def assess_status():
'''Assess status of current unit'''
application_version_set(get_upstream_version(VERSION_PACKAGE))
if is_unit_upgrading_set():
status_set(
"blocked", "Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
# Check that the no-bootstrap config option is set in conjunction with
# having the bootstrap-source relation established
if not config('no-bootstrap') and is_relation_made('bootstrap-source'):
status_set(
'blocked', 'Cannot join the bootstrap-source relation when '
'no-bootstrap is False')
return
moncount = int(config('monitor-count'))
units = get_peer_units()
# not enough peers and mon_count > 1
if len(units.keys()) < moncount:
status_set(
'blocked', 'Insufficient peer units to bootstrap'
' cluster (require {})'.format(moncount))
return
# mon_count > 1, peers, but no ceph-public-address
ready = sum(1 for unit_ready in units.values() if unit_ready)
if ready < moncount:
status_set('waiting', 'Peer units detected, waiting for addresses')
return
configured_rbd_features = config('default-rbd-features')
if has_rbd_mirrors() and configured_rbd_features:
if add_rbd_mirror_features(
configured_rbd_features) != configured_rbd_features:
# The configured RBD features bitmap does not contain the features
# required for RBD Mirroring
status_set(
'blocked', 'Configuration mismatch: RBD Mirroring '
'enabled but incorrect value set for '
'``default-rbd-features``')
return
# active - bootstrapped + quorum status check
if ceph.is_bootstrapped() and ceph.is_quorum():
expected_osd_count = config('expected-osd-count') or 3
if sufficient_osds(expected_osd_count):
status_set('active', 'Unit is ready and clustered')
else:
status_set(
'waiting', 'Monitor bootstrapped but waiting for number of'
' OSDs to reach expected-osd-count ({})'.format(
expected_osd_count))
else:
# Unit should be running and clustered, but no quorum
# TODO: should this be blocked or waiting?
status_set('blocked', 'Unit not clustered (no quorum)')
def assess_status_helper():
"""Assess status of unit
@returns status, message - status is workload status and message is any
corresponding messages
"""
if config('stonith_enabled') in ['true', 'True', True]:
return ('blocked',
'stonith_enabled config option is no longer supported')
if is_unit_upgrading_set():
return ("blocked",
"Ready for do-release-upgrade. Set complete when finished")
if is_waiting_unit_series_upgrade_set():
return ("blocked",
"HA services shutdown, peers are ready for series upgrade")
if is_unit_paused_set():
return ("maintenance",
"Paused. Use 'resume' action to resume normal service.")
node_count = int(config('cluster_count'))
status = 'active'
message = 'Unit is ready and clustered'
try:
try_pcmk_wait()
except pcmk.ServicesNotUp:
message = 'Pacemaker is down'
status = 'blocked'
for relid in relation_ids('hanode'):
if len(related_units(relid)) + 1 < node_count:
status = 'blocked'
message = ("Insufficient peer units for ha cluster "
"(require {})".format(node_count))
# if the status was not changed earlier, we verify the maintenance status
try:
if status == 'active':
prop = pcmk.get_property('maintenance-mode').strip()
except pcmk.PropertyNotFound:
# the property is not the output of 'crm configure show xml', so we use
# the default value for this property. For crmsh>=2.2.0 the default
# value is automatically provided by show-property or get-property.
prop = 'false'
if (status == 'active' and prop == 'true'):
# maintenance mode enabled in pacemaker
status = 'maintenance'
message = 'Pacemaker in maintenance mode'
for resource in get_resources().keys():
if not pcmk.is_resource_present(resource):
return ("waiting",
"Resource: {} not yet configured".format(resource))
if not pcmk.crm_res_running_on_node(resource, get_hostname()):
return ("blocked", "Resource: {} not running".format(resource))
return status, message
def assess_status():
"""Assess status of current unit"""
# check to see if the unit is paused.
application_version_set(get_upstream_version(VERSION_PACKAGE))
if is_unit_upgrading_set():
status_set("blocked",
"Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
if is_unit_paused_set():
status_set('maintenance',
"Paused. Use 'resume' action to resume normal service.")
return
# Check for mon relation
if len(relation_ids('mon')) < 1:
status_set('blocked', 'Missing relation: monitor')
return
# Check for monitors with presented addresses
# Check for bootstrap key presentation
monitors = get_mon_hosts()
if len(monitors) < 1 or not get_conf('osd_bootstrap_key'):
status_set('waiting', 'Incomplete relation: monitor')
return
# Check for vault
if use_vaultlocker():
if not relation_ids('secrets-storage'):
status_set('blocked', 'Missing relation: vault')
return
if not vaultlocker.vault_relation_complete():
status_set('waiting', 'Incomplete relation: vault')
return
# Check for OSD device creation parity i.e. at least some devices
# must have been presented and used for this charm to be operational
(prev_status, prev_message) = status_get()
running_osds = ceph.get_running_osds()
if not prev_message.startswith('Non-pristine'):
if not running_osds:
status_set('blocked',
'No block devices detected using current configuration')
else:
status_set('active',
'Unit is ready ({} OSD)'.format(len(running_osds)))
else:
pristine = True
osd_journals = get_journal_devices()
for dev in list(set(ceph.unmounted_disks()) - set(osd_journals)):
if (not ceph.is_active_bluestore_device(dev) and
not ceph.is_pristine_disk(dev)):
pristine = False
break
if pristine:
status_set('active',
'Unit is ready ({} OSD)'.format(len(running_osds)))
def assess_status():
'''Determine status of current unit'''
if is_unit_upgrading_set():
status_set('blocked',
'Ready for do-release-upgrade and reboot. '
'Set complete when finished.')
elif lxd_running():
status_set('active', 'Unit is ready')
else:
status_set('blocked', 'LXD is not running')
application_version_set(get_upstream_version(VERSION_PACKAGE))
def assess_status(self):
"""Determine the current application status for the charm"""
hookenv.application_version_set(self.application_version)
if not self.configuration_complete():
hookenv.status_set('blocked', 'LDAP configuration incomplete')
elif os_utils.is_unit_upgrading_set():
hookenv.status_set(
'blocked', 'Ready for do-release-upgrade and reboot. '
'Set complete when finished.')
else:
hookenv.status_set('active', 'Unit is ready')
def assess_status():
'''Determine status of current unit'''
if is_unit_upgrading_set():
status_set(
'blocked', 'Ready for do-release-upgrade and reboot. '
'Set complete when finished.')
elif lxd_running():
status_set('active', 'Unit is ready')
else:
status_set('blocked', 'LXD is not running')
application_version_set(get_upstream_version(VERSION_PACKAGE))
def assess_status():
'''Assess status of current unit'''
if is_unit_upgrading_set():
status_set("blocked",
"Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
if ready():
status_set('active', 'Ready to proxy settings')
else:
status_set('blocked', 'Ensure FSID and admin-key are set')
def assess_status():
"""Assess status of current unit"""
# check to see if the unit is paused.
application_version_set(get_upstream_version(VERSION_PACKAGE))
if is_unit_upgrading_set():
status_set(
"blocked", "Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
if is_unit_paused_set():
status_set('maintenance',
"Paused. Use 'resume' action to resume normal service.")
return
# Check for mon relation
if len(relation_ids('mon')) < 1:
status_set('blocked', 'Missing relation: monitor')
return
# Check for monitors with presented addresses
# Check for bootstrap key presentation
monitors = get_mon_hosts()
if len(monitors) < 1 or not get_conf('osd_bootstrap_key'):
status_set('waiting', 'Incomplete relation: monitor')
return
# Check for vault
if use_vaultlocker():
if not relation_ids('secrets-storage'):
status_set('blocked', 'Missing relation: vault')
return
if not vaultlocker.vault_relation_complete():
status_set('waiting', 'Incomplete relation: vault')
return
# Check for OSD device creation parity i.e. at least some devices
# must have been presented and used for this charm to be operational
(prev_status, prev_message) = status_get()
running_osds = ceph.get_running_osds()
if not prev_message.startswith('Non-pristine'):
if not running_osds:
status_set(
'blocked',
'No block devices detected using current configuration')
else:
status_set('active',
'Unit is ready ({} OSD)'.format(len(running_osds)))
def charm_check_func():
"""Custom function to assess the status of the current unit
@returns (status, message) - tuple of strings if an issue
"""
if is_unit_upgrading_set():
# Avoid looping through attempting to determine cluster_in_sync
return ("blocked", "Unit upgrading.")
@retry_on_exception(num_retries=10,
base_delay=2,
exc_type=DesyncedException)
def _cluster_in_sync():
'''Helper func to wait for a while for resync to occur
@raise DesynedException: raised if local unit is not in sync
with its peers
'''
if not cluster_in_sync():
raise DesyncedException()
min_size = config('min-cluster-size')
# Ensure that number of peers > cluster size configuration
if not is_sufficient_peers():
return ('blocked', 'Insufficient peers to bootstrap cluster')
if min_size and int(min_size) > 1:
# Once running, ensure that cluster is in sync
# and has the required peers
if not is_bootstrapped():
return ('waiting', 'Unit waiting for cluster bootstrap')
elif cluster_ready():
try:
_cluster_in_sync()
return ('active', 'Unit is ready and clustered')
except DesyncedException:
return ('blocked', 'Unit is not in sync')
else:
return ('waiting', 'Unit waiting on hacluster relation')
else:
if seeded():
return ('active', 'Unit is ready')
else:
return ('waiting', 'Unit waiting to bootstrap')
def assess_status_helper():
"""Assess status of unit
@returns status, message - status is workload status and message is any
corresponding messages
"""
if is_unit_upgrading_set():
return ("blocked",
"Ready for do-release-upgrde. Set complete when finished")
if is_unit_paused_set():
return ("maintenance",
"Paused. Use 'resume' action to resume normal service.")
node_count = int(config('cluster_count'))
status = 'active'
message = 'Unit is ready and clustered'
try:
try_pcmk_wait()
except pcmk.ServicesNotUp:
message = 'Pacemaker is down'
status = 'blocked'
for relid in relation_ids('hanode'):
if len(related_units(relid)) + 1 < node_count:
status = 'blocked'
message = ("Insufficient peer units for ha cluster "
"(require {})".format(node_count))
# if the status was not changed earlier, we verify the maintenance status
try:
if status == 'active':
prop = pcmk.get_property('maintenance-mode').strip()
except pcmk.PropertyNotFound:
# the property is not the output of 'crm configure show xml', so we use
# the default value for this property. For crmsh>=2.2.0 the default
# value is automatically provided by show-property or get-property.
prop = 'false'
if (status == 'active' and prop == 'true'):
# maintenance mode enabled in pacemaker
status = 'maintenance'
message = 'Pacemaker in maintenance mode'
return status, message
def assess_status():
'''Assess status of current unit'''
application_version_set(get_upstream_version(VERSION_PACKAGE))
if is_unit_upgrading_set():
status_set(
"blocked", "Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
# Check that the no-bootstrap config option is set in conjunction with
# having the bootstrap-source relation established
if not config('no-bootstrap') and is_relation_made('bootstrap-source'):
status_set(
'blocked', 'Cannot join the bootstrap-source relation when '
'no-bootstrap is False')
return
moncount = int(config('monitor-count'))
units = get_peer_units()
# not enough peers and mon_count > 1
if len(units.keys()) < moncount:
status_set(
'blocked', 'Insufficient peer units to bootstrap'
' cluster (require {})'.format(moncount))
return
# mon_count > 1, peers, but no ceph-public-address
ready = sum(1 for unit_ready in units.values() if unit_ready)
if ready < moncount:
status_set('waiting', 'Peer units detected, waiting for addresses')
return
# active - bootstrapped + quorum status check
if ceph.is_bootstrapped() and ceph.is_quorum():
status_set('active', 'Unit is ready and clustered')
else:
# Unit should be running and clustered, but no quorum
# TODO: should this be blocked or waiting?
status_set('blocked', 'Unit not clustered (no quorum)')
def pause_unit():
"""Pause services on this unit and update the units status
@returns None
"""
node_name = get_hostname()
messages = []
enter_standby_mode(node_name)
if not is_in_standby_mode(node_name):
messages.append("Node not in standby mode")
# some resources may take some time to be migrated out from the node. So 3
# retries are made with a 5 seconds wait between each one.
i = 0
ready = False
has_resources = False
while i < PCMKR_MAX_RETRIES and not ready:
if node_has_resources(node_name):
has_resources = True
i += 1
time.sleep(PCMKR_SLEEP_SECS)
else:
ready = True
has_resources = False
if has_resources:
messages.append("Resources still running on unit")
status, message = assess_status_helper()
# New status message will indicate the resource is not running
if status != 'active' and 'not running' not in message:
messages.append(message)
if messages and not is_unit_upgrading_set():
raise Exception("Couldn't pause: {}".format("; ".join(messages)))
else:
set_unit_paused()
status_set("maintenance",
"Paused. Use 'resume' action to resume normal service.")
def upgrade():
if is_leader():
if is_unit_paused_set() or is_unit_upgrading_set():
log('Unit is paused, skiping upgrade', level=INFO)
return
# Leader sets on upgrade
leader_set(**{'leader-ip': get_relation_ip('cluster')})
configure_sstuser(sst_password())
if not leader_get('root-password') and leader_get('mysql.passwd'):
leader_set(**{'root-password': leader_get('mysql.passwd')})
# On upgrade-charm we assume the cluster was complete at some point
kvstore = kv()
initial_clustered = kvstore.get(INITIAL_CLUSTERED_KEY, False)
if not initial_clustered:
kvstore.set(key=INITIAL_CLUSTERED_KEY, value=True)
kvstore.flush()
# broadcast the bootstrap-uuid
wsrep_ready = get_wsrep_value('wsrep_ready') or ""
if wsrep_ready.lower() in ['on', 'ready']:
cluster_state_uuid = get_wsrep_value('wsrep_cluster_state_uuid')
if cluster_state_uuid:
mark_seeded()
notify_bootstrapped(cluster_uuid=cluster_state_uuid)
else:
# Ensure all the peers have the bootstrap-uuid attribute set
# as this is all happening during the upgrade-charm hook is reasonable
# to expect the cluster is running.
# Wait until the leader has set the
try:
update_bootstrap_uuid()
except LeaderNoBootstrapUUIDError:
status_set('waiting', "Waiting for bootstrap-uuid set by leader")
def upgrade():
if is_leader():
if is_unit_paused_set() or is_unit_upgrading_set():
log('Unit is paused, skiping upgrade', level=INFO)
return
# Leader sets on upgrade
leader_set(**{'leader-ip': get_relation_ip('cluster')})
configure_sstuser(sst_password())
if not leader_get('root-password') and leader_get('mysql.passwd'):
leader_set(**{'root-password': leader_get('mysql.passwd')})
# On upgrade-charm we assume the cluster was complete at some point
kvstore = kv()
initial_clustered = kvstore.get(INITIAL_CLUSTERED_KEY, False)
if not initial_clustered:
kvstore.set(key=INITIAL_CLUSTERED_KEY, value=True)
kvstore.flush()
# broadcast the bootstrap-uuid
wsrep_ready = get_wsrep_value('wsrep_ready') or ""
if wsrep_ready.lower() in ['on', 'ready']:
cluster_state_uuid = get_wsrep_value('wsrep_cluster_state_uuid')
if cluster_state_uuid:
mark_seeded()
notify_bootstrapped(cluster_uuid=cluster_state_uuid)
else:
# Ensure all the peers have the bootstrap-uuid attribute set
# as this is all happening during the upgrade-charm hook is reasonable
# to expect the cluster is running.
# Wait until the leader has set the
try:
update_bootstrap_uuid()
except LeaderNoBootstrapUUIDError:
status_set('waiting', "Waiting for bootstrap-uuid set by leader")
def pause_unit():
"""Pause services on this unit and update the units status
@returns None
"""
node_name = get_hostname()
messages = []
enter_standby_mode(node_name)
if not is_in_standby_mode(node_name):
messages.append("Node not in standby mode")
# some resources may take some time to be migrated out from the node. So 3
# retries are made with a 5 seconds wait between each one.
i = 0
ready = False
has_resources = False
while i < PCMKR_MAX_RETRIES and not ready:
if node_has_resources(node_name):
has_resources = True
i += 1
time.sleep(PCMKR_SLEEP_SECS)
else:
ready = True
has_resources = False
if has_resources:
messages.append("Resources still running on unit")
status, message = assess_status_helper()
if status != 'active':
messages.append(message)
if messages and not is_unit_upgrading_set():
raise Exception("Couldn't pause: {}".format("; ".join(messages)))
else:
set_unit_paused()
status_set("maintenance",
"Paused. Use 'resume' action to resume normal service.")
def assess_status():
"""Assess status of current unit"""
# check to see if the unit is paused.
application_version_set(get_upstream_version(VERSION_PACKAGE))
if is_unit_upgrading_set():
status_set(
"blocked", "Ready for do-release-upgrade and reboot. "
"Set complete when finished.")
return
if is_unit_paused_set():
status_set('maintenance',
"Paused. Use 'resume' action to resume normal service.")
return
# Check for mon relation
if len(relation_ids('mon')) < 1:
status_set('blocked', 'Missing relation: monitor')
return
# Check for monitors with presented addresses
# Check for bootstrap key presentation
monitors = get_mon_hosts()
if len(monitors) < 1 or not get_conf('osd_bootstrap_key'):
status_set('waiting', 'Incomplete relation: monitor')
return
# Check for vault
if use_vaultlocker():
if not relation_ids('secrets-storage'):
status_set('blocked', 'Missing relation: vault')
return
try:
if not vaultlocker.vault_relation_complete():
status_set('waiting', 'Incomplete relation: vault')
return
except Exception as e:
status_set('blocked', "Warning: couldn't verify vault relation")
log("Exception when verifying vault relation - maybe it was "
"offline?:\n{}".format(str(e)))
log("Traceback: {}".format(traceback.format_exc()))
# Check for OSD device creation parity i.e. at least some devices
# must have been presented and used for this charm to be operational
(prev_status, prev_message) = status_get()
running_osds = ceph.get_running_osds()
if not prev_message.startswith('Non-pristine'):
if not running_osds:
status_set(
'blocked',
'No block devices detected using current configuration')
else:
status_set('active',
'Unit is ready ({} OSD)'.format(len(running_osds)))
else:
pristine = True
osd_journals = get_journal_devices()
for dev in list(set(ceph.unmounted_disks()) - set(osd_journals)):
if (not ceph.is_active_bluestore_device(dev)
and not ceph.is_pristine_disk(dev)):
pristine = False
break
if pristine:
status_set('active',
'Unit is ready ({} OSD)'.format(len(running_osds)))
try:
get_bdev_enable_discard()
except ValueError as ex:
status_set('blocked', str(ex))
try:
bluestore_compression = ch_context.CephBlueStoreCompressionContext()
bluestore_compression.validate()
except ValueError as e:
status_set('blocked', 'Invalid configuration: {}'.format(str(e)))
def config_changed():
# if we are paused or upgrading, delay doing any config changed hooks.
# It is forced on the resume.
if is_unit_paused_set() or is_unit_upgrading_set():
log("Unit is paused or upgrading. Skipping config_changed", "WARN")
return
# It is critical that the installation is attempted first before any
# rendering of the configuration files occurs.
# install_percona_xtradb_cluster has the code to decide if this is the
# leader or if the leader is bootstrapped and therefore ready for install.
install_percona_xtradb_cluster()
if config('prefer-ipv6'):
assert_charm_supports_ipv6()
hosts = get_cluster_hosts()
leader_bootstrapped = is_leader_bootstrapped()
leader_ip = leader_get('leader-ip')
# Cluster upgrade adds some complication
cluster_series_upgrading = leader_get("cluster_series_upgrading")
if cluster_series_upgrading:
leader = (leader_get('cluster_series_upgrade_leader') ==
get_relation_ip('cluster'))
leader_ip = leader_get('cluster_series_upgrade_leader')
else:
leader = is_leader()
leader_ip = leader_get('leader-ip')
# (re)install pcmkr agent
install_mysql_ocf()
if leader:
# If the cluster has not been fully bootstrapped once yet, use an empty
# hosts list to avoid restarting the leader node's mysqld during
# cluster buildup.
# After, the cluster has bootstrapped at least one time, it is much
# less likely to have restart collisions. It is then safe to use the
# full hosts list and have the leader node's mysqld restart.
# Empty hosts if cluster_series_upgrading
if not clustered_once() or cluster_series_upgrading:
hosts = []
log(
"Leader unit - bootstrap required={}".format(
not leader_bootstrapped), DEBUG)
render_config_restart_on_changed(hosts,
bootstrap=not leader_bootstrapped)
elif (leader_bootstrapped and is_sufficient_peers()
and not cluster_series_upgrading):
# Skip if cluster_series_upgrading
# Speed up cluster process by bootstrapping when the leader has
# bootstrapped if we have expected number of peers
# However, in a cold boot scenario do not add the "old" leader
# when it matches this host.
if (leader_ip not in hosts and leader_ip != get_cluster_host_ip()):
# Fix Bug #1738896
hosts = [leader_ip] + hosts
log("Leader is bootstrapped - configuring mysql on this node", DEBUG)
# Rendering the mysqld.cnf and restarting is bootstrapping for a
# non-leader node.
render_config_restart_on_changed(hosts)
# Assert we are bootstrapped. This will throw an
# InconsistentUUIDError exception if UUIDs do not match.
update_bootstrap_uuid()
else:
# Until the bootstrap-uuid attribute is set by the leader,
# cluster_ready() will evaluate to False. So it is necessary to
# feed this information to the user.
status_set('waiting', "Waiting for bootstrap-uuid set by leader")
log('Non-leader waiting on leader bootstrap, skipping render', DEBUG)
return
# Notify any changes to the access network
update_client_db_relations()
for rid in relation_ids('ha'):
# make sure all the HA resources are (re)created
ha_relation_joined(relation_id=rid)
if is_relation_made('nrpe-external-master'):
update_nrpe_config()
open_port(DEFAULT_MYSQL_PORT)
# the password needs to be updated only if the node was already
# bootstrapped
if is_bootstrapped():
if is_leader():
update_root_password()
set_ready_on_peers()
# NOTE(tkurek): re-set 'master' relation data
if relation_ids('master'):
master_joined()
def render_config(hosts=None):
if hosts is None:
hosts = []
config_file = resolve_cnf_file()
if not os.path.exists(os.path.dirname(config_file)):
os.makedirs(os.path.dirname(config_file))
context = {
'cluster_name': 'juju_cluster',
'private_address': get_cluster_host_ip(),
'cluster_hosts': ",".join(hosts),
'sst_method': config('sst-method'),
'sst_password': sst_password(),
'innodb_file_per_table': config('innodb-file-per-table'),
'table_open_cache': config('table-open-cache'),
'binlogs_path': config('binlogs-path'),
'enable_binlogs': config('enable-binlogs'),
'binlogs_max_size': config('binlogs-max-size'),
'binlogs_expire_days': config('binlogs-expire-days'),
'performance_schema': config('performance-schema'),
'is_leader': is_leader(),
'server_id': get_server_id(),
'series_upgrade': is_unit_upgrading_set(),
}
if config('prefer-ipv6'):
# NOTE(hopem): this is a kludge to get percona working with ipv6.
# See lp 1380747 for more info. This is intended as a stop gap until
# percona package is fixed to support ipv6.
context['bind_address'] = '::'
context['ipv6'] = True
else:
context['ipv6'] = False
wsrep_provider_options = get_wsrep_provider_options()
if wsrep_provider_options:
context['wsrep_provider_options'] = wsrep_provider_options
if config('wsrep-slave-threads') is not None:
context['wsrep_slave_threads'] = config('wsrep-slave-threads')
if CompareHostReleases(lsb_release()['DISTRIB_CODENAME']) < 'bionic':
# myisam_recover is not valid for PXC 5.7 (introduced in Bionic) so we
# only set it for PXC 5.6.
context['myisam_recover'] = 'BACKUP'
context['wsrep_provider'] = '/usr/lib/libgalera_smm.so'
if 'wsrep_slave_threads' not in context:
context['wsrep_slave_threads'] = 1
elif CompareHostReleases(lsb_release()['DISTRIB_CODENAME']) >= 'bionic':
context['wsrep_provider'] = '/usr/lib/galera3/libgalera_smm.so'
context['default_storage_engine'] = 'InnoDB'
context['wsrep_log_conflicts'] = True
context['innodb_autoinc_lock_mode'] = '2'
context['pxc_strict_mode'] = config('pxc-strict-mode')
if 'wsrep_slave_threads' not in context:
context['wsrep_slave_threads'] = 48
if config('databases-to-replicate'):
context['databases_to_replicate'] = get_databases_to_replicate()
context['server-id'] = get_server_id()
context.update(PerconaClusterHelper().parse_config())
render(os.path.basename(config_file), config_file, context, perms=0o444)
def render_config(hosts=None):
if hosts is None:
hosts = []
config_file = resolve_cnf_file()
if not os.path.exists(os.path.dirname(config_file)):
os.makedirs(os.path.dirname(config_file))
context = {
'cluster_name': 'juju_cluster',
'private_address': get_cluster_host_ip(),
'cluster_hosts': ",".join(hosts),
'sst_method': config('sst-method'),
'sst_password': sst_password(),
'innodb_file_per_table': config('innodb-file-per-table'),
'table_open_cache': config('table-open-cache'),
'binlogs_path': config('binlogs-path'),
'enable_binlogs': config('enable-binlogs'),
'binlogs_max_size': config('binlogs-max-size'),
'binlogs_expire_days': config('binlogs-expire-days'),
'performance_schema': config('performance-schema'),
'is_leader': is_leader(),
'server_id': get_server_id(),
'series_upgrade': is_unit_upgrading_set(),
}
if config('prefer-ipv6'):
# NOTE(hopem): this is a kludge to get percona working with ipv6.
# See lp 1380747 for more info. This is intended as a stop gap until
# percona package is fixed to support ipv6.
context['bind_address'] = '::'
context['ipv6'] = True
else:
context['ipv6'] = False
wsrep_provider_options = get_wsrep_provider_options()
if wsrep_provider_options:
context['wsrep_provider_options'] = wsrep_provider_options
if config('wsrep-slave-threads') is not None:
context['wsrep_slave_threads'] = config('wsrep-slave-threads')
if CompareHostReleases(lsb_release()['DISTRIB_CODENAME']) < 'bionic':
# myisam_recover is not valid for PXC 5.7 (introduced in Bionic) so we
# only set it for PXC 5.6.
context['myisam_recover'] = 'BACKUP'
context['wsrep_provider'] = '/usr/lib/libgalera_smm.so'
if 'wsrep_slave_threads' not in context:
context['wsrep_slave_threads'] = 1
elif CompareHostReleases(lsb_release()['DISTRIB_CODENAME']) >= 'bionic':
context['wsrep_provider'] = '/usr/lib/galera3/libgalera_smm.so'
context['default_storage_engine'] = 'InnoDB'
context['wsrep_log_conflicts'] = True
context['innodb_autoinc_lock_mode'] = '2'
context['pxc_strict_mode'] = config('pxc-strict-mode')
if 'wsrep_slave_threads' not in context:
context['wsrep_slave_threads'] = 48
if config('databases-to-replicate'):
context['databases_to_replicate'] = get_databases_to_replicate()
context['server-id'] = get_server_id()
context.update(PerconaClusterHelper().parse_config())
render(os.path.basename(config_file), config_file, context, perms=0o444)
def config_changed():
# if we are paused or upgrading, delay doing any config changed hooks.
# It is forced on the resume.
if is_unit_paused_set() or is_unit_upgrading_set():
log("Unit is paused or upgrading. Skipping config_changed", "WARN")
return
# It is critical that the installation is attempted first before any
# rendering of the configuration files occurs.
# install_percona_xtradb_cluster has the code to decide if this is the
# leader or if the leader is bootstrapped and therefore ready for install.
install_percona_xtradb_cluster()
if config('prefer-ipv6'):
assert_charm_supports_ipv6()
hosts = get_cluster_hosts()
leader_bootstrapped = is_leader_bootstrapped()
leader_ip = leader_get('leader-ip')
# Cluster upgrade adds some complication
cluster_series_upgrading = leader_get("cluster_series_upgrading")
if cluster_series_upgrading:
leader = (leader_get('cluster_series_upgrade_leader') ==
get_relation_ip('cluster'))
leader_ip = leader_get('cluster_series_upgrade_leader')
else:
leader = is_leader()
leader_ip = leader_get('leader-ip')
if leader:
# If the cluster has not been fully bootstrapped once yet, use an empty
# hosts list to avoid restarting the leader node's mysqld during
# cluster buildup.
# After, the cluster has bootstrapped at least one time, it is much
# less likely to have restart collisions. It is then safe to use the
# full hosts list and have the leader node's mysqld restart.
# Empty hosts if cluster_series_upgrading
if not clustered_once() or cluster_series_upgrading:
hosts = []
log("Leader unit - bootstrap required=%s" % (not leader_bootstrapped),
DEBUG)
render_config_restart_on_changed(hosts,
bootstrap=not leader_bootstrapped)
elif (leader_bootstrapped and
is_sufficient_peers() and not
cluster_series_upgrading):
# Skip if cluster_series_upgrading
# Speed up cluster process by bootstrapping when the leader has
# bootstrapped if we have expected number of peers
if leader_ip not in hosts:
# Fix Bug #1738896
hosts = [leader_ip] + hosts
log("Leader is bootstrapped - configuring mysql on this node",
DEBUG)
# Rendering the mysqld.cnf and restarting is bootstrapping for a
# non-leader node.
render_config_restart_on_changed(hosts)
# Assert we are bootstrapped. This will throw an
# InconsistentUUIDError exception if UUIDs do not match.
update_bootstrap_uuid()
else:
# Until the bootstrap-uuid attribute is set by the leader,
# cluster_ready() will evaluate to False. So it is necessary to
# feed this information to the user.
status_set('waiting', "Waiting for bootstrap-uuid set by leader")
log('Non-leader waiting on leader bootstrap, skipping render',
DEBUG)
return
# Notify any changes to the access network
update_client_db_relations()
# (re)install pcmkr agent
install_mysql_ocf()
for rid in relation_ids('ha'):
# make sure all the HA resources are (re)created
ha_relation_joined(relation_id=rid)
if is_relation_made('nrpe-external-master'):
update_nrpe_config()
open_port(DEFAULT_MYSQL_PORT)
# the password needs to be updated only if the node was already
# bootstrapped
if is_bootstrapped():
update_root_password()
set_ready_on_peers()
# NOTE(tkurek): re-set 'master' relation data
if relation_ids('master'):
master_joined()
