def setup_kafka_config(self):
'''
copy the default configuration files to kafka_conf property
defined in dist.yaml
'''
default_conf = self.dist_config.path('kafka') / 'config'
kafka_conf = self.dist_config.path('kafka_conf')
kafka_conf.rmtree_p()
default_conf.copytree(kafka_conf)
# Now remove the conf included in the tarball and symlink our real conf
# dir. we've seen issues where kafka still looks for config in
# KAFKA_HOME/config.
default_conf.rmtree_p()
kafka_conf.symlink(default_conf)
# Configure immutable bits
kafka_bin = self.dist_config.path('kafka') / 'bin'
with utils.environment_edit_in_place('/etc/environment') as env:
if kafka_bin not in env['PATH']:
env['PATH'] = ':'.join([env['PATH'], kafka_bin])
env['LOG_DIR'] = self.dist_config.path('kafka_app_logs')
# note: we set the advertised.host.name below to the public_address
# to ensure that external (non-Juju) clients can connect to Kafka
public_address = hookenv.unit_get('public-address')
private_ip = utils.resolve_private_address(
hookenv.unit_get('private-address'))
kafka_server_conf = self.dist_config.path(
'kafka_conf') / 'server.properties'
service, unit_num = os.environ['JUJU_UNIT_NAME'].split('/', 1)
utils.re_edit_in_place(
kafka_server_conf, {
r'^broker.id=.*':
'broker.id=%s' % unit_num,
r'^port=.*':
'port=%s' % self.dist_config.port('kafka'),
r'^log.dirs=.*':
'log.dirs=%s' % self.dist_config.path('kafka_data_logs'),
r'^#?advertised.host.name=.*':
'advertised.host.name=%s' % public_address,
})
kafka_log4j = self.dist_config.path('kafka_conf') / 'log4j.properties'
utils.re_edit_in_place(
kafka_log4j, {
r'^kafka.logs.dir=.*':
'kafka.logs.dir=%s' % self.dist_config.path('kafka_app_logs'),
})
# fix for lxc containers and some corner cases in manual provider
# ensure that public_address is resolvable internally by mapping it to the private IP
utils.update_kv_host(private_ip, public_address)
utils.manage_etc_hosts()
def trigger_puppet(self):
# If we can't reverse resolve the hostname (like on azure), support DN
# registration by IP address.
# NB: determine this *before* updating /etc/hosts below since
# gethostbyaddr will not fail if we have an /etc/hosts entry.
reverse_dns_bad = False
try:
socket.gethostbyaddr(utils.resolve_private_address(hookenv.unit_private_ip()))
except socket.herror:
reverse_dns_bad = True
# We know java7 has MAXHOSTNAMELEN of 64 char, so we cannot rely on
# java to do a hostname lookup on clouds that have >64 char fqdns
# (gce). Force short hostname (< 64 char) into /etc/hosts as workaround.
# Better fix may be to move to java8. See http://paste.ubuntu.com/16230171/
# NB: do this before the puppet apply, which may call java stuffs
# like format namenode, which will fail if we dont get this fix
# down early.
short_host = subprocess.check_output(['facter', 'hostname']).strip().decode()
private_ip = utils.resolve_private_address(hookenv.unit_private_ip())
if short_host and private_ip:
utils.update_kv_host(private_ip, short_host)
utils.manage_etc_hosts()
charm_dir = hookenv.charm_dir()
# TODO JIRA KWM: rm does not need Hdfs_init and will fail
rm_patch = Path(charm_dir) / 'resources/patch1_rm_init_hdfs.patch'
# TODO JIRA KWM: nm should not *need* mapred role. we could patch it
# with nm_patch, or adjust nm charm to include mapred role. for now,
# we're doing the latter. todo rfc from dev@bigtop list.
# nm_patch = Path(charm_dir) / 'resources/patch2_nm_core-site.patch'
# TODO JIRA KWM: client role needs common_yarn for yarn-site.xml
client_patch = Path(charm_dir) / 'resources/patch3_client_role_use_common_yarn.patch'
with chdir("{}".format(self.bigtop_base)):
# rm patch goes first
utils.run_as('root', 'patch', '-p1', '-s', '-i', rm_patch)
# skip nm_patch for now since nm charm is including mapred role
# utils.run_as('root', 'patch', '-p1', '-s', '-i', nm_patch)
# client patch goes last
utils.run_as('root', 'patch', '-p1', '-s', '-i', client_patch)
# TODO FIX ABOVE KWM
# puppet apply needs to be ran where recipes were unpacked
with chdir("{}".format(self.bigtop_base)):
utils.run_as('root', 'puppet', 'apply', '-d',
'--modulepath="bigtop-deploy/puppet/modules:/etc/puppet/modules"',
'bigtop-deploy/puppet/manifests/site.pp')
# Do any post-puppet config on the generated config files.
if reverse_dns_bad:
hdfs_site = Path('/etc/hadoop/conf/hdfs-site.xml')
with utils.xmlpropmap_edit_in_place(hdfs_site) as props:
props['dfs.namenode.datanode.registration.ip-hostname-check'] = 'false'
def trigger_puppet(self):
"""
Trigger Puppet to install the desired components.
"""
java_version = unitdata.kv().get('java_version', '')
if java_version.startswith('1.7.') and len(get_fqdn()) > 64:
# We know java7 has MAXHOSTNAMELEN of 64 char, so we cannot rely on
# java to do a hostname lookup on clouds that have >64 char FQDNs
# (e.g., gce). Attempt to work around this by putting the (hopefully
# short) hostname into /etc/hosts so that it will (hopefully) be
# used instead (see http://paste.ubuntu.com/16230171/).
# NB: do this before the puppet apply, which may call java stuffs
# like format namenode, which will fail if we dont get this fix
# down early.
short_host = subprocess.check_output(['facter', 'hostname']).strip().decode()
private_ip = utils.resolve_private_address(hookenv.unit_private_ip())
if short_host and private_ip:
utils.update_kv_host(private_ip, short_host)
utils.manage_etc_hosts()
# puppet args are bigtop-version depedent
if self.bigtop_version == '1.1.0':
puppet_args = [
'-d',
'--modulepath="bigtop-deploy/puppet/modules:/etc/puppet/modules"',
'bigtop-deploy/puppet/manifests/site.pp'
]
else:
puppet_args = [
'-d',
'--parser=future',
'--modulepath="bigtop-deploy/puppet/modules:/etc/puppet/modules"',
'bigtop-deploy/puppet/manifests'
]
# puppet apply runs from the root of the bigtop release source
with chdir(self.bigtop_base):
utils.run_as('root', 'puppet', 'apply', *puppet_args)
# Do any post-puppet config on the generated config files.
utils.re_edit_in_place('/etc/default/bigtop-utils', {
r'(# )?export JAVA_HOME.*': 'export JAVA_HOME={}'.format(
java_home()),
})
def trigger_puppet(self):
"""
Trigger Puppet to install the desired components.
"""
java_version = unitdata.kv().get('java_version', '')
if java_version.startswith('1.7.') and len(get_fqdn()) > 64:
# We know java7 has MAXHOSTNAMELEN of 64 char, so we cannot rely on
# java to do a hostname lookup on clouds that have >64 char FQDNs
# (e.g., gce). Attempt to work around this by putting the (hopefully
# short) hostname into /etc/hosts so that it will (hopefully) be
# used instead (see http://paste.ubuntu.com/16230171/).
# NB: do this before the puppet apply, which may call java stuffs
# like format namenode, which will fail if we dont get this fix
# down early.
short_host = subprocess.check_output(['facter', 'hostname']).strip().decode()
private_ip = utils.resolve_private_address(hookenv.unit_private_ip())
if short_host and private_ip:
utils.update_kv_host(private_ip, short_host)
utils.manage_etc_hosts()
# puppet args are bigtop-version depedent
if self.bigtop_version == '1.1.0':
puppet_args = [
'-d',
'--modulepath="bigtop-deploy/puppet/modules:/etc/puppet/modules"',
'bigtop-deploy/puppet/manifests/site.pp'
]
else:
puppet_args = [
'-d',
'--parser=future',
'--modulepath="bigtop-deploy/puppet/modules:/etc/puppet/modules"',
'bigtop-deploy/puppet/manifests'
]
# puppet apply runs from the root of the bigtop release source
with chdir(self.bigtop_base):
utils.run_as('root', 'puppet', 'apply', *puppet_args)
# Do any post-puppet config on the generated config files.
utils.re_edit_in_place('/etc/default/bigtop-utils', {
r'(# )?export JAVA_HOME.*': 'export JAVA_HOME={}'.format(
java_home()),
})
def setup_kafka_config(self):
'''
copy the default configuration files to kafka_conf property
defined in dist.yaml
'''
default_conf = self.dist_config.path('kafka') / 'config'
kafka_conf = self.dist_config.path('kafka_conf')
kafka_conf.rmtree_p()
default_conf.copytree(kafka_conf)
# Now remove the conf included in the tarball and symlink our real conf
# dir. we've seen issues where kafka still looks for config in
# KAFKA_HOME/config.
default_conf.rmtree_p()
kafka_conf.symlink(default_conf)
# Similarly, we've seen issues where kafka wants to write to
# KAFKA_HOME/logs regardless of the LOG_DIR, so make a symlink.
default_logs = self.dist_config.path('kafka') / 'logs'
kafka_logs = self.dist_config.path('kafka_app_logs')
default_logs.rmtree_p()
kafka_logs.symlink(default_logs)
# Configure environment
kafka_bin = self.dist_config.path('kafka') / 'bin'
with utils.environment_edit_in_place('/etc/environment') as env:
if kafka_bin not in env['PATH']:
env['PATH'] = ':'.join([env['PATH'], kafka_bin])
env['LOG_DIR'] = self.dist_config.path('kafka_app_logs')
# Configure server.properties
# note: we set the advertised.host.name below to the public_address
# to ensure that external (non-Juju) clients can connect to Kafka
public_address = hookenv.unit_get('public-address')
private_ip = utils.resolve_private_address(hookenv.unit_get('private-address'))
kafka_port = self.dist_config.port('kafka')
kafka_server_conf = self.dist_config.path('kafka_conf') / 'server.properties'
service, unit_num = os.environ['JUJU_UNIT_NAME'].split('/', 1)
utils.re_edit_in_place(kafka_server_conf, {
r'^broker.id=.*': 'broker.id=%s' % unit_num,
r'^port=.*': 'port=%s' % kafka_port,
r'^log.dirs=.*': 'log.dirs=%s' % self.dist_config.path('kafka_data_logs'),
r'^#?advertised.host.name=.*': 'advertised.host.name=%s' % public_address,
})
# Configure producer.properties
# note: we set the broker list host below to the public_address
# to ensure that external (non-Juju) clients can connect to Kafka.
# It must match our advertised.host.name from above.
kafka_producer_conf = self.dist_config.path('kafka_conf') / 'producer.properties'
utils.re_edit_in_place(kafka_producer_conf, {
r'^#?metadata.broker.list=.*': 'metadata.broker.list=%s:%s' % (public_address, kafka_port),
})
# Configure log properties
kafka_log4j = self.dist_config.path('kafka_conf') / 'log4j.properties'
utils.re_edit_in_place(kafka_log4j, {
r'^kafka.logs.dir=.*': 'kafka.logs.dir=%s' % self.dist_config.path('kafka_app_logs'),
})
template_name = 'upstart.conf'
template_path = '/etc/init/kafka.conf'
if host.init_is_systemd():
template_name = 'systemd.conf'
template_path = '/etc/systemd/system/kafka.service'
templating.render(
template_name,
template_path,
context={
'kafka_conf': self.dist_config.path('kafka_conf'),
'kafka_bin': '{}/bin'.format(self.dist_config.path('kafka'))
},
)
# fix for lxc containers and some corner cases in manual provider
# ensure that public_address is resolvable internally by mapping it to the private IP
utils.update_kv_host(private_ip, public_address)
utils.manage_etc_hosts()
def register_provided_hosts(self):
unit, data = any_ready_unit(self.relation_name)
provided_hosts = json.loads(data['etc_hosts'])
hookenv.log('Registering hosts from %s: %s' % (unit, provided_hosts))
for ip, name in provided_hosts.items():
utils.update_kv_host(ip, name)
def register_connected_hosts(self):
for unit, data in self.unfiltered_data().items():
ip = utils.resolve_private_address(data['private-address'])
name = unit.replace('/', '-')
utils.update_kv_host(ip, name)
def register_provided_hosts(self):
unit, data = any_ready_unit(self.relation_name)
provided_hosts = json.loads(data['etc_hosts'])
hookenv.log('Registering hosts from %s: %s' % (unit, provided_hosts))
for ip, name in provided_hosts.items():
utils.update_kv_host(ip, name)
def register_connected_hosts(self):
for unit, data in self.unfiltered_data().items():
ip = utils.resolve_private_address(data['private-address'])
name = unit.replace('/', '-')
utils.update_kv_host(ip, name)
