def main(args):
quiet = not args.verbose
print_topology_meta(args.topology)
if args.include_services and args.exclude_services:
raise ValueError(
'Cannot pass both --include-services and --exclude-services.')
image_prefix = '{}/{}/topology_hdp:hdp{}_ambari{}'.format(
args.registry, args.namespace or DEFAULT_NAMESPACE, args.hdp_version,
args.ambari_version)
primary_node_image = '{}_{}'.format(image_prefix, 'primary-node')
secondary_node_image = '{}_{}'.format(image_prefix, 'secondary-node')
clusterdock_config_host_dir = os.path.realpath(
os.path.expanduser(args.clusterdock_config_directory))
volumes = [{clusterdock_config_host_dir: CLUSTERDOCK_CLIENT_CONTAINER_DIR}]
primary_node = Node(hostname=args.primary_node[0],
group='primary',
volumes=volumes,
image=primary_node_image,
ports=[{
AMBARI_PORT: AMBARI_PORT
} if args.predictable else AMBARI_PORT])
secondary_nodes = [
Node(hostname=hostname,
group='secondary',
volumes=volumes,
image=secondary_node_image) for hostname in args.secondary_nodes
]
cluster = Cluster(primary_node, *secondary_nodes)
cluster.primary_node = primary_node
cluster.secondary_nodes = secondary_nodes
for node in cluster.nodes:
node.volumes.append({'/sys/fs/cgroup': '/sys/fs/cgroup'})
# do not use tempfile.mkdtemp, as systemd wont be able to bring services up when temp ends to be created in
# /var/tmp/ directory
node.volumes.append(['/run', '/run/lock'])
cluster.start(args.network)
hdp_version_tuple = version_tuple(args.hdp_version)
logger.debug('Starting PostgreSQL for Ambari server ...')
# Need this as init system in Docker misreports on postgres start initially
# Check https://github.com/docker-library/postgres/issues/146 for more
def condition():
primary_node.execute('service postgresql restart', quiet=quiet)
if '1 row' in primary_node.execute(
'PGPASSWORD=bigdata psql ambari '
'-U ambari -h localhost -c "select 1"',
quiet=quiet).output:
return True
wait_for_condition(condition=condition, time_between_checks=2)
def condition():
if 'running' in primary_node.execute('service postgresql status',
quiet=quiet).output:
return True
wait_for_condition(condition=condition)
time.sleep(
10
) # If images are set to start Ambari server/agents - give some time to recover the right status
_update_node_names(cluster, quiet=quiet)
# The HDP topology uses two pre-built images ('primary' and 'secondary'). If a cluster
# larger than 2 nodes is started, some modifications need to be done.
if len(secondary_nodes) > 1:
_remove_files(nodes=secondary_nodes[1:],
files=['/hadoop/hdfs/data/current/*'],
quiet=quiet)
logger.info('Starting Ambari server ...')
primary_node.execute('ambari-server start', quiet=quiet)
# Docker for Mac exposes ports that can be accessed only with ``localhost:<port>`` so
# use that instead of the hostname if the host name is ``moby``.
hostname = ('localhost'
if client.info().get('Name') == 'moby' else socket.getaddrinfo(
socket.gethostname(), 0, flags=socket.AI_CANONNAME)[0][3])
port = cluster.primary_node.host_ports.get(AMBARI_PORT)
server_url = 'http://{}:{}'.format(hostname, port)
logger.info('Ambari server is now reachable at %s', server_url)
logger.info('Starting Ambari agents ...')
for node in cluster:
logger.debug('Starting Ambari agent on %s ...', node.fqdn)
node.execute('ambari-agent start', quiet=quiet)
ambari = Ambari(server_url, username='******', password='******')
def condition(ambari, cluster):
cluster_hosts = {node.fqdn for node in cluster}
ambari_hosts = {host.host_name for host in ambari.hosts}
logger.debug('Cluster hosts: %s; Ambari hosts: %s', cluster_hosts,
ambari_hosts)
return cluster_hosts == ambari_hosts
wait_for_condition(condition=condition, condition_args=[ambari, cluster])
service_types_to_leave = (args.include_services.upper().split(',')
if args.include_services else [])
service_types_to_remove = (args.exclude_services.upper().split(',')
if args.exclude_services else [])
if service_types_to_leave or service_types_to_remove:
for service in list(ambari.clusters(DEFAULT_CLUSTER_NAME).services):
service_name = service.service_name.upper()
if (service_name in service_types_to_remove
or (service_types_to_leave
and service_name not in service_types_to_leave)):
logger.info('Removing cluster service (name = %s) ...',
service_name)
service.delete()
for node in secondary_nodes[1:]:
logger.info('Adding %s to cluster ...', node.fqdn)
ambari.clusters(DEFAULT_CLUSTER_NAME).hosts.create(node.fqdn)
secondary_node = ambari.clusters(DEFAULT_CLUSTER_NAME).hosts(
secondary_nodes[0].fqdn)
for component in secondary_node.components:
logger.debug('Adding component (%s) to cluster on host (%s) ...',
component.component_name, node.fqdn)
host_components = ambari.clusters(DEFAULT_CLUSTER_NAME).hosts(
node.fqdn).components
host_components.create(component.component_name).wait()
logger.debug('Installing all registered components on host (%s) ...',
node.fqdn)
ambari.clusters(DEFAULT_CLUSTER_NAME).hosts(
node.fqdn).components.install().wait()
logger.info('Waiting for all hosts to reach healthy state ...')
def condition(ambari):
health_report = ambari.clusters(DEFAULT_CLUSTER_NAME).health_report
logger.debug('Ambari cluster health report: %s ...', health_report)
return health_report.get('Host/host_state/HEALTHY') == len(
list(ambari.hosts))
wait_for_condition(condition=condition, condition_args=[ambari])
service_names = [
service['service_name'] for service in ambari.clusters(
DEFAULT_CLUSTER_NAME).services.to_dict()
]
if 'ATLAS' in service_names:
logger.info('Configuring Atlas required properties ...')
_configure_atlas(ambari,
args.hdp_version,
atlas_server_host=cluster.primary_node.fqdn)
if 'HIVE' in service_names:
primary_node.execute('touch /etc/hive/sys.db.created', quiet=quiet)
logger.info('Waiting for components to be ready ...')
def condition(ambari):
comps = ambari.clusters(
DEFAULT_CLUSTER_NAME).cluster.host_components.refresh()
for comp in comps:
if comp.state.upper() == 'UNKNOWN':
logger.debug('Not ready with component `%s` ...',
comp.component_name)
return False
else:
return True
wait_for_condition(condition=condition, condition_args=[ambari])
if not args.dont_start_cluster:
logger.info('Starting cluster services ...')
ambari.clusters(DEFAULT_CLUSTER_NAME).services.start().wait(
timeout=3600)
if 'HBASE' in service_names:
logger.info('Starting Thrift server ...')
if hdp_version_tuple <= (2, 0, 13, 0):
hbase_daemon_path = '/usr/lib/hbase/bin/hbase-daemon.sh'
else:
hbase_daemon_path = '/usr/hdp/current/hbase-master/bin/hbase-daemon.sh'
primary_node.execute('{} start thrift -p {} '
'--infoport {}'.format(
hbase_daemon_path,
HBASE_THRIFT_SERVER_PORT,
HBASE_THRIFT_SERVER_INFO_PORT),
quiet=quiet)
def main(args):
image_prefix = '{}/{}/topology_hdp:hdp{}_ambari{}'.format(
args.registry, args.namespace or DEFAULT_NAMESPACE, args.hdp_version,
args.ambari_version)
primary_node_image = '{}_{}'.format(image_prefix, 'primary-node')
secondary_node_image = '{}_{}'.format(image_prefix, 'secondary-node')
primary_node = Node(hostname=args.primary_node[0],
group='primary',
image=primary_node_image,
ports=[{
AMBARI_PORT: AMBARI_PORT
} if args.predictable else AMBARI_PORT])
secondary_nodes = [
Node(hostname=hostname, group='secondary', image=secondary_node_image)
for hostname in args.secondary_nodes
]
cluster = Cluster(primary_node, *secondary_nodes)
cluster.primary_node = primary_node
cluster.secondary_nodes = secondary_nodes
cluster.start(args.network)
logger.debug('Starting PostgreSQL for Ambari server ...')
primary_node.execute('service postgresql start', quiet=not args.verbose)
_update_node_names(cluster, quiet=not args.verbose)
# The HDP topology uses two pre-built images ('primary' and 'secondary'). If a cluster
# larger than 2 nodes is started, some modifications need to be done.
if len(secondary_nodes) > 1:
_remove_files(nodes=secondary_nodes[1:],
files=['/hadoop/hdfs/data/current/*'])
logger.info('Starting Ambari server ...')
primary_node.execute('ambari-server start', quiet=not args.verbose)
# Docker for Mac exposes ports that can be accessed only with ``localhost:<port>`` so
# use that instead of the hostname if the host name is ``moby``.
hostname = 'localhost' if client.info().get(
'Name') == 'moby' else socket.gethostname()
port = cluster.primary_node.host_ports.get(AMBARI_PORT)
server_url = 'http://{}:{}'.format(hostname, port)
logger.info('Ambari server is now reachable at %s', server_url)
logger.info('Starting Ambari agents ...')
for node in cluster:
logger.debug('Starting Ambari agent on %s ...', node.fqdn)
node.execute('ambari-agent start', quiet=not args.verbose)
ambari = Ambari(server_url, username='******', password='******')
def condition(ambari, cluster):
cluster_hosts = {node.fqdn for node in cluster}
ambari_hosts = {host.host_name for host in ambari.hosts}
logger.debug('Cluster hosts: %s; Ambari hosts: %s', cluster_hosts,
ambari_hosts)
return cluster_hosts == ambari_hosts
wait_for_condition(condition=condition, condition_args=[ambari, cluster])
for node in secondary_nodes[1:]:
logger.info('Adding %s to cluster ...', node.fqdn)
ambari.clusters('cluster').hosts.create(node.fqdn)
for component in ambari.clusters('cluster').hosts(
secondary_nodes[0].fqdn).components:
logger.debug('Adding component (%s) to cluster on host (%s) ...',
component.component_name, node.fqdn)
host_components = ambari.clusters('cluster').hosts(
node.fqdn).components
host_components.create(component.component_name).wait()
logger.debug('Installing all registered components on host (%s) ...',
node.fqdn)
ambari.clusters('cluster').hosts(node.fqdn).components.install().wait()
if not args.dont_start_cluster:
logger.debug(
'Waiting for all hosts to reach healthy state before starting cluster ...'
)
def condition(ambari):
health_report = ambari.clusters('cluster').health_report
logger.debug('Ambari cluster health report: %s ...', health_report)
return health_report.get('Host/host_state/HEALTHY') == len(
list(ambari.hosts))
wait_for_condition(condition=condition, condition_args=[ambari])
logger.info('Starting cluster services ...')
ambari.clusters('cluster').services.start().wait()