
-
-  cmd = ['gnt-cluster', 'version']
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+  AssertCommand(["gnt-cluster", "version"])
 
 
 def TestClusterRenewCrypto():
@@ -164,74 +720,73 @@ def TestClusterRenewCrypto():
     ["--new-cluster-domain-secret", "--cluster-domain-secret=/dev/null"],
     ]
   for i in conflicting:
-    AssertNotEqual(StartSSH(master["primary"],
-                            utils.ShellQuoteArgs(cmd + i)).wait(), 0)
+    AssertCommand(cmd + i, fail=True)
 
   # Invalid RAPI certificate
   cmd = ["gnt-cluster", "renew-crypto", "--force",
          "--rapi-certificate=/dev/null"]
-  AssertNotEqual(StartSSH(master["primary"],
-                          utils.ShellQuoteArgs(cmd)).wait(), 0)
+  AssertCommand(cmd, fail=True)
 
-  # Custom RAPI certificate
-  fh = tempfile.NamedTemporaryFile()
+  rapi_cert_backup = qa_utils.BackupFile(master.primary,
+                                         pathutils.RAPI_CERT_FILE)
+  try:
+    # Custom RAPI certificate
+    fh = tempfile.NamedTemporaryFile()
 
-  # Ensure certificate doesn't cause "gnt-cluster verify" to complain
-  validity = constants.SSL_CERT_EXPIRATION_WARN * 3
+    # Ensure certificate doesn't cause "gnt-cluster verify" to complain
+    validity = constants.SSL_CERT_EXPIRATION_WARN * 3
 
-  utils.GenerateSelfSignedSslCert(fh.name, validity=validity)
+    utils.GenerateSelfSignedSslCert(fh.name, validity=validity)
 
-  tmpcert = qa_utils.UploadFile(master["primary"], fh.name)
-  try:
-    cmd = ["gnt-cluster", "renew-crypto", "--force",
-           "--rapi-certificate=%s" % tmpcert]
-    AssertEqual(StartSSH(master["primary"],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
-  finally:
-    cmd = ["rm", "-f", tmpcert]
-    AssertEqual(StartSSH(master["primary"],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+    tmpcert = qa_utils.UploadFile(master.primary, fh.name)
+    try:
+      AssertCommand(["gnt-cluster", "renew-crypto", "--force",
+                     "--rapi-certificate=%s" % tmpcert])
+    finally:
+      AssertCommand(["rm", "-f", tmpcert])
 
-  # Custom cluster domain secret
-  cds_fh = tempfile.NamedTemporaryFile()
-  cds_fh.write(utils.GenerateSecret())
-  cds_fh.write("\n")
-  cds_fh.flush()
+    # Custom cluster domain secret
+    cds_fh = tempfile.NamedTemporaryFile()
+    cds_fh.write(utils.GenerateSecret())
+    cds_fh.write("\n")
+    cds_fh.flush()
 
-  tmpcds = qa_utils.UploadFile(master["primary"], cds_fh.name)
-  try:
-    cmd = ["gnt-cluster", "renew-crypto", "--force",
-           "--cluster-domain-secret=%s" % tmpcds]
-    AssertEqual(StartSSH(master["primary"],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
-  finally:
-    cmd = ["rm", "-f", tmpcds]
-    AssertEqual(StartSSH(master["primary"],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+    tmpcds = qa_utils.UploadFile(master.primary, cds_fh.name)
+    try:
+      AssertCommand(["gnt-cluster", "renew-crypto", "--force",
+                     "--cluster-domain-secret=%s" % tmpcds])
+    finally:
+      AssertCommand(["rm", "-f", tmpcds])
 
-  # Normal case
-  cmd = ["gnt-cluster", "renew-crypto", "--force",
-         "--new-cluster-certificate", "--new-confd-hmac-key",
-         "--new-rapi-certificate", "--new-cluster-domain-secret"]
-  AssertEqual(StartSSH(master["primary"],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+    # Normal case
+    AssertCommand(["gnt-cluster", "renew-crypto", "--force",
+                   "--new-cluster-certificate", "--new-confd-hmac-key",
+                   "--new-rapi-certificate", "--new-cluster-domain-secret"])
+
+    # Restore RAPI certificate
+    AssertCommand(["gnt-cluster", "renew-crypto", "--force",
+                   "--rapi-certificate=%s" % rapi_cert_backup])
+  finally:
+    AssertCommand(["rm", "-f", rapi_cert_backup])
 
 
 def TestClusterBurnin():
   """Burnin"""
   master = qa_config.GetMasterNode()
 
-  options = qa_config.get('options', {})
-  disk_template = options.get('burnin-disk-template', 'drbd')
-  parallel = options.get('burnin-in-parallel', False)
-  check_inst = options.get('burnin-check-instances', False)
-  do_rename = options.get('burnin-rename', '')
+  options = qa_config.get("options", {})
+  disk_template = options.get("burnin-disk-template", constants.DT_DRBD8)
+  parallel = options.get("burnin-in-parallel", False)
+  check_inst = options.get("burnin-check-instances", False)
+  do_rename = options.get("burnin-rename", "")
+  do_reboot = options.get("burnin-reboot", True)
+  reboot_types = options.get("reboot-types", constants.REBOOT_TYPES)
 
   # Get as many instances as we need
   instances = []
   try:
     try:
-      num = qa_config.get('options', {}).get('burnin-instances', 1)
+      num = qa_config.get("options", {}).get("burnin-instances", 1)
       for _ in range(0, num):
         instances.append(qa_config.AcquireInstance())
     except qa_error.OutOfInstancesError:
@@ -240,48 +795,97 @@ def TestClusterBurnin():
     if len(instances) < 1:
       raise qa_error.Error("Burnin needs at least one instance")
 
-    script = qa_utils.UploadFile(master['primary'], '../tools/burnin')
+    script = qa_utils.UploadFile(master.primary, "../tools/burnin")
     try:
+      disks = qa_config.GetDiskOptions()
       # Run burnin
       cmd = [script,
-             '-p',
-             '--os=%s' % qa_config.get('os'),
-             '--disk-size=%s' % ",".join(qa_config.get('disk')),
-             '--disk-growth=%s' % ",".join(qa_config.get('disk-growth')),
-             '--disk-template=%s' % disk_template]
+             "--os=%s" % qa_config.get("os"),
+             "--minmem-size=%s" % qa_config.get(constants.BE_MINMEM),
+             "--maxmem-size=%s" % qa_config.get(constants.BE_MAXMEM),
+             "--disk-size=%s" % ",".join([d.get("size") for d in disks]),
+             "--disk-growth=%s" % ",".join([d.get("growth") for d in disks]),
+             "--disk-template=%s" % disk_template]
       if parallel:
-        cmd.append('--parallel')
+        cmd.append("--parallel")
+        cmd.append("--early-release")
       if check_inst:
-        cmd.append('--http-check')
+        cmd.append("--http-check")
       if do_rename:
-        cmd.append('--rename=%s' % do_rename)
-      cmd += [inst['name'] for inst in instances]
-      AssertEqual(StartSSH(master['primary'],
-                           utils.ShellQuoteArgs(cmd)).wait(), 0)
+        cmd.append("--rename=%s" % do_rename)
+      if not do_reboot:
+        cmd.append("--no-reboot")
+      else:
+        cmd.append("--reboot-types=%s" % ",".join(reboot_types))
+      cmd += [inst.name for inst in instances]
+      AssertCommand(cmd)
     finally:
-      cmd = ['rm', '-f', script]
-      AssertEqual(StartSSH(master['primary'],
-                           utils.ShellQuoteArgs(cmd)).wait(), 0)
+      AssertCommand(["rm", "-f", script])
+
   finally:
     for inst in instances:
-      qa_config.ReleaseInstance(inst)
+      inst.Release()
 
 
 def TestClusterMasterFailover():
-  """gnt-cluster masterfailover"""
+  """gnt-cluster master-failover"""
   master = qa_config.GetMasterNode()
+  failovermaster = qa_config.AcquireNode(exclude=master)
+
+  cmd = ["gnt-cluster", "master-failover"]
+  node_list_cmd = ["gnt-node", "list"]
+  try:
+    AssertCommand(cmd, node=failovermaster)
+    AssertCommand(node_list_cmd, node=failovermaster)
+    # Back to original master node
+    AssertCommand(cmd, node=master)
+    AssertCommand(node_list_cmd, node=master)
+  finally:
+    failovermaster.Release()
+
+
+def _NodeQueueDrainFile(node):
+  """Returns path to queue drain file for a node.
+
+  """
+  return qa_utils.MakeNodePath(node, pathutils.JOB_QUEUE_DRAIN_FILE)
+
+
+def _AssertDrainFile(node, **kwargs):
+  """Checks for the queue drain file.
+
+  """
+  AssertCommand(["test", "-f", _NodeQueueDrainFile(node)], node=node, **kwargs)
+
 
+def TestClusterMasterFailoverWithDrainedQueue():
+  """gnt-cluster master-failover with drained queue"""
+  master = qa_config.GetMasterNode()
   failovermaster = qa_config.AcquireNode(exclude=master)
+
+  # Ensure queue is not drained
+  for node in [master, failovermaster]:
+    _AssertDrainFile(node, fail=True)
+
+  # Drain queue on failover master
+  AssertCommand(["touch", _NodeQueueDrainFile(failovermaster)],
+                node=failovermaster)
+
+  cmd = ["gnt-cluster", "master-failover"]
   try:
-    cmd = ['gnt-cluster', 'masterfailover']
-    AssertEqual(StartSSH(failovermaster['primary'],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+    _AssertDrainFile(failovermaster)
+    AssertCommand(cmd, node=failovermaster)
+    _AssertDrainFile(master, fail=True)
+    _AssertDrainFile(failovermaster, fail=True)
 
-    cmd = ['gnt-cluster', 'masterfailover']
-    AssertEqual(StartSSH(master['primary'],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+    # Back to original master node
+    AssertCommand(cmd, node=master)
   finally:
-    qa_config.ReleaseNode(failovermaster)
+    failovermaster.Release()
+
+  # Ensure queue is not drained
+  for node in [master, failovermaster]:
+    _AssertDrainFile(node, fail=True)
 
 
 def TestClusterCopyfile():
@@ -297,12 +901,10 @@ def TestClusterCopyfile():
   f.seek(0)
 
   # Upload file to master node
-  testname = qa_utils.UploadFile(master['primary'], f.name)
+  testname = qa_utils.UploadFile(master.primary, f.name)
   try:
     # Copy file to all nodes
-    cmd = ['gnt-cluster', 'copyfile', testname]
-    AssertEqual(StartSSH(master['primary'],
-                         utils.ShellQuoteArgs(cmd)).wait(), 0)
+    AssertCommand(["gnt-cluster", "copyfile", testname])
     _CheckFileOnAllNodes(testname, uniqueid)
   finally:
     _RemoveFileFromAllNodes(testname)
@@ -310,16 +912,14 @@ def TestClusterCopyfile():
 
 def TestClusterCommand():
   """gnt-cluster command"""
-  master = qa_config.GetMasterNode()
-
   uniqueid = utils.NewUUID()
   rfile = "/tmp/gnt%s" % utils.NewUUID()
-  rcmd = utils.ShellQuoteArgs(['echo', '-n', uniqueid])
-  cmd = utils.ShellQuoteArgs(['gnt-cluster', 'command',
+  rcmd = utils.ShellQuoteArgs(["echo", "-n", uniqueid])
+  cmd = utils.ShellQuoteArgs(["gnt-cluster", "command",
                               "%s >%s" % (rcmd, rfile)])
 
   try:
-    AssertEqual(StartSSH(master['primary'], cmd).wait(), 0)
+    AssertCommand(cmd)
     _CheckFileOnAllNodes(rfile, uniqueid)
   finally:
     _RemoveFileFromAllNodes(rfile)
@@ -327,8 +927,48 @@ def TestClusterCommand():
 
 def TestClusterDestroy():
   """gnt-cluster destroy"""
-  master = qa_config.GetMasterNode()
+  AssertCommand(["gnt-cluster", "destroy", "--yes-do-it"])
+
+
+def TestClusterRepairDiskSizes():
+  """gnt-cluster repair-disk-sizes"""
+  AssertCommand(["gnt-cluster", "repair-disk-sizes"])
 
-  cmd = ['gnt-cluster', 'destroy', '--yes-do-it']
-  AssertEqual(StartSSH(master['primary'],
-                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+
+def TestSetExclStorCluster(newvalue):
+  """Set the exclusive_storage node parameter at the cluster level.
+
+  @type newvalue: bool
+  @param newvalue: New value of exclusive_storage
+  @rtype: bool
+  @return: The old value of exclusive_storage
+
+  """
+  es_path = ["Default node parameters", "exclusive_storage"]
+  oldvalue = _GetClusterField(es_path)
+  AssertCommand(["gnt-cluster", "modify", "--node-parameters",
+                 "exclusive_storage=%s" % newvalue])
+  effvalue = _GetClusterField(es_path)
+  if effvalue != newvalue:
+    raise qa_error.Error("exclusive_storage has the wrong value: %s instead"
+                         " of %s" % (effvalue, newvalue))
+  qa_config.SetExclusiveStorage(newvalue)
+  return oldvalue
+
+
+def TestExclStorSharedPv(node):
+  """cluster-verify reports LVs that share the same PV with exclusive_storage.
+
+  """
+  vgname = qa_config.get("vg-name", constants.DEFAULT_VG)
+  lvname1 = _QA_LV_PREFIX + "vol1"
+  lvname2 = _QA_LV_PREFIX + "vol2"
+  node_name = node.primary
+  AssertCommand(["lvcreate", "-L1G", "-n", lvname1, vgname], node=node_name)
+  AssertClusterVerify(fail=True, errors=[constants.CV_ENODEORPHANLV])
+  AssertCommand(["lvcreate", "-L1G", "-n", lvname2, vgname], node=node_name)
+  AssertClusterVerify(fail=True, errors=[constants.CV_ENODELVM,
+                                         constants.CV_ENODEORPHANLV])
+  AssertCommand(["lvremove", "-f", "/".join([vgname, lvname1])], node=node_name)
+  AssertCommand(["lvremove", "-f", "/".join([vgname, lvname2])], node=node_name)
+  AssertClusterVerify()