pvc/client-cli/cli_lib/provisioner.py

1032 lines
38 KiB
Python

#!/usr/bin/env python3
# provisioner.py - PVC CLI client function library, Provisioner functions
# Part of the Parallel Virtual Cluster (PVC) system
#
# Copyright (C) 2018-2019 Joshua M. Boniface <joshua@boniface.me>
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program. If not, see <https://www.gnu.org/licenses/>.
#
###############################################################################
import time
import re
import subprocess
import click
import requests
import cli_lib.ansiprint as ansiprint
def get_request_uri(config, endpoint):
"""
Return the fully-formed URI for {endpoint}
"""
uri = '{}://{}{}{}'.format(
config['api_scheme'],
config['api_host'],
config['api_prefix'],
endpoint
)
return uri
#
# Primary functions
#
def template_info(config, template, template_type):
"""
Get information about template
API endpoint: GET /api/v1/provisioner/template/{template_type}/{template}
API arguments:
API schema: {json_data_object}
"""
request_uri = get_request_uri(config, '/provisioner/template/{template_type}/{template}'.format(template_type=template_type, template=template))
response = requests.get(
request_uri
)
if config['debug']:
print('API endpoint: GET {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
return True, response.json()
else:
return False, response.json()['message']
def template_list(config, limit, template_type=None):
"""
Get list information about templates (limited by {limit})
API endpoint: GET /api/v1/provisioner/template/{template_type}
API arguments: limit={limit}
API schema: [{json_data_object},{json_data_object},etc.]
"""
params = dict()
if limit:
params['limit'] = limit
if template_type is not None:
request_uri = get_request_uri(config, '/provisioner/template/{template_type}'.format(template_type=template_type))
else:
request_uri = get_request_uri(config, '/provisioner/template')
response = requests.get(
request_uri,
params=params
)
if config['debug']:
print('API endpoint: GET {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
return True, response.json()
else:
return False, response.json()['message']
def format_list_template(template_data, template_type=None):
"""
Format the returned template data
template_type can be used to only display part of the full list, allowing function
reuse with more limited output options.
"""
template_types = [ 'system', 'network', 'storage' ]
if template_type in template_types:
template_types = [ template_type ]
if 'system' in template_types:
click.echo('System templates:')
click.echo()
format_list_template_system(template_data['system_templates'])
if len(template_types) > 1:
click.echo()
if 'network' in template_types:
click.echo('Network templates:')
click.echo()
format_list_template_network(template_data['network_templates'])
if len(template_types) > 1:
click.echo()
# if 'storage' in template_types:
# format_list_template_storage(template_data['storage_templates'])
def format_list_template_system(template_data):
if isinstance(template_data, dict):
template_data = [ template_data ]
template_list_output = []
# Determine optimal column widths
template_name_length = 5
template_id_length = 4
template_vcpu_length = 6
template_vram_length = 10
template_serial_length = 7
template_vnc_length = 4
template_vnc_bind_length = 10
template_node_limit_length = 9
template_node_selector_length = 11
template_node_autostart_length = 11
for template in template_data:
# template_name column
_template_name_length = len(str(template['name'])) + 1
if _template_name_length > template_name_length:
template_name_length = _template_name_length
# template_id column
_template_id_length = len(str(template['id'])) + 1
if _template_id_length > template_id_length:
template_id_length = _template_id_length
# template_vcpu column
_template_vcpu_length = len(str(template['vcpu_count'])) + 1
if _template_vcpu_length > template_vcpu_length:
template_vcpu_length = _template_vcpu_length
# template_vram column
_template_vram_length = len(str(template['vram_mb'])) + 1
if _template_vram_length > template_vram_length:
template_vram_length = _template_vram_length
# template_serial column
_template_serial_length = len(str(template['serial'])) + 1
if _template_serial_length > template_serial_length:
template_serial_length = _template_serial_length
# template_vnc column
_template_vnc_length = len(str(template['vnc'])) + 1
if _template_vnc_length > template_vnc_length:
template_vnc_length = _template_vnc_length
# template_vnc_bind column
_template_vnc_bind_length = len(str(template['vnc_bind'])) + 1
if _template_vnc_bind_length > template_vnc_bind_length:
template_vnc_bind_length = _template_vnc_bind_length
# template_node_limit column
_template_node_limit_length = len(str(template['node_limit'])) + 1
if _template_node_limit_length > template_node_limit_length:
template_node_limit_length = _template_node_limit_length
# template_node_selector column
_template_node_selector_length = len(str(template['node_selector'])) + 1
if _template_node_selector_length > template_node_selector_length:
template_node_selector_length = _template_node_selector_length
# template_node_autostart column
_template_node_autostart_length = len(str(template['node_autostart'])) + 1
if _template_node_autostart_length > template_node_autostart_length:
template_node_autostart_length = _template_node_autostart_length
# Format the string (header)
template_list_output.append(
'{bold}{template_name: <{template_name_length}} {template_id: <{template_id_length}} \
{template_vcpu: <{template_vcpu_length}} \
{template_vram: <{template_vram_length}} \
Consoles: {template_serial: <{template_serial_length}} \
{template_vnc: <{template_vnc_length}} \
{template_vnc_bind: <{template_vnc_bind_length}} \
Metadata: {template_node_limit: <{template_node_limit_length}} \
{template_node_selector: <{template_node_selector_length}} \
{template_node_autostart: <{template_node_autostart_length}}{end_bold}'.format(
template_name_length=template_name_length,
template_id_length=template_id_length,
template_vcpu_length=template_vcpu_length,
template_vram_length=template_vram_length,
template_serial_length=template_serial_length,
template_vnc_length=template_vnc_length,
template_vnc_bind_length=template_vnc_bind_length,
template_node_limit_length=template_node_limit_length,
template_node_selector_length=template_node_selector_length,
template_node_autostart_length=template_node_autostart_length,
bold=ansiprint.bold(),
end_bold=ansiprint.end(),
template_state_colour='',
end_colour='',
template_name='Name',
template_id='ID',
template_vcpu='vCPUs',
template_vram='vRAM [MB]',
template_serial='Serial',
template_vnc='VNC',
template_vnc_bind='VNC bind',
template_node_limit='Limit',
template_node_selector='Selector',
template_node_autostart='Autostart'
)
)
# Keep track of nets we found to be valid to cut down on duplicate API hits
valid_net_list = []
# Format the string (elements)
for template in template_data:
template_list_output.append(
'{bold}{template_name: <{template_name_length}} {template_id: <{template_id_length}} \
{template_vcpu: <{template_vcpu_length}} \
{template_vram: <{template_vram_length}} \
{template_serial: <{template_serial_length}} \
{template_vnc: <{template_vnc_length}} \
{template_vnc_bind: <{template_vnc_bind_length}} \
{template_node_limit: <{template_node_limit_length}} \
{template_node_selector: <{template_node_selector_length}} \
{template_node_autostart: <{template_node_autostart_length}}{end_bold}'.format(
template_name_length=template_name_length,
template_id_length=template_id_length,
template_vcpu_length=template_vcpu_length,
template_vram_length=template_vram_length,
template_serial_length=template_serial_length,
template_vnc_length=template_vnc_length,
template_vnc_bind_length=template_vnc_bind_length,
template_node_limit_length=template_node_limit_length,
template_node_selector_length=template_node_selector_length,
template_node_autostart_length=template_node_autostart_length,
bold='',
end_bold='',
template_name=str(template['name']),
template_id=str(template['id']),
template_vcpu=str(template['vcpu_count']),
template_vram=str(template['vram_mb']),
template_serial=str(template['serial']),
template_vnc=str(template['vnc']),
template_vnc_bind=str(template['vnc_bind']),
template_node_limit=str(template['node_limit']),
template_node_selector=str(template['node_selector']),
template_node_autostart=str(template['node_autostart'])
)
)
click.echo('\n'.join(sorted(template_list_output)))
return True, ''
def format_list_template_network(template_data):
if isinstance(template_data, dict):
template_data = [ template_data ]
template_list_output = []
# Determine optimal column widths
template_name_length = 5
template_id_length = 4
template_mac_template_length = 13
template_networks_length = 10
for template in template_data:
# Join the networks elements into a single list of VNIs
network_list = list()
for network in template['networks']:
network_list.append(str(network['vni']))
template['networks_csv'] = ','.join(network_list)
for template in template_data:
# template_name column
_template_name_length = len(str(template['name'])) + 1
if _template_name_length > template_name_length:
template_name_length = _template_name_length
# template_id column
_template_id_length = len(str(template['id'])) + 1
if _template_id_length > template_id_length:
template_id_length = _template_id_length
# template_mac_template column
_template_mac_template_length = len(str(template['mac_template'])) + 1
if _template_mac_template_length > template_mac_template_length:
template_mac_template_length = _template_mac_template_length
# template_networks column
_template_networks_length = len(str(template['networks_csv'])) + 1
if _template_networks_length > template_networks_length:
template_networks_length = _template_networks_length
# Format the string (header)
template_list_output.append(
'{bold}{template_name: <{template_name_length}} {template_id: <{template_id_length}} \
{template_mac_template: <{template_mac_template_length}} \
{template_networks: <{template_networks_length}}{end_bold}'.format(
template_name_length=template_name_length,
template_id_length=template_id_length,
template_mac_template_length=template_mac_template_length,
template_networks_length=template_networks_length,
bold=ansiprint.bold(),
end_bold=ansiprint.end(),
template_state_colour='',
end_colour='',
template_name='Name',
template_id='ID',
template_mac_template='MAC template',
template_networks='Network VNIs'
)
)
# Format the string (elements)
for template in template_data:
template_list_output.append(
'{bold}{template_name: <{template_name_length}} {template_id: <{template_id_length}} \
{template_mac_template: <{template_mac_template_length}} \
{template_networks: <{template_networks_length}}{end_bold}'.format(
template_name_length=template_name_length,
template_id_length=template_id_length,
template_mac_template_length=template_mac_template_length,
template_networks_length=template_networks_length,
bold='',
end_bold='',
template_state_colour='',
end_colour='',
template_name=str(template['name']),
template_id=str(template['id']),
template_mac_template=str(template['mac_template']),
template_networks=str(template['networks_csv'])
)
)
click.echo('\n'.join(sorted(template_list_output)))
return True, ''
def vm_info(config, vm):
"""
Get information about VM
API endpoint: GET /api/v1/vm/{vm}
API arguments:
API schema: {json_data_object}
"""
request_uri = get_request_uri(config, '/vm/{vm}'.format(vm=vm))
response = requests.get(
request_uri
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
return True, response.json()
else:
return False, response.json()['message']
def vm_list(config, limit, target_node, target_state):
"""
Get list information about nodes (limited by {limit}, {target_node}, or {target_state})
API endpoint: GET /api/v1/vm
API arguments: limit={limit}, node={target_node}, state={target_state}
API schema: [{json_data_object},{json_data_object},etc.]
"""
params = dict()
if limit:
params['limit'] = limit
if target_node:
params['node'] = target_node
if target_state:
params['state'] = target_state
request_uri = get_request_uri(config, '/vm')
response = requests.get(
request_uri,
params=params
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
return True, response.json()
else:
return False, response.json()['message']
def vm_define(config, xml, node, node_limit, node_selector, node_autostart):
"""
Define a new VM on the cluster
API endpoint: POST /vm
API arguments: xml={xml}, node={node}, limit={node_limit}, selector={node_selector}, autostart={node_autostart}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm')
response = requests.post(
request_uri,
params={
'xml': xml,
'node': node,
'limit': node_limit,
'selector': node_selector,
'autostart': node_autostart
}
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_modify(config, vm, xml, restart):
"""
Modify the configuration of VM
API endpoint: POST /vm/{vm}
API arguments: xml={xml}, restart={restart}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}'.format(vm=vm))
response = requests.post(
request_uri,
params={
'xml': xml,
'restart': restart
}
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_metadata(config, vm, node_limit, node_selector, node_autostart):
"""
Modify PVC metadata of a VM
API endpoint: GET /vm/{vm}/meta, POST /vm/{vm}/meta
API arguments: limit={node_limit}, selector={node_selector}, autostart={node_autostart}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/meta'.format(vm=vm))
# Get the existing metadata so we can perform a fully dynamic update
response = requests.get(
request_uri
)
if config['debug']:
print('API endpoint: GET {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
metadata = response.json()
# Update any params that we've sent
if node_limit is not None:
metadata['node_limit'] = node_limit
else:
# Collapse the existing list back down to a CSV
metadata['node_limit'] = ','.join(metadata['node_limit'])
if node_selector is not None:
metadata['node_selector'] = node_selector
if node_autostart is not None:
metadata['node_autostart'] = node_autostart
# Write the new metadata
print(metadata['node_limit'])
response = requests.post(
request_uri,
params={
'limit': metadata['node_limit'],
'selector': metadata['node_selector'],
'autostart': metadata['node_autostart']
}
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_remove(config, vm, delete_disks=False):
"""
Remove a VM
API endpoint: DELETE /vm/{vm}
API arguments: delete_disks={delete_disks}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}'.format(vm=vm))
response = requests.delete(
request_uri,
params={
'delete_disks': delete_disks
}
)
if config['debug']:
print('API endpoint: DELETE {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_state(config, vm, target_state):
"""
Modify the current state of VM
API endpoint: POST /vm/{vm}/state
API arguments: state={state}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/state'.format(vm=vm))
response = requests.post(
request_uri,
params={
'state': target_state,
}
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_node(config, vm, target_node, action, force=False):
"""
Modify the current node of VM via {action}
API endpoint: POST /vm/{vm}/node
API arguments: node={target_node}, action={action}, force={force}
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/node'.format(vm=vm))
response = requests.post(
request_uri,
params={
'node': target_node,
'action': action,
'force': force
}
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def vm_locks(config, vm):
"""
Flush RBD locks of (stopped) VM
API endpoint: POST /vm/{vm}/locks
API arguments:
API schema: {"message":"{data}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/locks'.format(vm=vm))
response = requests.post(
request_uri
)
if config['debug']:
print('API endpoint: POST {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
if response.status_code == 200:
retstatus = True
else:
retstatus = False
return retstatus, response.json()['message']
def view_console_log(config, vm, lines=100):
"""
Return console log lines from the API and display them in a pager
API endpoint: GET /vm/{vm}/console
API arguments: lines={lines}
API schema: {"name":"{vmname}","data":"{console_log}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/console'.format(vm=vm))
response = requests.get(
request_uri,
params={'lines': lines}
)
if config['debug']:
print('API endpoint: GET {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
console_log = response.json()['data']
# Shrink the log buffer to length lines
shrunk_log = console_log.split('\n')[-lines:]
loglines = '\n'.join(shrunk_log)
# Show it in the pager (less)
try:
pager = subprocess.Popen(['less', '-R'], stdin=subprocess.PIPE)
pager.communicate(input=loglines.encode('utf8'))
except FileNotFoundError:
click.echo("Error: `less` pager not found, dumping log ({} lines) to stdout".format(lines))
return True, loglines
return True, ''
def follow_console_log(config, vm, lines=10):
"""
Return and follow console log lines from the API
API endpoint: GET /vm/{vm}/console
API arguments: lines={lines}
API schema: {"name":"{vmname}","data":"{console_log}"}
"""
request_uri = get_request_uri(config, '/vm/{vm}/console'.format(vm=vm))
response = requests.get(
request_uri,
params={'lines': lines}
)
if config['debug']:
print('API endpoint: GET {}'.format(request_uri))
print('Response code: {}'.format(response.status_code))
print('Response headers: {}'.format(response.headers))
console_log = response.json()['data']
# Shrink the log buffer to length lines
shrunk_log = console_log.split('\n')[-lines:]
loglines = '\n'.join(shrunk_log)
# Print the initial data and begin following
print(loglines, end='')
while True:
# Grab the next line set
# Get the (initial) data from the API
response = requests.get(
'{}://{}{}{}'.format(
config['api_scheme'],
config['api_host'],
config['api_prefix'],
'/vm/{}/console'.format(vm)
),
params={'lines': lines}
)
if config['debug']:
print(
'Response code: {}'.format(
response.status_code
)
)
print(
'Response headers: {}'.format(
response.headers
)
)
new_console_log = response.json()['data']
# Split the new and old log strings into constitutent lines
old_console_loglines = console_log.split('\n')
new_console_loglines = new_console_log.split('\n')
# Set the console log to the new log value for the next iteration
console_log = new_console_log
# Remove the lines from the old log until we hit the first line of the new log; this
# ensures that the old log is a string that we can remove from the new log entirely
for index, line in enumerate(old_console_loglines, start=0):
if line == new_console_loglines[0]:
del old_console_loglines[0:index]
break
# Rejoin the log lines into strings
old_console_log = '\n'.join(old_console_loglines)
new_console_log = '\n'.join(new_console_loglines)
# Remove the old lines from the new log
diff_console_log = new_console_log.replace(old_console_log, "")
# If there's a difference, print it out
if diff_console_log:
print(diff_console_log, end='')
# Wait a second
time.sleep(1)
return True, ''
#
# Output display functions
#
def format_info(config, domain_information, long_output):
# Format a nice output; do this line-by-line then concat the elements at the end
ainformation = []
ainformation.append('{}Virtual machine information:{}'.format(ansiprint.bold(), ansiprint.end()))
ainformation.append('')
# Basic information
ainformation.append('{}UUID:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['uuid']))
ainformation.append('{}Name:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['name']))
ainformation.append('{}Description:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['description']))
ainformation.append('{}Profile:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['profile']))
ainformation.append('{}Memory (M):{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['memory']))
ainformation.append('{}vCPUs:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['vcpu']))
ainformation.append('{}Topology (S/C/T):{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['vcpu_topology']))
if long_output == True:
# Virtualization information
ainformation.append('')
ainformation.append('{}Emulator:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['emulator']))
ainformation.append('{}Type:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['type']))
ainformation.append('{}Arch:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['arch']))
ainformation.append('{}Machine:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['machine']))
ainformation.append('{}Features:{} {}'.format(ansiprint.purple(), ansiprint.end(), ' '.join(domain_information['features'])))
# PVC cluster information
ainformation.append('')
dstate_colour = {
'start': ansiprint.green(),
'restart': ansiprint.yellow(),
'shutdown': ansiprint.yellow(),
'stop': ansiprint.red(),
'disable': ansiprint.blue(),
'fail': ansiprint.red(),
'migrate': ansiprint.blue(),
'unmigrate': ansiprint.blue()
}
ainformation.append('{}State:{} {}{}{}'.format(ansiprint.purple(), ansiprint.end(), dstate_colour[domain_information['state']], domain_information['state'], ansiprint.end()))
ainformation.append('{}Current Node:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['node']))
if not domain_information['last_node']:
domain_information['last_node'] = "N/A"
ainformation.append('{}Previous Node:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['last_node']))
# Get a failure reason if applicable
if domain_information['failed_reason']:
ainformation.append('')
ainformation.append('{}Failure reason:{} {}'.format(ansiprint.purple(), ansiprint.end(), domain_information['failed_reason']))
if not domain_information['node_selector']:
formatted_node_selector = "False"
else:
formatted_node_selector = domain_information['node_selector']
if not domain_information['node_limit']:
formatted_node_limit = "False"
else:
formatted_node_limit = ', '.join(domain_information['node_limit'])
if not domain_information['node_autostart']:
formatted_node_autostart = "False"
else:
formatted_node_autostart = domain_information['node_autostart']
ainformation.append('{}Migration selector:{} {}'.format(ansiprint.purple(), ansiprint.end(), formatted_node_selector))
ainformation.append('{}Node limit:{} {}'.format(ansiprint.purple(), ansiprint.end(), formatted_node_limit))
ainformation.append('{}Autostart:{} {}'.format(ansiprint.purple(), ansiprint.end(), formatted_node_autostart))
# Network list
net_list = []
for net in domain_information['networks']:
# Split out just the numerical (VNI) part of the brXXXX name
net_vnis = re.findall(r'\d+', net['source'])
if net_vnis:
net_vni = net_vnis[0]
else:
net_vni = re.sub('br', '', net['source'])
request_uri = get_request_uri(config, '/network/{net}'.format(net=net_vni))
response = requests.get(
request_uri
)
if response.status_code != 200 and net_vni != 'cluster':
net_list.append(ansiprint.red() + net_vni + ansiprint.end() + ' [invalid]')
else:
net_list.append(net_vni)
ainformation.append('')
ainformation.append('{}Networks:{} {}'.format(ansiprint.purple(), ansiprint.end(), ', '.join(net_list)))
if long_output == True:
# Disk list
ainformation.append('')
name_length = 0
for disk in domain_information['disks']:
_name_length = len(disk['name']) + 1
if _name_length > name_length:
name_length = _name_length
ainformation.append('{0}Disks:{1} {2}ID Type {3: <{width}} Dev Bus{4}'.format(ansiprint.purple(), ansiprint.end(), ansiprint.bold(), 'Name', ansiprint.end(), width=name_length))
for disk in domain_information['disks']:
ainformation.append(' {0: <3} {1: <5} {2: <{width}} {3: <4} {4: <5}'.format(domain_information['disks'].index(disk), disk['type'], disk['name'], disk['dev'], disk['bus'], width=name_length))
ainformation.append('')
ainformation.append('{}Interfaces:{} {}ID Type Source Model MAC{}'.format(ansiprint.purple(), ansiprint.end(), ansiprint.bold(), ansiprint.end()))
for net in domain_information['networks']:
ainformation.append(' {0: <3} {1: <8} {2: <10} {3: <8} {4}'.format(domain_information['networks'].index(net), net['type'], net['source'], net['model'], net['mac']))
# Controller list
ainformation.append('')
ainformation.append('{}Controllers:{} {}ID Type Model{}'.format(ansiprint.purple(), ansiprint.end(), ansiprint.bold(), ansiprint.end()))
for controller in domain_information['controllers']:
ainformation.append(' {0: <3} {1: <14} {2: <8}'.format(domain_information['controllers'].index(controller), controller['type'], controller['model']))
# Join it all together
information = '\n'.join(ainformation)
click.echo(information)
click.echo('')
def format_list(config, vm_list, raw):
# Function to strip the "br" off of nets and return a nicer list
def getNiceNetID(domain_information):
# Network list
net_list = []
for net in domain_information['networks']:
# Split out just the numerical (VNI) part of the brXXXX name
net_vnis = re.findall(r'\d+', net['source'])
if net_vnis:
net_vni = net_vnis[0]
else:
net_vni = re.sub('br', '', net['source'])
net_list.append(net_vni)
return net_list
# Handle raw mode since it just lists the names
if raw:
for vm in sorted(item['name'] for item in vm_list):
click.echo(vm)
return True, ''
vm_list_output = []
# Determine optimal column widths
# Dynamic columns: node_name, node, migrated
vm_name_length = 5
vm_uuid_length = 37
vm_state_length = 6
vm_nets_length = 9
vm_ram_length = 8
vm_vcpu_length = 6
vm_node_length = 8
vm_migrated_length = 10
for domain_information in vm_list:
net_list = getNiceNetID(domain_information)
# vm_name column
_vm_name_length = len(domain_information['name']) + 1
if _vm_name_length > vm_name_length:
vm_name_length = _vm_name_length
# vm_state column
_vm_state_length = len(domain_information['state']) + 1
if _vm_state_length > vm_state_length:
vm_state_length = _vm_state_length
# vm_nets column
_vm_nets_length = len(','.join(net_list)) + 1
if _vm_nets_length > vm_nets_length:
vm_nets_length = _vm_nets_length
# vm_node column
_vm_node_length = len(domain_information['node']) + 1
if _vm_node_length > vm_node_length:
vm_node_length = _vm_node_length
# vm_migrated column
_vm_migrated_length = len(domain_information['migrated']) + 1
if _vm_migrated_length > vm_migrated_length:
vm_migrated_length = _vm_migrated_length
# Format the string (header)
vm_list_output.append(
'{bold}{vm_name: <{vm_name_length}} {vm_uuid: <{vm_uuid_length}} \
{vm_state_colour}{vm_state: <{vm_state_length}}{end_colour} \
{vm_networks: <{vm_nets_length}} \
{vm_memory: <{vm_ram_length}} {vm_vcpu: <{vm_vcpu_length}} \
{vm_node: <{vm_node_length}} \
{vm_migrated: <{vm_migrated_length}}{end_bold}'.format(
vm_name_length=vm_name_length,
vm_uuid_length=vm_uuid_length,
vm_state_length=vm_state_length,
vm_nets_length=vm_nets_length,
vm_ram_length=vm_ram_length,
vm_vcpu_length=vm_vcpu_length,
vm_node_length=vm_node_length,
vm_migrated_length=vm_migrated_length,
bold=ansiprint.bold(),
end_bold=ansiprint.end(),
vm_state_colour='',
end_colour='',
vm_name='Name',
vm_uuid='UUID',
vm_state='State',
vm_networks='Networks',
vm_memory='RAM (M)',
vm_vcpu='vCPUs',
vm_node='Node',
vm_migrated='Migrated'
)
)
# Keep track of nets we found to be valid to cut down on duplicate API hits
valid_net_list = []
# Format the string (elements)
for domain_information in vm_list:
if domain_information['state'] == 'start':
vm_state_colour = ansiprint.green()
elif domain_information['state'] == 'restart':
vm_state_colour = ansiprint.yellow()
elif domain_information['state'] == 'shutdown':
vm_state_colour = ansiprint.yellow()
elif domain_information['state'] == 'stop':
vm_state_colour = ansiprint.red()
elif domain_information['state'] == 'fail':
vm_state_colour = ansiprint.red()
else:
vm_state_colour = ansiprint.blue()
# Handle colouring for an invalid network config
raw_net_list = getNiceNetID(domain_information)
net_list = []
vm_net_colour = ''
for net_vni in raw_net_list:
if not net_vni in valid_net_list:
request_uri = get_request_uri(config, '/network/{net}'.format(net=net_vni))
response = requests.get(
request_uri
)
if response.status_code != 200 and net_vni != 'cluster':
vm_net_colour = ansiprint.red()
else:
valid_net_list.append(net_vni)
net_list.append(net_vni)
vm_list_output.append(
'{bold}{vm_name: <{vm_name_length}} {vm_uuid: <{vm_uuid_length}} \
{vm_state_colour}{vm_state: <{vm_state_length}}{end_colour} \
{vm_net_colour}{vm_networks: <{vm_nets_length}}{end_colour} \
{vm_memory: <{vm_ram_length}} {vm_vcpu: <{vm_vcpu_length}} \
{vm_node: <{vm_node_length}} \
{vm_migrated: <{vm_migrated_length}}{end_bold}'.format(
vm_name_length=vm_name_length,
vm_uuid_length=vm_uuid_length,
vm_state_length=vm_state_length,
vm_nets_length=vm_nets_length,
vm_ram_length=vm_ram_length,
vm_vcpu_length=vm_vcpu_length,
vm_node_length=vm_node_length,
vm_migrated_length=vm_migrated_length,
bold='',
end_bold='',
vm_state_colour=vm_state_colour,
end_colour=ansiprint.end(),
vm_name=domain_information['name'],
vm_uuid=domain_information['uuid'],
vm_state=domain_information['state'],
vm_net_colour=vm_net_colour,
vm_networks=','.join(net_list),
vm_memory=domain_information['memory'],
vm_vcpu=domain_information['vcpu'],
vm_node=domain_information['node'],
vm_migrated=domain_information['migrated']
)
)
click.echo('\n'.join(sorted(vm_list_output)))
return True, ''