def enable_sbd(
lib_env, default_watchdog, watchdog_dict, sbd_options,
default_device_list=None, node_device_dict=None, allow_unknown_opts=False,
ignore_offline_nodes=False, no_watchdog_validation=False,
allow_invalid_option_values=False,
):
# pylint: disable=too-many-arguments, too-many-locals
"""
Enable SBD on all nodes in cluster.
lib_env -- LibraryEnvironment
default_watchdog -- watchdog for nodes which are not specified in
watchdog_dict. Uses default value from settings if None.
watchdog_dict -- dictionary with node names as keys and watchdog path
as value
sbd_options -- dictionary in format: <SBD config option>: <value>
default_device_list -- list of devices for all nodes
node_device_dict -- dictionary with node names as keys and list of devices
as value
allow_unknown_opts -- if True, accept also unknown options.
ignore_offline_nodes -- if True, omit offline nodes
no_watchdog_validation -- it True, do not validate existance of a watchdog
on the nodes
allow_invalid_option_values -- if True, invalid values of some options will
be treated as warning instead of errors
"""
using_devices = not (
default_device_list is None and node_device_dict is None
)
if default_device_list is None:
default_device_list = []
if node_device_dict is None:
node_device_dict = {}
if not default_watchdog:
default_watchdog = settings.sbd_watchdog_default
sbd_options = {opt.upper(): val for opt, val in sbd_options.items()}
corosync_conf = lib_env.get_corosync_conf()
node_list, get_nodes_report_list = get_existing_nodes_names(corosync_conf)
if not node_list:
get_nodes_report_list.append(reports.corosync_config_no_nodes_defined())
target_list = lib_env.get_node_target_factory().get_target_list(
node_list, skip_non_existing=ignore_offline_nodes,
)
full_watchdog_dict = _get_full_target_dict(
target_list, watchdog_dict, default_watchdog
)
full_device_dict = _get_full_target_dict(
target_list, node_device_dict, default_device_list
)
lib_env.report_processor.process_list(
get_nodes_report_list
+
[
reports.node_not_found(node)
for node in (
set(list(watchdog_dict.keys()) + list(node_device_dict.keys()))
-
set(node_list)
)
]
+
_validate_watchdog_dict(full_watchdog_dict)
+
(sbd.validate_nodes_devices(full_device_dict) if using_devices else [])
+
_validate_sbd_options(
sbd_options, allow_unknown_opts, allow_invalid_option_values
)
)
com_cmd = GetOnlineTargets(
lib_env.report_processor, ignore_offline_targets=ignore_offline_nodes,
)
com_cmd.set_targets(target_list)
online_targets = run_and_raise(lib_env.get_node_communicator(), com_cmd)
# check if SBD can be enabled
if no_watchdog_validation:
lib_env.report_processor.report(
reports.sbd_watchdog_validation_inactive()
)
com_cmd = CheckSbd(lib_env.report_processor)
for target in online_targets:
com_cmd.add_request(
target,
(
# Do not send watchdog if validation is turned off. Listing of
# available watchdogs in pcsd may restart the machine in some
# corner cases.
"" if no_watchdog_validation
else full_watchdog_dict[target.label]
),
full_device_dict[target.label] if using_devices else [],
)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# enable ATB if needed
if not using_devices:
if sbd.atb_has_to_be_enabled_pre_enable_check(corosync_conf):
lib_env.report_processor.process(
reports.corosync_quorum_atb_will_be_enabled_due_to_sbd()
)
corosync_conf.set_quorum_options({"auto_tie_breaker": "1"})
lib_env.push_corosync_conf(corosync_conf, ignore_offline_nodes)
# distribute SBD configuration
config = sbd.get_default_sbd_config()
config.update(sbd_options)
com_cmd = SetSbdConfig(lib_env.report_processor)
for target in online_targets:
com_cmd.add_request(
target,
sbd.create_sbd_config(
config,
target.label,
full_watchdog_dict[target.label],
full_device_dict[target.label]
)
)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# remove cluster prop 'stonith_watchdog_timeout'
com_cmd = RemoveStonithWatchdogTimeout(lib_env.report_processor)
com_cmd.set_targets(online_targets)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# enable SBD service an all nodes
com_cmd = EnableSbdService(lib_env.report_processor)
com_cmd.set_targets(online_targets)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
lib_env.report_processor.process(
reports.cluster_restart_required_to_apply_changes()
)
def enable_sbd(
lib_env,
default_watchdog,
watchdog_dict,
sbd_options,
default_device_list=None,
node_device_dict=None,
allow_unknown_opts=False,
ignore_offline_nodes=False,
no_watchdog_validation=False,
allow_invalid_option_values=False,
):
# pylint: disable=too-many-arguments, too-many-locals
"""
Enable SBD on all nodes in cluster.
lib_env -- LibraryEnvironment
default_watchdog -- watchdog for nodes which are not specified in
watchdog_dict. Uses default value from settings if None.
watchdog_dict -- dictionary with node names as keys and watchdog path
as value
sbd_options -- dictionary in format: <SBD config option>: <value>
default_device_list -- list of devices for all nodes
node_device_dict -- dictionary with node names as keys and list of devices
as value
allow_unknown_opts -- if True, accept also unknown options.
ignore_offline_nodes -- if True, omit offline nodes
no_watchdog_validation -- it True, do not validate existance of a watchdog
on the nodes
allow_invalid_option_values -- if True, invalid values of some options will
be treated as warning instead of errors
"""
using_devices = not (default_device_list is None
and node_device_dict is None)
if default_device_list is None:
default_device_list = []
if node_device_dict is None:
node_device_dict = {}
if not default_watchdog:
default_watchdog = settings.sbd_watchdog_default
sbd_options = {opt.upper(): val for opt, val in sbd_options.items()}
corosync_conf = lib_env.get_corosync_conf()
node_list, get_nodes_report_list = get_existing_nodes_names(corosync_conf)
if not node_list:
get_nodes_report_list.append(
reports.corosync_config_no_nodes_defined())
target_list = lib_env.get_node_target_factory().get_target_list(
node_list,
skip_non_existing=ignore_offline_nodes,
)
full_watchdog_dict = _get_full_target_dict(target_list, watchdog_dict,
default_watchdog)
full_device_dict = _get_full_target_dict(target_list, node_device_dict,
default_device_list)
if lib_env.report_processor.report_list(
get_nodes_report_list +
[
reports.node_not_found(node) for node in
(set(list(watchdog_dict.keys()) + list(node_device_dict.keys())) -
set(node_list))
] + _validate_watchdog_dict(full_watchdog_dict) +
(sbd.validate_nodes_devices(full_device_dict) if using_devices else []
) + _validate_sbd_options(sbd_options, allow_unknown_opts,
allow_invalid_option_values)).has_errors:
raise LibraryError()
com_cmd = GetOnlineTargets(
lib_env.report_processor,
ignore_offline_targets=ignore_offline_nodes,
)
com_cmd.set_targets(target_list)
online_targets = run_and_raise(lib_env.get_node_communicator(), com_cmd)
# check if SBD can be enabled
if no_watchdog_validation:
lib_env.report_processor.report(
reports.sbd_watchdog_validation_inactive())
com_cmd = CheckSbd(lib_env.report_processor)
for target in online_targets:
com_cmd.add_request(
target,
(
# Do not send watchdog if validation is turned off. Listing of
# available watchdogs in pcsd may restart the machine in some
# corner cases.
"" if no_watchdog_validation else
full_watchdog_dict[target.label]),
full_device_dict[target.label] if using_devices else [],
)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# enable ATB if needed
if not using_devices:
if sbd.atb_has_to_be_enabled_pre_enable_check(corosync_conf):
lib_env.report_processor.report(
reports.corosync_quorum_atb_will_be_enabled_due_to_sbd())
corosync_conf.set_quorum_options({"auto_tie_breaker": "1"})
lib_env.push_corosync_conf(corosync_conf, ignore_offline_nodes)
# distribute SBD configuration
config = sbd.get_default_sbd_config()
config.update(sbd_options)
com_cmd = SetSbdConfig(lib_env.report_processor)
for target in online_targets:
com_cmd.add_request(
target,
sbd.create_sbd_config(config, target.label,
full_watchdog_dict[target.label],
full_device_dict[target.label]))
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# remove cluster prop 'stonith_watchdog_timeout'
com_cmd = RemoveStonithWatchdogTimeout(lib_env.report_processor)
com_cmd.set_targets(online_targets)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
# enable SBD service an all nodes
com_cmd = EnableSbdService(lib_env.report_processor)
com_cmd.set_targets(online_targets)
run_and_raise(lib_env.get_node_communicator(), com_cmd)
lib_env.report_processor.report(
reports.cluster_restart_required_to_apply_changes())
def add_nodes(
env,
nodes,
wait=False,
start=False,
enable=False,
force=False,
force_unresolvable=False,
skip_offline_nodes=False,
no_watchdog_validation=False,
):
# pylint: disable=too-many-locals
"""
Add specified nodes to the local cluster
Raise LibraryError on any error.
env LibraryEnvironment
nodes list -- list of dicts which represents node.
Supported keys are: name (required), addrs (list), devices (list),
watchdog
wait -- specifies if command should try to wait for cluster to start up.
Has no effect start is False. If set to False command will not wait for
cluster to start. If None command will wait for some default timeout.
If int wait set timeout to int value of seconds.
start bool -- if True start cluster when it is set up
enable bool -- if True enable cluster when it is set up
force bool -- if True some validations errors are treated as warnings
force_unresolvable bool -- if True not resolvable addresses of nodes are
treated as warnings
skip_offline_nodes bool -- if True non fatal connection failures to other
hosts are treated as warnings
no_watchdog_validation bool -- if True do not validate specified watchdogs
on remote hosts
"""
_ensure_live_env(env) # raises if env is not live
report_processor = SimpleReportProcessor(env.report_processor)
target_factory = env.get_node_target_factory()
is_sbd_enabled = sbd.is_sbd_enabled(env.cmd_runner())
corosync_conf = env.get_corosync_conf()
cluster_nodes_names = corosync_conf.get_nodes_names()
corosync_node_options = {"name", "addrs"}
sbd_node_options = {"devices", "watchdog"}
keys_to_normalize = {"addrs"}
if is_sbd_enabled:
keys_to_normalize |= sbd_node_options
new_nodes = [_normalize_dict(node, keys_to_normalize) for node in nodes]
# get targets for existing nodes
target_report_list, cluster_nodes_target_list = (
target_factory.get_target_list_with_reports(
cluster_nodes_names,
skip_non_existing=skip_offline_nodes,
))
report_processor.report_list(target_report_list)
# get a target for qnetd if needed
qdevice_model, qdevice_model_options, _, _ = (
corosync_conf.get_quorum_device_settings())
if qdevice_model == "net":
try:
qnetd_target = target_factory.get_target(
qdevice_model_options["host"])
except HostNotFound:
report_processor.report(
reports.host_not_found([qdevice_model_options["host"]]))
# Get targets for new nodes and report unknown (== not-authorized) nodes.
# If a node doesn't contain the 'name' key, validation of inputs reports it.
# That means we don't report missing names but cannot rely on them being
# present either.
target_report_list, new_nodes_target_list = (
target_factory.get_target_list_with_reports(
[node["name"] for node in new_nodes if "name" in node],
allow_skip=False,
))
report_processor.report_list(target_report_list)
# Set default values for not-specified node options.
# Use an address defined in known-hosts for each node with no addresses
# specified. This allows users not to specify node addresses at all which
# simplifies the whole node add command / form significantly.
new_nodes_target_dict = {
target.label: target
for target in new_nodes_target_list
}
addrs_defaulter = _get_addrs_defaulter(report_processor,
new_nodes_target_dict)
new_nodes_defaulters = {"addrs": addrs_defaulter}
if is_sbd_enabled:
watchdog_defaulter = _get_watchdog_defaulter(report_processor,
new_nodes_target_dict)
new_nodes_defaulters["devices"] = lambda _: []
new_nodes_defaulters["watchdog"] = watchdog_defaulter
new_nodes = [
_set_defaults_in_dict(node, new_nodes_defaulters) for node in new_nodes
]
new_nodes_dict = {
node["name"]: node
for node in new_nodes if "name" in node
}
# Validate inputs - node options names
# We do not want to make corosync validators know about SBD options and
# vice versa. Therefore corosync and SBD validators get only valid corosync
# and SBD options respectively, and we need to check for any surplus
# options here.
report_processor.report_list(
validate_names_in(
corosync_node_options | sbd_node_options,
set([
option for node_options in [node.keys() for node in new_nodes]
for option in node_options
]),
option_type="node"))
# Validate inputs - corosync part
try:
cib = env.get_cib()
cib_nodes = get_remote_nodes(cib) + get_guest_nodes(cib)
except LibraryError:
cib_nodes = []
report_processor.report(
reports.get_problem_creator(
report_codes.FORCE_LOAD_NODES_FROM_CIB,
force)(reports.cib_load_error_get_nodes_for_validation))
# corosync validator rejects non-corosync keys
new_nodes_corosync = [{
key: node[key]
for key in corosync_node_options if key in node
} for node in new_nodes]
report_processor.report_list(
config_validators.add_nodes(new_nodes_corosync,
corosync_conf.get_nodes(),
cib_nodes,
force_unresolvable=force_unresolvable))
# Validate inputs - SBD part
if is_sbd_enabled:
report_processor.report_list(
sbd.validate_new_nodes_devices({
node["name"]: node["devices"]
for node in new_nodes if "name" in node
}))
else:
for node in new_nodes:
sbd_options = sbd_node_options.intersection(node.keys())
if sbd_options and "name" in node:
report_processor.report(
reports.sbd_not_used_cannot_set_sbd_options(
sbd_options, node["name"]))
# Validate inputs - flags part
wait_timeout = _get_validated_wait_timeout(report_processor, wait, start)
# Get online cluster nodes
# This is the only call in which we accept skip_offline_nodes option for the
# cluster nodes. In all the other actions we communicate only with the
# online nodes. This allows us to simplify code as any communication issue
# is considered an error, ends the command processing and is not possible
# to skip it by skip_offline_nodes. We do not have to care about a situation
# when a communication command cannot connect to some nodes and then the
# next command can connect but fails due to the previous one did not
# succeed.
online_cluster_target_list = []
if cluster_nodes_target_list:
com_cmd = GetOnlineTargets(
report_processor,
ignore_offline_targets=skip_offline_nodes,
)
com_cmd.set_targets(cluster_nodes_target_list)
online_cluster_target_list = run_com(env.get_node_communicator(),
com_cmd)
offline_cluster_target_list = [
target for target in cluster_nodes_target_list
if target not in online_cluster_target_list
]
if len(online_cluster_target_list) == 0:
report_processor.report(
reports.unable_to_perform_operation_on_any_node())
elif offline_cluster_target_list and skip_offline_nodes:
# TODO: report (warn) how to fix offline nodes when they come online
# report_processor.report(None)
pass
# Validate existing cluster nodes status
atb_has_to_be_enabled = sbd.atb_has_to_be_enabled(env.cmd_runner(),
corosync_conf,
len(new_nodes))
if atb_has_to_be_enabled:
report_processor.report(
reports.corosync_quorum_atb_will_be_enabled_due_to_sbd())
if online_cluster_target_list:
com_cmd = CheckCorosyncOffline(
report_processor,
allow_skip_offline=False,
)
com_cmd.set_targets(online_cluster_target_list)
run_com(env.get_node_communicator(), com_cmd)
# Validate new nodes. All new nodes have to be online.
com_cmd = GetHostInfo(report_processor)
com_cmd.set_targets(new_nodes_target_list)
report_processor.report_list(
_host_check_cluster_setup(
run_com(env.get_node_communicator(), com_cmd),
force,
# version of services may not be the same across the existing
# cluster nodes, so it's not easy to make this check properly
check_services_versions=False,
))
# Validate SBD on new nodes
if is_sbd_enabled:
if no_watchdog_validation:
report_processor.report(reports.sbd_watchdog_validation_inactive())
com_cmd = CheckSbd(report_processor)
for new_node_target in new_nodes_target_list:
new_node = new_nodes_dict[new_node_target.label]
# Do not send watchdog if validation is turned off. Listing of
# available watchdogs in pcsd may restart the machine in some
# corner cases.
com_cmd.add_request(
new_node_target,
watchdog=""
if no_watchdog_validation else new_node["watchdog"],
device_list=new_node["devices"],
)
run_com(env.get_node_communicator(), com_cmd)
if report_processor.has_errors:
raise LibraryError()
# Validation done. If errors occured, an exception has been raised and we
# don't get below this line.
# First set up everything else than corosync. Once the new nodes are present
# in corosync.conf, they're considered part of a cluster and the node add
# command cannot be run again. So we need to minimize the amout of actions
# (and therefore possible failures) after adding the nodes to corosync.
# distribute auth tokens of all cluster nodes (including the new ones) to
# all new nodes
com_cmd = UpdateKnownHosts(
env.report_processor,
known_hosts_to_add=env.get_known_hosts(cluster_nodes_names +
list(new_nodes_dict.keys())),
known_hosts_to_remove=[],
)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
# qdevice setup
if qdevice_model == "net":
qdevice_net.set_up_client_certificates(
env.cmd_runner(),
env.report_processor,
env.communicator_factory,
qnetd_target,
corosync_conf.get_cluster_name(),
new_nodes_target_list,
# we don't want to allow skiping offline nodes which are being
# added, otherwise qdevice will not work properly
skip_offline_nodes=False,
allow_skip_offline=False)
# sbd setup
if is_sbd_enabled:
sbd_cfg = environment_file_to_dict(sbd.get_local_sbd_config())
com_cmd = SetSbdConfig(env.report_processor)
for new_node_target in new_nodes_target_list:
new_node = new_nodes_dict[new_node_target.label]
com_cmd.add_request(
new_node_target,
sbd.create_sbd_config(
sbd_cfg,
new_node["name"],
watchdog=new_node["watchdog"],
device_list=new_node["devices"],
))
run_and_raise(env.get_node_communicator(), com_cmd)
com_cmd = EnableSbdService(env.report_processor)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
else:
com_cmd = DisableSbdService(env.report_processor)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
# booth setup
booth_sync.send_all_config_to_node(
env.get_node_communicator(),
env.report_processor,
new_nodes_target_list,
rewrite_existing=force,
skip_wrong_config=force,
)
# distribute corosync and pacemaker authkeys
files_action = {}
forceable_io_error_creator = reports.get_problem_creator(
report_codes.SKIP_FILE_DISTRIBUTION_ERRORS, force)
if os.path.isfile(settings.corosync_authkey_file):
try:
files_action.update(
node_communication_format.corosync_authkey_file(
open(settings.corosync_authkey_file, "rb").read()))
except EnvironmentError as e:
report_processor.report(
forceable_io_error_creator(
reports.file_io_error,
env_file_role_codes.COROSYNC_AUTHKEY,
file_path=settings.corosync_authkey_file,
operation="read",
reason=format_environment_error(e)))
if os.path.isfile(settings.pacemaker_authkey_file):
try:
files_action.update(
node_communication_format.pcmk_authkey_file(
open(settings.pacemaker_authkey_file, "rb").read()))
except EnvironmentError as e:
report_processor.report(
forceable_io_error_creator(
reports.file_io_error,
env_file_role_codes.PACEMAKER_AUTHKEY,
file_path=settings.pacemaker_authkey_file,
operation="read",
reason=format_environment_error(e)))
# pcs_settings.conf was previously synced using pcsdcli send_local_configs.
# This has been changed temporarily until new system for distribution and
# syncronization of configs will be introduced.
if os.path.isfile(settings.pcsd_settings_conf_location):
try:
files_action.update(
node_communication_format.pcs_settings_conf_file(
open(settings.pcsd_settings_conf_location, "r").read()))
except EnvironmentError as e:
report_processor.report(
forceable_io_error_creator(
reports.file_io_error,
env_file_role_codes.PCS_SETTINGS_CONF,
file_path=settings.pcsd_settings_conf_location,
operation="read",
reason=format_environment_error(e)))
# stop here if one of the files could not be loaded and it was not forced
if report_processor.has_errors:
raise LibraryError()
if files_action:
com_cmd = DistributeFilesWithoutForces(env.report_processor,
files_action)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
# Distribute and reload pcsd SSL certificate
report_processor.report(
reports.pcsd_ssl_cert_and_key_distribution_started(
[target.label for target in new_nodes_target_list]))
try:
with open(settings.pcsd_cert_location, "r") as f:
ssl_cert = f.read()
except EnvironmentError as e:
report_processor.report(
reports.file_io_error(
env_file_role_codes.PCSD_SSL_CERT,
file_path=settings.pcsd_cert_location,
reason=format_environment_error(e),
operation="read",
))
try:
with open(settings.pcsd_key_location, "r") as f:
ssl_key = f.read()
except EnvironmentError as e:
report_processor.report(
reports.file_io_error(
env_file_role_codes.PCSD_SSL_KEY,
file_path=settings.pcsd_key_location,
reason=format_environment_error(e),
operation="read",
))
if report_processor.has_errors:
raise LibraryError()
com_cmd = SendPcsdSslCertAndKey(env.report_processor, ssl_cert, ssl_key)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
# When corosync >= 2 is in use, the procedure for adding a node is:
# 1. add the new node to corosync.conf on all existing nodes
# 2. reload corosync.conf before the new node is started
# 3. start the new node
# If done otherwise, membership gets broken and qdevice hangs. Cluster
# will recover after a minute or so but still it's a wrong way.
corosync_conf.add_nodes(new_nodes_corosync)
if atb_has_to_be_enabled:
corosync_conf.set_quorum_options(dict(auto_tie_breaker="1"))
com_cmd = DistributeCorosyncConf(
env.report_processor,
corosync_conf.config.export(),
allow_skip_offline=False,
)
com_cmd.set_targets(online_cluster_target_list + new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
com_cmd = ReloadCorosyncConf(env.report_processor)
com_cmd.set_targets(online_cluster_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
# Optionally enable and start cluster services.
if enable:
com_cmd = EnableCluster(env.report_processor)
com_cmd.set_targets(new_nodes_target_list)
run_and_raise(env.get_node_communicator(), com_cmd)
if start:
_start_cluster(
env.communicator_factory,
env.report_processor,
new_nodes_target_list,
wait_timeout=wait_timeout,
)