VMware: Refactor vmware_cluster (#37072)

* Update cluster logic
* Refactor code
* Documentation update

Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
pull/45340/head
Abhijeet Kasurde 6 years ago committed by ansibot
parent b964c46235
commit 8cd395f595

@ -113,17 +113,6 @@ def find_dvspg_by_name(dv_switch, portgroup_name):
return None
# Maintain for legacy, or remove with 2.1 ?
# Should be replaced with find_cluster_by_name
def find_cluster_by_name_datacenter(datacenter, cluster_name):
host_folder = datacenter.hostFolder
for folder in host_folder.childEntity:
if folder.name == cluster_name:
return folder
return None
def find_object_by_name(content, name, obj_type, folder=None, recurse=True):
if not isinstance(obj_type, list):
obj_type = [obj_type]

@ -2,60 +2,180 @@
# -*- coding: utf-8 -*-
# Copyright: (c) 2015, Joseph Callen <jcallen () csc.com>
# Copyright: (c) 2018, Ansible Project
#
# 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'}
ANSIBLE_METADATA = {
'metadata_version': '1.1',
'status': ['preview'],
'supported_by': 'community'
}
DOCUMENTATION = r'''
---
module: vmware_cluster
short_description: Manage VMware vSphere clusters
description:
- Add or remove VMware vSphere clusters.
- This module can be used to add, remove and update VMware vSphere clusters and its configurations.
- Module can manage HA, DRS and VSAN related configurations.
- All values and VMware object names are case sensitive.
version_added: '2.0'
author:
- Joseph Callen (@jcpowermac)
- Abhijeet Kasurde (@Akasurde)
requirements:
- Tested on ESXi 5.5
- PyVmomi installed
- Tested on ESXi 5.5 and 6.5.
- PyVmomi installed.
options:
cluster_name:
description:
- The name of the cluster that will be created.
required: yes
datacenter_name:
description:
- The name of the datacenter the cluster will be created in.
required: yes
description:
- The name of the cluster to be managed.
required: yes
datacenter:
description:
- The name of the datacenter.
required: yes
aliases: [ datacenter_name ]
enable_drs:
description:
- If set to C(yes) will enable DRS when the cluster is created.
type: bool
default: 'no'
description:
- If set to C(yes), will enable DRS when the cluster is created.
type: bool
default: 'no'
drs_enable_vm_behavior_overrides:
description:
- Determines whether DRS Behavior overrides for individual virtual machines are enabled.
- If set to C(True), overrides C(drs_default_vm_behavior).
type: bool
default: True
version_added: 2.8
drs_default_vm_behavior:
description:
- Specifies the cluster-wide default DRS behavior for virtual machines.
- If set to C(partiallyAutomated), then vCenter generate recommendations for virtual machine migration and
for the placement with a host. vCenter automatically implement placement at power on.
- If set to C(manual), then vCenter generate recommendations for virtual machine migration and
for the placement with a host. vCenter should not implement the recommendations automatically.
- If set to C(fullyAutomated), then vCenter should automate both the migration of virtual machines
and their placement with a host at power on.
default: fullyAutomated
choices: [ fullyAutomated, manual, partiallyAutomated ]
version_added: 2.8
drs_vmotion_rate:
description:
- Threshold for generated ClusterRecommendations.
default: 3
choices: [ 1, 2, 3, 4, 5 ]
version_added: 2.8
enable_ha:
description:
- If set to C(yes) will enable HA when the cluster is created.
type: bool
default: 'no'
description:
- If set to C(yes) will enable HA when the cluster is created.
type: bool
default: 'no'
ha_host_monitoring:
description:
- Indicates whether HA restarts virtual machines after a host fails.
- If set to C(enabled), HA restarts virtual machines after a host fails.
- If set to C(disabled), HA does not restart virtual machines after a host fails.
- If C(enable_ha) is set to C(no), then this value is ignored.
choices: [ 'enabled', 'disabled' ]
default: 'enabled'
version_added: 2.8
ha_vm_monitoring:
description:
- Indicates the state of virtual machine health monitoring service.
- If set to C(vmAndAppMonitoring), HA response to both virtual machine and application heartbeat failure.
- If set to C(vmMonitoringDisabled), virtual machine health monitoring is disabled.
- If set to C(vmMonitoringOnly), HA response to virtual machine heartbeat failure.
- If C(enable_ha) is set to C(no), then this value is ignored.
choices: ['vmAndAppMonitoring', 'vmMonitoringOnly', 'vmMonitoringDisabled']
default: 'vmMonitoringDisabled'
version_added: 2.8
ha_failover_level:
description:
- Number of host failures that should be tolerated, still guaranteeing sufficient resources to
restart virtual machines on available hosts.
- Accepts integer values only.
default: 2
version_added: 2.8
ha_admission_control_enabled:
description:
- Determines if strict admission control is enabled.
- It is recommended to set this parameter to C(True), please refer documentation for more details.
default: True
type: bool
version_added: 2.8
ha_vm_failure_interval:
description:
- The number of seconds after which virtual machine is declared as failed
if no heartbeat has been received.
- This setting is only valid if C(ha_vm_monitoring) is set to, either C(vmAndAppMonitoring) or C(vmMonitoringOnly).
- Unit is seconds.
default: 30
version_added: 2.8
ha_vm_min_up_time:
description:
- The number of seconds for the virtual machine's heartbeats to stabilize after
the virtual machine has been powered on.
- This setting is only valid if C(ha_vm_monitoring) is set to, either C(vmAndAppMonitoring) or C(vmMonitoringOnly).
- Unit is seconds.
default: 120
version_added: 2.8
ha_vm_max_failures:
description:
- Maximum number of failures and automated resets allowed during the time
that C(ha_vm_max_failure_window) specifies.
- This setting is only valid if C(ha_vm_monitoring) is set to, either C(vmAndAppMonitoring) or C(vmMonitoringOnly).
default: 3
version_added: 2.8
ha_vm_max_failure_window:
description:
- The number of seconds for the window during which up to C(ha_vm_max_failures) resets
can occur before automated responses stop.
- This setting is only valid if C(ha_vm_monitoring) is set to, either C(vmAndAppMonitoring) or C(vmMonitoringOnly).
- Unit is seconds.
- Default specifies no failure window.
default: -1
version_added: 2.8
ha_restart_priority:
description:
- Determines the preference that HA gives to a virtual machine if sufficient capacity is not available
to power on all failed virtual machines.
- This setting is only valid if C(ha_vm_monitoring) is set to, either C(vmAndAppMonitoring) or C(vmMonitoringOnly).
- If set to C(disabled), then HA is disabled for this virtual machine.
- If set to C(high), then virtual machine with this priority have a higher chance of powering on after a failure,
when there is insufficient capacity on hosts to meet all virtual machine needs.
- If set to C(medium), then virtual machine with this priority have an intermediate chance of powering on after a failure,
when there is insufficient capacity on hosts to meet all virtual machine needs.
- If set to C(low), then virtual machine with this priority have a lower chance of powering on after a failure,
when there is insufficient capacity on hosts to meet all virtual machine needs.
default: 'medium'
version_added: 2.8
choices: [ 'disabled', 'high', 'low', 'medium' ]
enable_vsan:
description:
- If set to C(yes) will enable vSAN when the cluster is created.
type: bool
default: 'no'
description:
- If set to C(yes) will enable vSAN when the cluster is created.
type: bool
default: 'no'
vsan_auto_claim_storage:
description:
- Determines whether the VSAN service is configured to automatically claim local storage
on VSAN-enabled hosts in the cluster.
type: bool
default: False
version_added: 2.8
state:
description:
- Create (C(present)) or remove (C(absent)) a VMware vSphere cluster.
choices: [absent, present]
default: present
description:
- Create C(present) or remove C(absent) a VMware vSphere cluster.
choices: [ absent, present ]
default: present
extends_documentation_fragment: vmware.documentation
'''
EXAMPLES = r'''
EXAMPLES = r"""
- name: Create Cluster
vmware_cluster:
hostname: '{{ vcenter_hostname }}'
@ -67,47 +187,73 @@ EXAMPLES = r'''
enable_drs: yes
enable_vsan: yes
delegate_to: localhost
'''
- name: Create Cluster with additional changes
vmware_cluster:
hostname: "{{ vcenter_server }}"
username: "{{ vcenter_user }}"
password: "{{ vcenter_pass }}"
validate_certs: no
datacenter_name: DC0
cluster_name: "{{ cluster_name }}"
enable_ha: True
ha_vm_monitoring: vmMonitoringOnly
enable_drs: True
drs_default_vm_behavior: partiallyAutomated
enable_vsan: True
register: cl_result
delegate_to: localhost
- name: Delete Cluster
vmware_cluster:
hostname: "{{ vcenter_server }}"
username: "{{ vcenter_user }}"
password: "{{ vcenter_pass }}"
datacenter_name: datacenter
cluster_name: cluster
enable_ha: yes
enable_drs: yes
enable_vsan: yes
state: absent
"""
RETURN = r"""#
"""
try:
from pyVmomi import vim, vmodl
HAS_PYVMOMI = True
except ImportError:
HAS_PYVMOMI = False
pass
from ansible.module_utils.basic import AnsibleModule
from ansible.module_utils.vmware import (HAS_PYVMOMI,
TaskError,
connect_to_api,
find_cluster_by_name_datacenter,
find_datacenter_by_name,
vmware_argument_spec,
wait_for_task
)
from ansible.module_utils.vmware import (PyVmomi, TaskError, find_datacenter_by_name,
vmware_argument_spec, wait_for_task)
from ansible.module_utils._text import to_native
class VMwareCluster(object):
class VMwareCluster(PyVmomi):
def __init__(self, module):
self.module = module
super(VMwareCluster, self).__init__(module)
self.cluster_name = module.params['cluster_name']
self.datacenter_name = module.params['datacenter_name']
self.datacenter_name = module.params['datacenter']
self.enable_drs = module.params['enable_drs']
self.enable_ha = module.params['enable_ha']
self.enable_vsan = module.params['enable_vsan']
self.desired_state = module.params['state']
self.datacenter = None
self.cluster = None
self.content = connect_to_api(module)
def process_state(self):
"""
Manage internal states of cluster
"""
cluster_states = {
'absent': {
'present': self.state_destroy_cluster,
'absent': self.state_exit_unchanged,
},
'present': {
'update': self.state_update_cluster,
'present': self.state_exit_unchanged,
'present': self.state_update_cluster,
'absent': self.state_create_cluster,
}
}
@ -117,27 +263,71 @@ class VMwareCluster(object):
cluster_states[self.desired_state][current_state]()
def configure_ha(self):
"""
Manage HA Configuration
Returns: Cluster DAS configuration spec
"""
das_config = vim.cluster.DasConfigInfo()
das_config.enabled = self.enable_ha
das_config.admissionControlPolicy = vim.cluster.FailoverLevelAdmissionControlPolicy()
das_config.admissionControlPolicy.failoverLevel = 2
das_config.admissionControlPolicy.failoverLevel = self.params.get('ha_failover_level')
ha_vm_monitoring = self.params.get('ha_vm_monitoring')
das_vm_config = None
if ha_vm_monitoring in ['vmMonitoringOnly', 'vmAndAppMonitoring']:
vm_tool_spec = vim.cluster.VmToolsMonitoringSettings()
vm_tool_spec.enabled = True
vm_tool_spec.vmMonitoring = ha_vm_monitoring
vm_tool_spec.failureInterval = self.params.get('ha_vm_failure_interval')
vm_tool_spec.minUpTime = self.params.get('ha_vm_min_up_time')
vm_tool_spec.maxFailures = self.params.get('ha_vm_max_failures')
vm_tool_spec.maxFailureWindow = self.params.get('ha_vm_max_failure_window')
das_vm_config = vim.cluster.DasVmSettings()
das_vm_config.restartPriority = self.params.get('ha_restart_priority')
das_vm_config.isolationResponse = None
das_vm_config.vmToolsMonitoringSettings = vm_tool_spec
das_config.admissionControlEnabled = self.params.get('ha_admission_control_enabled')
das_config.hostMonitoring = self.params.get('ha_host_monitoring')
das_config.vmMonitoring = ha_vm_monitoring
das_config.defaultVmSettings = das_vm_config
return das_config
def configure_drs(self):
"""
Manage DRS configuration
Returns: Cluster DRS configuration spec
"""
drs_config = vim.cluster.DrsConfigInfo()
drs_config.enabled = self.enable_drs
# Set to partially automated
drs_config.vmotionRate = 3
drs_config.enableVmBehaviorOverrides = self.params.get('drs_enable_vm_behavior_overrides')
drs_config.defaultVmBehavior = self.params.get('drs_default_vm_behavior')
drs_config.vmotionRate = self.params.get('drs_vmotion_rate')
return drs_config
def configure_vsan(self):
"""
Manage VSAN configuration
Returns: Cluster VSAN configuration spec
"""
vsan_config = vim.vsan.cluster.ConfigInfo()
vsan_config.enabled = self.enable_vsan
vsan_config.defaultConfig = vim.vsan.cluster.ConfigInfo.HostDefaultInfo()
vsan_config.defaultConfig.autoClaimStorage = False
vsan_config.defaultConfig.autoClaimStorage = self.params.get('vsan_auto_claim_storage')
return vsan_config
def state_create_cluster(self):
"""
Create cluster with given configuration
"""
try:
cluster_config_spec = vim.cluster.ConfigSpecEx()
cluster_config_spec.dasConfig = self.configure_ha()
@ -148,23 +338,33 @@ class VMwareCluster(object):
self.datacenter.hostFolder.CreateClusterEx(self.cluster_name, cluster_config_spec)
self.module.exit_json(changed=True)
except vim.fault.DuplicateName:
self.module.fail_json(msg="A cluster with the name %s already exists" % self.cluster_name)
except vmodl.fault.InvalidArgument:
self.module.fail_json(msg="Cluster configuration specification parameter is invalid")
except vim.fault.InvalidName:
self.module.fail_json(msg="%s is an invalid name for a cluster" % self.cluster_name)
except vmodl.fault.NotSupported:
# To match other vmware_* modules
pass
except vmodl.fault.InvalidArgument as invalid_args:
self.module.fail_json(msg="Cluster configuration specification"
" parameter is invalid : %s" % to_native(invalid_args.msg))
except vim.fault.InvalidName as invalid_name:
self.module.fail_json(msg="'%s' is an invalid name for a"
" cluster : %s" % (self.cluster_name,
to_native(invalid_name.msg)))
except vmodl.fault.NotSupported as not_supported:
# This should never happen
self.module.fail_json(msg="Trying to create a cluster on an incorrect folder object")
self.module.fail_json(msg="Trying to create a cluster on an incorrect"
" folder object : %s" % to_native(not_supported.msg))
except vmodl.RuntimeFault as runtime_fault:
self.module.fail_json(msg=runtime_fault.msg)
self.module.fail_json(msg=to_native(runtime_fault.msg))
except vmodl.MethodFault as method_fault:
# This should never happen either
self.module.fail_json(msg=method_fault.msg)
self.module.fail_json(msg=to_native(method_fault.msg))
except Exception as generic_exc:
self.module.fail_json(msg="Failed to create cluster"
" due to generic exception %s" % to_native(generic_exc))
def state_destroy_cluster(self):
changed = True
result = None
"""
Destroy cluster
"""
changed, result = False, None
try:
if not self.module.check_mode:
@ -172,79 +372,167 @@ class VMwareCluster(object):
changed, result = wait_for_task(task)
self.module.exit_json(changed=changed, result=result)
except vim.fault.VimFault as vim_fault:
self.module.fail_json(msg=vim_fault.msg)
self.module.fail_json(msg=to_native(vim_fault.msg))
except vmodl.RuntimeFault as runtime_fault:
self.module.fail_json(msg=runtime_fault.msg)
self.module.fail_json(msg=to_native(runtime_fault.msg))
except vmodl.MethodFault as method_fault:
self.module.fail_json(msg=method_fault.msg)
self.module.fail_json(msg=to_native(method_fault.msg))
except Exception as generic_exc:
self.module.fail_json(msg="Failed to destroy cluster"
" due to generic exception %s" % to_native(generic_exc))
def state_exit_unchanged(self):
"""
Exit without any change
"""
self.module.exit_json(changed=False)
def state_update_cluster(self):
"""
Update cluster configuration of existing cluster
"""
changed, result = False, None
cluster_config_spec = vim.cluster.ConfigSpecEx()
changed = True
result = None
if self.cluster.configurationEx.dasConfig.enabled != self.enable_ha:
diff = False # Triggers Reconfigure Task only when there is a change
if self.check_ha_config_diff():
cluster_config_spec.dasConfig = self.configure_ha()
if self.cluster.configurationEx.drsConfig.enabled != self.enable_drs:
diff = True
if self.check_drs_config_diff():
cluster_config_spec.drsConfig = self.configure_drs()
if self.cluster.configurationEx.vsanConfigInfo.enabled != self.enable_vsan:
diff = True
if self.check_vsan_config_diff():
cluster_config_spec.vsanConfig = self.configure_vsan()
diff = True
try:
if not self.module.check_mode:
if not self.module.check_mode and diff:
task = self.cluster.ReconfigureComputeResource_Task(cluster_config_spec, True)
changed, result = wait_for_task(task)
self.module.exit_json(changed=changed, result=result)
except vmodl.RuntimeFault as runtime_fault:
self.module.fail_json(msg=runtime_fault.msg)
self.module.fail_json(msg=to_native(runtime_fault.msg))
except vmodl.MethodFault as method_fault:
self.module.fail_json(msg=method_fault.msg)
self.module.fail_json(msg=to_native(method_fault.msg))
except TaskError as task_e:
self.module.fail_json(msg=str(task_e))
self.module.fail_json(msg=to_native(task_e))
except Exception as generic_exc:
self.module.fail_json(msg="Failed to update cluster"
" due to generic exception %s" % to_native(generic_exc))
def check_ha_config_diff(self):
"""
Check HA configuration diff
Returns: True if there is diff, else False
"""
das_config = self.cluster.configurationEx.dasConfig
if das_config.enabled != self.enable_ha or \
das_config.admissionControlPolicy.failoverLevel != self.params.get('ha_failover_level') or \
das_config.vmMonitoring != self.params.get('ha_vm_monitoring') or \
das_config.hostMonitoring != self.params.get('ha_host_monitoring') or \
das_config.admissionControlPolicy.failoverLevel != self.params.get('ha_failover_level') or \
das_config.admissionControlEnabled != self.params.get('ha_admission_control_enabled') or \
das_config.defaultVmSettings.restartPriority != self.params.get('ha_restart_priority') or \
das_config.defaultVmSettings.vmToolsMonitoringSettings.vmMonitoring != self.params.get('ha_vm_monitoring') or \
das_config.defaultVmSettings.vmToolsMonitoringSettings.failureInterval != self.params.get('ha_vm_failure_interval') or \
das_config.defaultVmSettings.vmToolsMonitoringSettings.minUpTime != self.params.get('ha_vm_min_up_time') or \
das_config.defaultVmSettings.vmToolsMonitoringSettings.maxFailures != self.params.get('ha_vm_max_failures') or \
das_config.defaultVmSettings.vmToolsMonitoringSettings.maxFailureWindow != self.params.get('ha_vm_max_failure_window'):
return True
return False
def check_drs_config_diff(self):
"""
Check DRS configuration diff
Returns: True if there is diff, else False
"""
drs_config = self.cluster.configurationEx.drsConfig
if drs_config.enabled != self.enable_drs or \
drs_config.enableVmBehaviorOverrides != self.params.get('drs_enable_vm_behavior_overrides') or \
drs_config.defaultVmBehavior != self.params.get('drs_default_vm_behavior') or \
drs_config.vmotionRate != self.params.get('drs_vmotion_rate'):
return True
return False
def check_vsan_config_diff(self):
"""
Check VSAN configuration diff
Returns: True if there is diff, else False
"""
vsan_config = self.cluster.configurationEx.vsanConfigInfo
if vsan_config.enabled != self.enable_vsan or \
vsan_config.defaultConfig.autoClaimStorage != self.params.get('vsan_auto_claim_storage'):
return True
return False
def check_cluster_configuration(self):
"""
Check cluster configuration
Returns: 'Present' if cluster exists, else 'absent'
"""
try:
self.datacenter = find_datacenter_by_name(self.content, self.datacenter_name)
if self.datacenter is None:
self.module.fail_json(msg="Datacenter %s does not exist, "
"please create first with Ansible Module vmware_datacenter or manually."
% self.datacenter_name)
self.cluster = find_cluster_by_name_datacenter(self.datacenter, self.cluster_name)
self.module.fail_json(msg="Datacenter %s does not exist." % self.datacenter_name)
self.cluster = self.find_cluster_by_name(cluster_name=self.cluster_name)
if self.cluster is None:
return 'absent'
else:
desired_state = (self.enable_ha,
self.enable_drs,
self.enable_vsan)
current_state = (self.cluster.configurationEx.dasConfig.enabled,
self.cluster.configurationEx.drsConfig.enabled,
self.cluster.configurationEx.vsanConfigInfo.enabled)
if desired_state != current_state:
return 'update'
else:
return 'present'
return 'present'
except vmodl.RuntimeFault as runtime_fault:
self.module.fail_json(msg=runtime_fault.msg)
self.module.fail_json(msg=to_native(runtime_fault.msg))
except vmodl.MethodFault as method_fault:
self.module.fail_json(msg=method_fault.msg)
self.module.fail_json(msg=to_native(method_fault.msg))
except Exception as generic_exc:
self.module.fail_json(msg="Failed to check configuration"
" due to generic exception %s" % to_native(generic_exc))
def main():
argument_spec = vmware_argument_spec()
argument_spec.update(dict(
cluster_name=dict(type='str', required=True),
datacenter_name=dict(type='str', required=True),
datacenter=dict(type='str', required=True, aliases=['datacenter_name']),
state=dict(type='str',
default='present',
choices=['absent', 'present']),
# DRS
enable_drs=dict(type='bool', default=False),
drs_enable_vm_behavior_overrides=dict(type='bool', default=True),
drs_default_vm_behavior=dict(type='str',
choices=['fullyAutomated', 'manual', 'partiallyAutomated'],
default='fullyAutomated'),
drs_vmotion_rate=dict(type='int',
choices=range(1, 6),
default=3),
# HA
enable_ha=dict(type='bool', default=False),
ha_failover_level=dict(type='int', default=2),
ha_host_monitoring=dict(type='str',
default='enabled',
choices=['enabled', 'disabled']),
# HA VM Monitoring related parameters
ha_vm_monitoring=dict(type='str',
choices=['vmAndAppMonitoring', 'vmMonitoringOnly', 'vmMonitoringDisabled'],
default='vmMonitoringDisabled'),
ha_vm_failure_interval=dict(type='int', default=30),
ha_vm_min_up_time=dict(type='int', default=120),
ha_vm_max_failures=dict(type='int', default=3),
ha_vm_max_failure_window=dict(type='int', default=-1),
ha_restart_priority=dict(type='str',
choices=['high', 'low', 'medium', 'disabled'],
default='medium'),
ha_admission_control_enabled=dict(type='bool', default=True),
# VSAN
enable_vsan=dict(type='bool', default=False),
state=dict(type='str', default='present', choices=['absent', 'present']),
vsan_auto_claim_storage=dict(type='bool', default=False),
))
module = AnsibleModule(
@ -252,9 +540,6 @@ def main():
supports_check_mode=True,
)
if not HAS_PYVMOMI:
module.fail_json(msg='pyvmomi is required for this module')
vmware_cluster = VMwareCluster(module)
vmware_cluster.process_state()

Loading…
Cancel
Save