community.general/lib/ansible/modules/cloud/amazon/redshift.py

541 lines
19 KiB
Python

#!/usr/bin/python
# Copyright 2014 Jens Carl, Hothead Games Inc.
# GNU General Public License v3.0+ (see COPYING or https://www.gnu.org/licenses/gpl-3.0.txt)
from __future__ import absolute_import, division, print_function
__metaclass__ = type
ANSIBLE_METADATA = {'metadata_version': '1.1',
'status': ['preview'],
'supported_by': 'community'}
DOCUMENTATION = '''
---
author:
- "Jens Carl (@j-carl), Hothead Games Inc."
- "Rafael Driutti (@rafaeldriutti)"
module: redshift
version_added: "2.2"
short_description: create, delete, or modify an Amazon Redshift instance
description:
- Creates, deletes, or modifies amazon Redshift cluster instances.
options:
command:
description:
- Specifies the action to take.
required: true
choices: [ 'create', 'facts', 'delete', 'modify' ]
identifier:
description:
- Redshift cluster identifier.
required: true
node_type:
description:
- The node type of the cluster. Must be specified when command=create.
choices: ['ds1.xlarge', 'ds1.8xlarge', 'ds2.xlarge', 'ds2.8xlarge', 'dc1.large', 'dc1.8xlarge', 'dc2.large', 'dc2.8xlarge',
'dw1.xlarge', 'dw1.8xlarge', 'dw2.large', 'dw2.8xlarge']
username:
description:
- Master database username. Used only when command=create.
password:
description:
- Master database password. Used only when command=create.
cluster_type:
description:
- The type of cluster.
choices: ['multi-node', 'single-node' ]
default: 'single-node'
db_name:
description:
- Name of the database.
availability_zone:
description:
- availability zone in which to launch cluster
aliases: ['zone', 'aws_zone']
number_of_nodes:
description:
- Number of nodes. Only used when cluster_type=multi-node.
cluster_subnet_group_name:
description:
- which subnet to place the cluster
aliases: ['subnet']
cluster_security_groups:
description:
- in which security group the cluster belongs
aliases: ['security_groups']
vpc_security_group_ids:
description:
- VPC security group
aliases: ['vpc_security_groups']
skip_final_cluster_snapshot:
description:
- skip a final snapshot before deleting the cluster. Used only when command=delete.
aliases: ['skip_final_snapshot']
default: 'no'
version_added: "2.4"
final_cluster_snapshot_identifier:
description:
- identifier of the final snapshot to be created before deleting the cluster. If this parameter is provided,
final_cluster_snapshot_identifier must be false. Used only when command=delete.
aliases: ['final_snapshot_id']
version_added: "2.4"
preferred_maintenance_window:
description:
- maintenance window
aliases: ['maintance_window', 'maint_window']
cluster_parameter_group_name:
description:
- name of the cluster parameter group
aliases: ['param_group_name']
automated_snapshot_retention_period:
description:
- period when the snapshot take place
aliases: ['retention_period']
port:
description:
- which port the cluster is listining
cluster_version:
description:
- which version the cluster should have
aliases: ['version']
choices: ['1.0']
allow_version_upgrade:
description:
- flag to determinate if upgrade of version is possible
aliases: ['version_upgrade']
default: 'yes'
publicly_accessible:
description:
- if the cluster is accessible publicly or not
default: 'no'
encrypted:
description:
- if the cluster is encrypted or not
default: 'no'
elastic_ip:
description:
- if the cluster has an elastic IP or not
new_cluster_identifier:
description:
- Only used when command=modify.
aliases: ['new_identifier']
wait:
description:
- When command=create, modify or restore then wait for the database to enter the 'available' state.
When command=delete wait for the database to be terminated.
type: bool
default: 'no'
wait_timeout:
description:
- how long before wait gives up, in seconds
default: 300
requirements: [ 'boto3' ]
extends_documentation_fragment:
- aws
- ec2
'''
EXAMPLES = '''
# Basic cluster provisioning example
- redshift: >
command=create
node_type=ds1.xlarge
identifier=new_cluster
username=cluster_admin
password=1nsecure
# Cluster delete example
- redshift:
command: delete
identifier: new_cluster
skip_final_cluster_snapshot: true
wait: true
'''
RETURN = '''
cluster:
description: dictionary containing all the cluster information
returned: success
type: complex
contains:
identifier:
description: Id of the cluster.
returned: success
type: str
sample: "new_redshift_cluster"
create_time:
description: Time of the cluster creation as timestamp.
returned: success
type: float
sample: 1430158536.308
status:
description: Stutus of the cluster.
returned: success
type: str
sample: "available"
db_name:
description: Name of the database.
returned: success
type: str
sample: "new_db_name"
availability_zone:
description: Amazon availability zone where the cluster is located. "None" until cluster is available.
returned: success
type: str
sample: "us-east-1b"
maintenance_window:
description: Time frame when maintenance/upgrade are done.
returned: success
type: str
sample: "sun:09:30-sun:10:00"
private_ip_address:
description: Private IP address of the main node.
returned: success
type: str
sample: "10.10.10.10"
public_ip_address:
description: Public IP address of the main node. "None" when enhanced_vpc_routing is enabled.
returned: success
type: str
sample: "0.0.0.0"
port:
description: Port of the cluster. "None" until cluster is available.
returned: success
type: int
sample: 5439
url:
description: FQDN of the main cluster node. "None" until cluster is available.
returned: success
type: str
sample: "new-redshift_cluster.jfkdjfdkj.us-east-1.redshift.amazonaws.com"
enhanced_vpc_routing:
description: status of the enhanced vpc routing feature.
returned: success
type: boolean
'''
try:
import botocore
except ImportError:
pass # handled by AnsibleAWSModule
from ansible.module_utils.ec2 import ec2_argument_spec, snake_dict_to_camel_dict
from ansible.module_utils.aws.core import AnsibleAWSModule, is_boto3_error_code
def _collect_facts(resource):
"""Transfrom cluster information to dict."""
facts = {
'identifier': resource['ClusterIdentifier'],
'status': resource['ClusterStatus'],
'username': resource['MasterUsername'],
'db_name': resource['DBName'],
'maintenance_window': resource['PreferredMaintenanceWindow'],
'enhanced_vpc_routing': resource['EnhancedVpcRouting']
}
for node in resource['ClusterNodes']:
if node['NodeRole'] in ('SHARED', 'LEADER'):
facts['private_ip_address'] = node['PrivateIPAddress']
if facts['enhanced_vpc_routing'] is False:
facts['public_ip_address'] = node['PublicIPAddress']
else:
facts['public_ip_address'] = None
break
# Some parameters are not ready instantly if you don't wait for available
# cluster status
facts['create_time'] = None
facts['url'] = None
facts['port'] = None
facts['availability_zone'] = None
if resource['ClusterStatus'] != "creating":
facts['create_time'] = resource['ClusterCreateTime']
facts['url'] = resource['Endpoint']['Address']
facts['port'] = resource['Endpoint']['Port']
facts['availability_zone'] = resource['AvailabilityZone']
return facts
def create_cluster(module, redshift):
"""
Create a new cluster
module: AnsibleModule object
redshift: authenticated redshift connection object
Returns:
"""
identifier = module.params.get('identifier')
node_type = module.params.get('node_type')
username = module.params.get('username')
password = module.params.get('password')
d_b_name = module.params.get('db_name')
wait = module.params.get('wait')
wait_timeout = module.params.get('wait_timeout')
changed = True
# Package up the optional parameters
params = {}
for p in ('cluster_type', 'cluster_security_groups',
'vpc_security_group_ids', 'cluster_subnet_group_name',
'availability_zone', 'preferred_maintenance_window',
'cluster_parameter_group_name',
'automated_snapshot_retention_period', 'port',
'cluster_version', 'allow_version_upgrade',
'number_of_nodes', 'publicly_accessible',
'encrypted', 'elastic_ip', 'enhanced_vpc_routing'):
# https://github.com/boto/boto3/issues/400
if module.params.get(p) is not None:
params[p] = module.params.get(p)
if d_b_name:
params['d_b_name'] = d_b_name
try:
redshift.describe_clusters(ClusterIdentifier=identifier)['Clusters'][0]
changed = False
except is_boto3_error_code('ClusterNotFound'):
try:
redshift.create_cluster(ClusterIdentifier=identifier,
NodeType=node_type,
MasterUsername=username,
MasterUserPassword=password,
**snake_dict_to_camel_dict(params, capitalize_first=True))
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json_aws(e, msg="Failed to create cluster")
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e: # pylint: disable=duplicate-except
module.fail_json_aws(e, msg="Failed to describe cluster")
if wait:
attempts = wait_timeout // 60
waiter = redshift.get_waiter('cluster_available')
try:
waiter.wait(
ClusterIdentifier=identifier,
WaiterConfig=dict(MaxAttempts=attempts)
)
except (botocore.exceptions.ClientError, botocore.exceptions.BotoCoreError) as e:
module.fail_json_aws(e, msg="Timeout waiting for the cluster creation")
try:
resource = redshift.describe_clusters(ClusterIdentifier=identifier)['Clusters'][0]
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json_aws(e, msg="Failed to describe cluster")
return(changed, _collect_facts(resource))
def describe_cluster(module, redshift):
"""
Collect data about the cluster.
module: Ansible module object
redshift: authenticated redshift connection object
"""
identifier = module.params.get('identifier')
try:
resource = redshift.describe_clusters(ClusterIdentifier=identifier)['Clusters'][0]
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json_aws(e, msg="Error describing cluster")
return(True, _collect_facts(resource))
def delete_cluster(module, redshift):
"""
Delete a cluster.
module: Ansible module object
redshift: authenticated redshift connection object
"""
identifier = module.params.get('identifier')
wait = module.params.get('wait')
wait_timeout = module.params.get('wait_timeout')
params = {}
for p in ('skip_final_cluster_snapshot',
'final_cluster_snapshot_identifier'):
if p in module.params:
# https://github.com/boto/boto3/issues/400
if module.params.get(p) is not None:
params[p] = module.params.get(p)
try:
redshift.delete_cluster(
ClusterIdentifier=identifier,
**snake_dict_to_camel_dict(params, capitalize_first=True)
)
except is_boto3_error_code('ClusterNotFound'):
return(False, {})
except (botocore.exceptions.ClientError, botocore.exceptions.BotoCoreError) as e: # pylint: disable=duplicate-except
module.fail_json_aws(e, msg="Failed to delete cluster")
if wait:
attempts = wait_timeout // 60
waiter = redshift.get_waiter('cluster_deleted')
try:
waiter.wait(
ClusterIdentifier=identifier,
WaiterConfig=dict(MaxAttempts=attempts)
)
except (botocore.exceptions.ClientError, botocore.exceptions.BotoCoreError) as e:
module.fail_json_aws(e, msg="Timeout deleting the cluster")
return(True, {})
def modify_cluster(module, redshift):
"""
Modify an existing cluster.
module: Ansible module object
redshift: authenticated redshift connection object
"""
identifier = module.params.get('identifier')
wait = module.params.get('wait')
wait_timeout = module.params.get('wait_timeout')
# Package up the optional parameters
params = {}
for p in ('cluster_type', 'cluster_security_groups',
'vpc_security_group_ids', 'cluster_subnet_group_name',
'availability_zone', 'preferred_maintenance_window',
'cluster_parameter_group_name',
'automated_snapshot_retention_period', 'port', 'cluster_version',
'allow_version_upgrade', 'number_of_nodes', 'new_cluster_identifier'):
# https://github.com/boto/boto3/issues/400
if module.params.get(p) is not None:
params[p] = module.params.get(p)
# enhanced_vpc_routing parameter change needs an exclusive request
if module.params.get('enhanced_vpc_routing') is not None:
try:
redshift.modify_cluster(ClusterIdentifier=identifier,
EnhancedVpcRouting=module.params.get('enhanced_vpc_routing'))
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json_aws(e, msg="Couldn't modify redshift cluster %s " % identifier)
if wait:
attempts = wait_timeout // 60
waiter = redshift.get_waiter('cluster_available')
try:
waiter.wait(
ClusterIdentifier=identifier,
WaiterConfig=dict(MaxAttempts=attempts)
)
except (botocore.exceptions.ClientError, botocore.exceptions.BotoCoreError) as e:
module.fail_json_aws(e,
msg="Timeout waiting for cluster enhanced vpc routing modification"
)
# change the rest
try:
redshift.modify_cluster(ClusterIdentifier=identifier,
**snake_dict_to_camel_dict(params, capitalize_first=True))
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json_aws(e, msg="Couldn't modify redshift cluster %s " % identifier)
if module.params.get('new_cluster_identifier'):
identifier = module.params.get('new_cluster_identifier')
if wait:
attempts = wait_timeout // 60
waiter2 = redshift.get_waiter('cluster_available')
try:
waiter2.wait(
ClusterIdentifier=identifier,
WaiterConfig=dict(MaxAttempts=attempts)
)
except (botocore.exceptions.ClientError, botocore.exceptions.BotoCoreError) as e:
module.fail_json_aws(e, msg="Timeout waiting for cluster modification")
try:
resource = redshift.describe_clusters(ClusterIdentifier=identifier)['Clusters'][0]
except (botocore.exceptions.BotoCoreError, botocore.exceptions.ClientError) as e:
module.fail_json(e, msg="Couldn't modify redshift cluster %s " % identifier)
return(True, _collect_facts(resource))
def main():
argument_spec = ec2_argument_spec()
argument_spec.update(dict(
command=dict(choices=['create', 'facts', 'delete', 'modify'], required=True),
identifier=dict(required=True),
node_type=dict(choices=['ds1.xlarge', 'ds1.8xlarge', 'ds2.xlarge',
'ds2.8xlarge', 'dc1.large', 'dc2.large',
'dc1.8xlarge', 'dw1.xlarge', 'dw1.8xlarge',
'dw2.large', 'dw2.8xlarge'], required=False),
username=dict(required=False),
password=dict(no_log=True, required=False),
db_name=dict(require=False),
cluster_type=dict(choices=['multi-node', 'single-node'], default='single-node'),
cluster_security_groups=dict(aliases=['security_groups'], type='list'),
vpc_security_group_ids=dict(aliases=['vpc_security_groups'], type='list'),
skip_final_cluster_snapshot=dict(aliases=['skip_final_snapshot'],
type='bool', default=False),
final_cluster_snapshot_identifier=dict(aliases=['final_snapshot_id'], required=False),
cluster_subnet_group_name=dict(aliases=['subnet']),
availability_zone=dict(aliases=['aws_zone', 'zone']),
preferred_maintenance_window=dict(aliases=['maintance_window', 'maint_window']),
cluster_parameter_group_name=dict(aliases=['param_group_name']),
automated_snapshot_retention_period=dict(aliases=['retention_period'], type='int'),
port=dict(type='int'),
cluster_version=dict(aliases=['version'], choices=['1.0']),
allow_version_upgrade=dict(aliases=['version_upgrade'], type='bool', default=True),
number_of_nodes=dict(type='int'),
publicly_accessible=dict(type='bool', default=False),
encrypted=dict(type='bool', default=False),
elastic_ip=dict(required=False),
new_cluster_identifier=dict(aliases=['new_identifier']),
enhanced_vpc_routing=dict(type='bool', default=False),
wait=dict(type='bool', default=False),
wait_timeout=dict(type='int', default=300),
))
required_if = [
('command', 'delete', ['skip_final_cluster_snapshot']),
('command', 'create', ['node_type',
'username',
'password'])
]
module = AnsibleAWSModule(
argument_spec=argument_spec,
required_if=required_if
)
command = module.params.get('command')
skip_final_cluster_snapshot = module.params.get('skip_final_cluster_snapshot')
final_cluster_snapshot_identifier = module.params.get('final_cluster_snapshot_identifier')
# can't use module basic required_if check for this case
if command == 'delete' and skip_final_cluster_snapshot is False and final_cluster_snapshot_identifier is None:
module.fail_json(msg="Need to specifiy final_cluster_snapshot_identifier if skip_final_cluster_snapshot is False")
conn = module.client('redshift')
changed = True
if command == 'create':
(changed, cluster) = create_cluster(module, conn)
elif command == 'facts':
(changed, cluster) = describe_cluster(module, conn)
elif command == 'delete':
(changed, cluster) = delete_cluster(module, conn)
elif command == 'modify':
(changed, cluster) = modify_cluster(module, conn)
module.exit_json(changed=changed, cluster=cluster)
if __name__ == '__main__':
main()