-# Copyright (C) 2007 Google Inc.
+#
+#
+
+# Copyright (C) 2007, 2010, 2011, 2012 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 tempfile
+import os.path
+from ganeti import constants
+from ganeti import compat
from ganeti import utils
import qa_config
import qa_utils
import qa_error
-from qa_utils import AssertEqual, StartSSH
+from qa_utils import AssertEqual, AssertCommand, GetCommandOutput
+
+
+#: 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):
"""
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)
+
+# 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",
+ ]
-@qa_utils.DefineHook('cluster-init')
-def TestClusterInit():
+
+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()
- cmd = ['gnt-cluster', 'init']
+ rapi_dir = os.path.dirname(constants.RAPI_USERS_FILE)
- if master.get('secondary', None):
- cmd.append('--secondary-ip=%s' % master['secondary'])
+ # First create the RAPI credentials
+ fh = tempfile.NamedTemporaryFile()
+ try:
+ fh.write("%s %s write\n" % (rapi_user, rapi_secret))
+ fh.flush()
- bridge = qa_config.get('bridge', None)
+ tmpru = qa_utils.UploadFile(master["primary"], fh.name)
+ try:
+ AssertCommand(["mkdir", "-p", rapi_dir])
+ AssertCommand(["mv", tmpru, constants.RAPI_USERS_FILE])
+ 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()),
+ ]
+
+ 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:
+ cmd.append("--specs-%s=%s=%d" % (spec_type, spec_val, spec))
+
+ if master.get("secondary", None):
+ cmd.append("--secondary-ip=%s" % master["secondary"])
+
+ bridge = qa_config.get("bridge", None)
if bridge:
- cmd.append('--bridge=%s' % bridge)
- cmd.append('--master-netdev=%s' % bridge)
+ cmd.append("--bridge=%s" % bridge)
+ cmd.append("--master-netdev=%s" % bridge)
- cmd.append(qa_config.get('name'))
+ cmd.append(qa_config.get("name"))
+ AssertCommand(cmd)
- AssertEqual(StartSSH(master['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+ cmd = ["gnt-cluster", "modify"]
+ # 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])
-@qa_utils.DefineHook('cluster-verify')
-def TestClusterVerify():
- """gnt-cluster verify"""
+ 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"""
+ cmd = ["gnt-cluster", "rename", "-f"]
+
+ 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')
+ return
+
+ for data in [
+ cmd + [rename_target],
+ _CLUSTER_VERIFY,
+ cmd + [original_name],
+ _CLUSTER_VERIFY,
+ ]:
+ AssertCommand(data)
+
+
+def TestClusterOob():
+ """out-of-band framework"""
+ oob_path_exists = "/tmp/ganeti-qa-oob-does-exist-%s" % utils.NewUUID()
+
+ AssertCommand(_CLUSTER_VERIFY)
+ AssertCommand(["gnt-cluster", "modify", "--node-parameters",
+ "oob_program=/tmp/ganeti-qa-oob-does-not-exist-%s" %
+ utils.NewUUID()])
+
+ AssertCommand(_CLUSTER_VERIFY, fail=True)
+
+ 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])
+
+ 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()
- cmd = ['gnt-cluster', 'verify']
- AssertEqual(StartSSH(master['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+ # 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"""
+ 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"""
+ for fail, cmd in [
+ (False, _CLUSTER_VERIFY),
+ (False, ["gnt-cluster", "modify", "--reserved-lvs", ""]),
+ (False, ["lvcreate", "-L1G", "-nqa-test", "xenvg"]),
+ (True, _CLUSTER_VERIFY),
+ (False, ["gnt-cluster", "modify", "--reserved-lvs",
+ "xenvg/qa-test,.*/other-test"]),
+ (False, _CLUSTER_VERIFY),
+ (False, ["gnt-cluster", "modify", "--reserved-lvs", ".*/qa-.*"]),
+ (False, _CLUSTER_VERIFY),
+ (False, ["gnt-cluster", "modify", "--reserved-lvs", ""]),
+ (True, _CLUSTER_VERIFY),
+ (False, ["lvremove", "-f", "xenvg/qa-test"]),
+ (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 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])
-@qa_utils.DefineHook('cluster-info')
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"])
-@qa_utils.DefineHook('cluster-getmaster')
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"])
-@qa_utils.DefineHook('cluster-version')
def TestClusterVersion():
"""gnt-cluster version"""
+ AssertCommand(["gnt-cluster", "version"])
+
+
+def TestClusterRenewCrypto():
+ """gnt-cluster renew-crypto"""
master = qa_config.GetMasterNode()
- cmd = ['gnt-cluster', 'version']
- AssertEqual(StartSSH(master['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+ # Conflicting options
+ cmd = ["gnt-cluster", "renew-crypto", "--force",
+ "--new-cluster-certificate", "--new-confd-hmac-key"]
+ conflicting = [
+ ["--new-rapi-certificate", "--rapi-certificate=/dev/null"],
+ ["--new-cluster-domain-secret", "--cluster-domain-secret=/dev/null"],
+ ]
+ for i in conflicting:
+ AssertCommand(cmd + i, fail=True)
+
+ # Invalid RAPI certificate
+ cmd = ["gnt-cluster", "renew-crypto", "--force",
+ "--rapi-certificate=/dev/null"]
+ AssertCommand(cmd, fail=True)
+
+ rapi_cert_backup = qa_utils.BackupFile(master["primary"],
+ constants.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
+
+ utils.GenerateSelfSignedSslCert(fh.name, validity=validity)
+
+ 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()
+
+ 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
+ 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])
-@qa_utils.DefineHook('cluster-burnin')
def TestClusterBurnin():
"""Burnin"""
master = qa_config.GetMasterNode()
- disk_template = (qa_config.get('options', {}).
- get('burnin-disk-template', 'remote_raid1'))
+ 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", "")
+ 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)
- for _ in xrange(0, num):
+ num = qa_config.get("options", {}).get("burnin-instances", 1)
+ for _ in range(0, num):
instances.append(qa_config.AcquireInstance())
except qa_error.OutOfInstancesError:
print "Not enough instances, continuing anyway."
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:
# Run burnin
cmd = [script,
- '--os=%s' % qa_config.get('os'),
- '--os-size=%s' % qa_config.get('os-size'),
- '--swap-size=%s' % qa_config.get('swap-size'),
- '--disk-template=%s' % disk_template]
- cmd += [inst['name'] for inst in instances]
- AssertEqual(StartSSH(master['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+ "--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(qa_config.get("disk")),
+ "--disk-growth=%s" % ",".join(qa_config.get("disk-growth")),
+ "--disk-template=%s" % disk_template]
+ if parallel:
+ cmd.append("--parallel")
+ cmd.append("--early-release")
+ if check_inst:
+ cmd.append("--http-check")
+ if do_rename:
+ 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)
-@qa_utils.DefineHook('cluster-master-failover')
def TestClusterMasterFailover():
- """gnt-cluster masterfailover"""
+ """gnt-cluster master-failover"""
master = qa_config.GetMasterNode()
+ failovermaster = qa_config.AcquireNode(exclude=master)
+
+ cmd = ["gnt-cluster", "master-failover"]
+ try:
+ AssertCommand(cmd, node=failovermaster)
+ # Back to original master node
+ AssertCommand(cmd, node=master)
+ finally:
+ qa_config.ReleaseNode(failovermaster)
+
+def TestClusterMasterFailoverWithDrainedQueue():
+ """gnt-cluster master-failover with drained queue"""
+ drain_check = ["test", "-f", constants.JOB_QUEUE_DRAIN_FILE]
+
+ master = qa_config.GetMasterNode()
failovermaster = qa_config.AcquireNode(exclude=master)
+
+ # Ensure queue is not drained
+ for node in [master, failovermaster]:
+ AssertCommand(drain_check, node=node, fail=True)
+
+ # Drain queue on failover master
+ AssertCommand(["touch", constants.JOB_QUEUE_DRAIN_FILE], node=failovermaster)
+
+ cmd = ["gnt-cluster", "master-failover"]
try:
- cmd = ['gnt-cluster', 'masterfailover']
- AssertEqual(StartSSH(failovermaster['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+ AssertCommand(drain_check, node=failovermaster)
+ AssertCommand(cmd, node=failovermaster)
+ AssertCommand(drain_check, fail=True)
+ AssertCommand(drain_check, node=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)
+ AssertCommand(drain_check, fail=True)
+ AssertCommand(drain_check, node=failovermaster, fail=True)
+
-@qa_utils.DefineHook('cluster-copyfile')
def TestClusterCopyfile():
"""gnt-cluster copyfile"""
master = qa_config.GetMasterNode()
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)
-@qa_utils.DefineHook('cluster-command')
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)
-@qa_utils.DefineHook('cluster-destroy')
def TestClusterDestroy():
"""gnt-cluster destroy"""
- master = qa_config.GetMasterNode()
+ AssertCommand(["gnt-cluster", "destroy", "--yes-do-it"])
+
- cmd = ['gnt-cluster', 'destroy', '--yes-do-it']
- AssertEqual(StartSSH(master['primary'],
- utils.ShellQuoteArgs(cmd)).wait(), 0)
+def TestClusterRepairDiskSizes():
+ """gnt-cluster repair-disk-sizes"""
+ AssertCommand(["gnt-cluster", "repair-disk-sizes"])