pvc/daemon-common/vm.py

1474 lines
51 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
# vm.py - PVC client function library, VM fuctions
# Part of the Parallel Virtual Cluster (PVC) system
#
2022-10-06 11:55:27 -04:00
# Copyright (C) 2018-2022 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, version 3.
#
# 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 os.path
import lxml.objectify
import lxml.etree
from distutils.util import strtobool
from uuid import UUID
from concurrent.futures import ThreadPoolExecutor
from datetime import datetime
from json import dump as jdump
import daemon_lib.common as common
import daemon_lib.ceph as ceph
2021-06-21 22:21:54 -04:00
from daemon_lib.network import set_sriov_vf_vm, unset_sriov_vf_vm
#
# Cluster search functions
#
2021-05-29 21:17:19 -04:00
def getClusterDomainList(zkhandler):
# Get a list of UUIDs by listing the children of /domains
uuid_list = zkhandler.children("base.domain")
name_list = []
# For each UUID, get the corresponding name from the data
for uuid in uuid_list:
name_list.append(zkhandler.read(("domain", uuid)))
return uuid_list, name_list
2021-05-29 21:17:19 -04:00
def searchClusterByUUID(zkhandler, uuid):
try:
# Get the lists
2021-05-29 21:17:19 -04:00
uuid_list, name_list = getClusterDomainList(zkhandler)
# We're looking for UUID, so find that element ID
index = uuid_list.index(uuid)
# Get the name_list element at that index
name = name_list[index]
except ValueError:
# We didn't find anything
return None
return name
2021-05-29 21:17:19 -04:00
def searchClusterByName(zkhandler, name):
try:
# Get the lists
2021-05-29 21:17:19 -04:00
uuid_list, name_list = getClusterDomainList(zkhandler)
# We're looking for name, so find that element ID
index = name_list.index(name)
# Get the uuid_list element at that index
uuid = uuid_list[index]
except ValueError:
# We didn't find anything
return None
return uuid
2021-05-29 21:17:19 -04:00
def getDomainUUID(zkhandler, domain):
# Validate that VM exists in cluster
if common.validateUUID(domain):
2021-05-29 21:17:19 -04:00
dom_name = searchClusterByUUID(zkhandler, domain)
dom_uuid = searchClusterByName(zkhandler, dom_name)
else:
2021-05-29 21:17:19 -04:00
dom_uuid = searchClusterByName(zkhandler, domain)
dom_name = searchClusterByUUID(zkhandler, dom_uuid)
return dom_uuid
2021-05-29 21:17:19 -04:00
def getDomainName(zkhandler, domain):
# Validate that VM exists in cluster
2018-09-25 02:26:37 -04:00
if common.validateUUID(domain):
2021-05-29 21:17:19 -04:00
dom_name = searchClusterByUUID(zkhandler, domain)
dom_uuid = searchClusterByName(zkhandler, dom_name)
2018-09-25 02:26:37 -04:00
else:
2021-05-29 21:17:19 -04:00
dom_uuid = searchClusterByName(zkhandler, domain)
dom_name = searchClusterByUUID(zkhandler, dom_uuid)
2018-09-25 02:26:37 -04:00
return dom_name
2021-05-29 21:17:19 -04:00
#
# Helper functions
#
def change_state(zkhandler, dom_uuid, new_state):
lock = zkhandler.exclusivelock(("domain.state", dom_uuid))
with lock:
zkhandler.write([(("domain.state", dom_uuid), new_state)])
2021-05-29 21:17:19 -04:00
# Wait for 1/2 second to allow state to flow to all nodes
time.sleep(0.5)
2021-05-29 21:17:19 -04:00
#
# Direct functions
#
2021-05-29 21:17:19 -04:00
def is_migrated(zkhandler, domain):
2019-07-25 14:33:50 -04:00
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
2019-07-25 14:33:50 -04:00
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
last_node = zkhandler.read(("domain.last_node", dom_uuid))
2019-07-25 14:33:50 -04:00
if last_node:
return True
2019-07-25 15:45:45 -04:00
else:
2019-07-25 14:33:50 -04:00
return False
2021-05-29 21:17:19 -04:00
def flush_locks(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Verify that the VM is in a stopped state; freeing locks is not safe otherwise
state = zkhandler.read(("domain.state", dom_uuid))
if state != "stop":
return (
False,
'ERROR: VM "{}" is not in stopped state; flushing RBD locks on a running VM is dangerous.'.format(
domain
),
)
# Tell the cluster to create a new OSD for the host
flush_locks_string = "flush_locks {}".format(dom_uuid)
zkhandler.write([("base.cmd.domain", flush_locks_string)])
# Wait 1/2 second for the cluster to get the message and start working
time.sleep(0.5)
# Acquire a read lock, so we get the return exclusively
lock = zkhandler.readlock("base.cmd.domain")
with lock:
try:
result = zkhandler.read("base.cmd.domain").split()[0]
if result == "success-flush_locks":
message = 'Flushed locks on VM "{}"'.format(domain)
success = True
else:
message = 'ERROR: Failed to flush locks on VM "{}"; check node logs for details.'.format(
domain
)
success = False
except Exception:
message = "ERROR: Command ignored by node."
success = False
# Acquire a write lock to ensure things go smoothly
lock = zkhandler.writelock("base.cmd.domain")
with lock:
time.sleep(0.5)
zkhandler.write([("base.cmd.domain", "")])
return success, message
def define_vm(
zkhandler,
config_data,
target_node,
node_limit,
node_selector,
node_autostart,
migration_method=None,
profile=None,
tags=[],
initial_state="stop",
):
# Parse the XML data
try:
parsed_xml = lxml.objectify.fromstring(config_data)
except Exception:
return False, "ERROR: Failed to parse XML data."
dom_uuid = parsed_xml.uuid.text
dom_name = parsed_xml.name.text
# Ensure that the UUID and name are unique
if searchClusterByUUID(zkhandler, dom_uuid) or searchClusterByName(
zkhandler, dom_name
):
return (
False,
'ERROR: Specified VM "{}" or UUID "{}" matches an existing VM on the cluster'.format(
dom_name, dom_uuid
),
)
2019-06-24 13:37:56 -04:00
if not target_node:
2021-05-29 21:17:19 -04:00
target_node = common.findTargetNode(zkhandler, dom_uuid)
2019-06-24 13:25:24 -04:00
else:
# Verify node is valid
2021-05-29 21:17:19 -04:00
valid_node = common.verifyNode(zkhandler, target_node)
2019-06-24 13:25:24 -04:00
if not valid_node:
return False, 'ERROR: Specified node "{}" is invalid.'.format(target_node)
# Validate the new RAM against the current active node
node_total_memory = int(zkhandler.read(("node.memory.total", target_node)))
if int(parsed_xml.memory.text) >= node_total_memory:
return (
False,
'ERROR: VM configuration specifies more memory ({} MiB) than node "{}" has available ({} MiB).'.format(
parsed_xml.memory.text, target_node, node_total_memory
),
)
# Validate the number of vCPUs against the current active node
node_total_cpus = int(zkhandler.read(("node.data.static", target_node)).split()[0])
if (node_total_cpus - 2) <= int(parsed_xml.vcpu.text):
return (
False,
'ERROR: VM configuration specifies more vCPUs ({}) than node "{}" has available ({} minus 2).'.format(
parsed_xml.vcpu.text, target_node, node_total_cpus
),
)
2021-06-21 22:21:54 -04:00
# If a SR-IOV network device is being added, set its used state
dnetworks = common.getDomainNetworks(parsed_xml, {})
for network in dnetworks:
if network["type"] in ["direct", "hostdev"]:
dom_node = zkhandler.read(("domain.node", dom_uuid))
2021-06-21 22:21:54 -04:00
# Check if the network is already in use
is_used = zkhandler.read(
("node.sriov.vf", dom_node, "sriov_vf.used", network["source"])
)
if is_used == "True":
used_by_name = searchClusterByUUID(
zkhandler,
zkhandler.read(
(
"node.sriov.vf",
dom_node,
"sriov_vf.used_by",
network["source"],
)
),
)
return (
False,
'ERROR: Attempted to use SR-IOV network "{}" which is already used by VM "{}" on node "{}".'.format(
network["source"], used_by_name, dom_node
),
)
2021-06-21 22:21:54 -04:00
# We must update the "used" section
set_sriov_vf_vm(
zkhandler,
dom_uuid,
dom_node,
network["source"],
network["mac"],
network["type"],
)
2021-06-21 22:21:54 -04:00
# Obtain the RBD disk list using the common functions
2020-06-07 00:40:21 -04:00
ddisks = common.getDomainDisks(parsed_xml, {})
rbd_list = []
for disk in ddisks:
if disk["type"] == "rbd":
rbd_list.append(disk["name"])
# Join the limit
if isinstance(node_limit, list) and node_limit:
formatted_node_limit = ",".join(node_limit)
else:
formatted_node_limit = ""
# Join the RBD list
if isinstance(rbd_list, list) and rbd_list:
formatted_rbd_list = ",".join(rbd_list)
else:
formatted_rbd_list = ""
# Add the new domain to Zookeeper
zkhandler.write(
[
(("domain", dom_uuid), dom_name),
(("domain.xml", dom_uuid), config_data),
(("domain.state", dom_uuid), initial_state),
(("domain.profile", dom_uuid), profile),
(("domain.stats", dom_uuid), ""),
(("domain.node", dom_uuid), target_node),
(("domain.last_node", dom_uuid), ""),
(("domain.failed_reason", dom_uuid), ""),
(("domain.storage.volumes", dom_uuid), formatted_rbd_list),
(("domain.console.log", dom_uuid), ""),
(("domain.console.vnc", dom_uuid), ""),
(("domain.meta.autostart", dom_uuid), node_autostart),
(("domain.meta.migrate_method", dom_uuid), str(migration_method).lower()),
(("domain.meta.node_limit", dom_uuid), formatted_node_limit),
(("domain.meta.node_selector", dom_uuid), str(node_selector).lower()),
(("domain.meta.tags", dom_uuid), ""),
(("domain.migrate.sync_lock", dom_uuid), ""),
]
)
for tag in tags:
tag_name = tag["name"]
zkhandler.write(
[
(("domain.meta.tags", dom_uuid, "tag.name", tag_name), tag["name"]),
(("domain.meta.tags", dom_uuid, "tag.type", tag_name), tag["type"]),
(
("domain.meta.tags", dom_uuid, "tag.protected", tag_name),
tag["protected"],
),
]
)
return True, 'Added new VM with Name "{}" and UUID "{}" to database.'.format(
dom_name, dom_uuid
)
def attach_vm_device(zkhandler, domain, device_spec_xml):
# Validate that VM exists in cluster
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Verify that the VM is in a stopped state; freeing locks is not safe otherwise
state = zkhandler.read(("domain.state", dom_uuid))
if state != "start":
return (
False,
'ERROR: VM "{}" is not in started state; live-add unneccessary.'.format(
domain
),
)
# Tell the cluster to attach the device
attach_device_string = "attach_device {} {}".format(dom_uuid, device_spec_xml)
zkhandler.write([("base.cmd.domain", attach_device_string)])
# Wait 1/2 second for the cluster to get the message and start working
time.sleep(0.5)
# Acquire a read lock, so we get the return exclusively
lock = zkhandler.readlock("base.cmd.domain")
with lock:
try:
result = zkhandler.read("base.cmd.domain").split()[0]
if result == "success-attach_device":
message = 'Attached device on VM "{}"'.format(domain)
success = True
else:
message = 'ERROR: Failed to attach device on VM "{}"; check node logs for details.'.format(
domain
)
success = False
except Exception:
message = "ERROR: Command ignored by node."
success = False
# Acquire a write lock to ensure things go smoothly
lock = zkhandler.writelock("base.cmd.domain")
with lock:
time.sleep(0.5)
zkhandler.write([("base.cmd.domain", "")])
return success, message
def detach_vm_device(zkhandler, domain, device_spec_xml):
# Validate that VM exists in cluster
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Verify that the VM is in a stopped state; freeing locks is not safe otherwise
state = zkhandler.read(("domain.state", dom_uuid))
if state != "start":
return (
False,
'ERROR: VM "{}" is not in started state; live-add unneccessary.'.format(
domain
),
)
# Tell the cluster to detach the device
detach_device_string = "detach_device {} {}".format(dom_uuid, device_spec_xml)
zkhandler.write([("base.cmd.domain", detach_device_string)])
# Wait 1/2 second for the cluster to get the message and start working
time.sleep(0.5)
# Acquire a read lock, so we get the return exclusively
lock = zkhandler.readlock("base.cmd.domain")
with lock:
try:
result = zkhandler.read("base.cmd.domain").split()[0]
if result == "success-detach_device":
message = 'Attached device on VM "{}"'.format(domain)
success = True
else:
message = 'ERROR: Failed to detach device on VM "{}"; check node logs for details.'.format(
domain
)
success = False
except Exception:
message = "ERROR: Command ignored by node."
success = False
# Acquire a write lock to ensure things go smoothly
lock = zkhandler.writelock("base.cmd.domain")
with lock:
time.sleep(0.5)
zkhandler.write([("base.cmd.domain", "")])
return success, message
def modify_vm_metadata(
zkhandler,
domain,
node_limit,
node_selector,
node_autostart,
provisioner_profile,
migration_method,
):
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
2021-06-13 13:41:21 -04:00
update_list = list()
if node_limit is not None:
update_list.append((("domain.meta.node_limit", dom_uuid), node_limit))
if node_selector is not None:
update_list.append(
(("domain.meta.node_selector", dom_uuid), str(node_selector).lower())
)
if node_autostart is not None:
update_list.append((("domain.meta.autostart", dom_uuid), node_autostart))
if provisioner_profile is not None:
update_list.append((("domain.profile", dom_uuid), provisioner_profile))
if migration_method is not None:
update_list.append(
(("domain.meta.migrate_method", dom_uuid), str(migration_method).lower())
)
2021-06-13 13:41:21 -04:00
if len(update_list) < 1:
return False, "ERROR: No updates to apply."
2021-06-13 13:41:21 -04:00
zkhandler.write(update_list)
return True, 'Successfully modified PVC metadata of VM "{}".'.format(domain)
def modify_vm_tag(zkhandler, domain, action, tag, protected=False):
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
if action == "add":
zkhandler.write(
[
(("domain.meta.tags", dom_uuid, "tag.name", tag), tag),
(("domain.meta.tags", dom_uuid, "tag.type", tag), "user"),
(("domain.meta.tags", dom_uuid, "tag.protected", tag), protected),
]
)
return True, 'Successfully added tag "{}" to VM "{}".'.format(tag, domain)
elif action == "remove":
if not zkhandler.exists(("domain.meta.tags", dom_uuid, "tag", tag)):
return False, 'The tag "{}" does not exist.'.format(tag)
if zkhandler.read(("domain.meta.tags", dom_uuid, "tag.type", tag)) != "user":
return (
False,
'The tag "{}" is not a user tag and cannot be removed.'.format(tag),
)
if bool(
strtobool(
zkhandler.read(("domain.meta.tags", dom_uuid, "tag.protected", tag))
)
):
return False, 'The tag "{}" is protected and cannot be removed.'.format(tag)
zkhandler.delete([(("domain.meta.tags", dom_uuid, "tag", tag))])
return True, 'Successfully removed tag "{}" from VM "{}".'.format(tag, domain)
else:
return False, "Specified tag action is not available."
2021-05-29 21:17:19 -04:00
def modify_vm(zkhandler, domain, restart, new_vm_config):
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
2021-05-29 21:17:19 -04:00
dom_name = getDomainName(zkhandler, domain)
# Parse and valiate the XML
try:
parsed_xml = lxml.objectify.fromstring(new_vm_config)
except Exception:
return False, "ERROR: Failed to parse new XML data."
2021-06-21 22:21:54 -04:00
# Get our old network list for comparison purposes
old_vm_config = zkhandler.read(("domain.xml", dom_uuid))
old_parsed_xml = lxml.objectify.fromstring(old_vm_config)
old_dnetworks = common.getDomainNetworks(old_parsed_xml, {})
# Validate the new RAM against the current active node
node_name = zkhandler.read(("domain.node", dom_uuid))
node_total_memory = int(zkhandler.read(("node.memory.total", node_name)))
if int(parsed_xml.memory.text) >= node_total_memory:
return (
False,
'ERROR: Updated VM configuration specifies more memory ({} MiB) than node "{}" has available ({} MiB).'.format(
parsed_xml.memory.text, node_name, node_total_memory
),
)
# Validate the number of vCPUs against the current active node
node_total_cpus = int(zkhandler.read(("node.data.static", node_name)).split()[0])
if (node_total_cpus - 2) <= int(parsed_xml.vcpu.text):
return (
False,
'ERROR: Updated VM configuration specifies more vCPUs ({}) than node "{}" has available ({} minus 2).'.format(
parsed_xml.vcpu.text, node_name, node_total_cpus
),
)
2021-06-21 22:21:54 -04:00
# If a SR-IOV network device is being added, set its used state
dnetworks = common.getDomainNetworks(parsed_xml, {})
for network in dnetworks:
# Ignore networks that are already there
if network["source"] in [net["source"] for net in old_dnetworks]:
continue
if network["type"] in ["direct", "hostdev"]:
dom_node = zkhandler.read(("domain.node", dom_uuid))
2021-06-21 22:21:54 -04:00
# Check if the network is already in use
is_used = zkhandler.read(
("node.sriov.vf", dom_node, "sriov_vf.used", network["source"])
)
if is_used == "True":
used_by_name = searchClusterByUUID(
zkhandler,
zkhandler.read(
(
"node.sriov.vf",
dom_node,
"sriov_vf.used_by",
network["source"],
)
),
)
return (
False,
'ERROR: Attempted to use SR-IOV network "{}" which is already used by VM "{}" on node "{}".'.format(
network["source"], used_by_name, dom_node
),
)
2021-06-21 22:21:54 -04:00
# We must update the "used" section
set_sriov_vf_vm(
zkhandler,
dom_uuid,
dom_node,
network["source"],
network["mac"],
network["type"],
)
2021-06-21 22:21:54 -04:00
# If a SR-IOV network device is being removed, unset its used state
for network in old_dnetworks:
if network["type"] in ["direct", "hostdev"]:
if network["mac"] not in [n["mac"] for n in dnetworks]:
dom_node = zkhandler.read(("domain.node", dom_uuid))
2021-06-21 22:21:54 -04:00
# We must update the "used" section
unset_sriov_vf_vm(zkhandler, dom_node, network["source"])
# Obtain the RBD disk list using the common functions
2020-06-07 00:40:21 -04:00
ddisks = common.getDomainDisks(parsed_xml, {})
rbd_list = []
for disk in ddisks:
if disk["type"] == "rbd":
rbd_list.append(disk["name"])
# Join the RBD list
if isinstance(rbd_list, list) and rbd_list:
formatted_rbd_list = ",".join(rbd_list)
else:
formatted_rbd_list = ""
# Add the modified config to Zookeeper
zkhandler.write(
[
(("domain", dom_uuid), dom_name),
(("domain.storage.volumes", dom_uuid), formatted_rbd_list),
(("domain.xml", dom_uuid), new_vm_config),
]
)
if restart:
change_state(zkhandler, dom_uuid, "restart")
return True, 'Successfully modified configuration of VM "{}".'.format(domain)
2021-05-29 21:17:19 -04:00
def dump_vm(zkhandler, domain):
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
2019-03-12 21:09:54 -04:00
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Gram the domain XML and dump it to stdout
vm_xml = zkhandler.read(("domain.xml", dom_uuid))
2019-03-12 21:09:54 -04:00
2019-05-20 22:15:28 -04:00
return True, vm_xml
2019-03-12 21:09:54 -04:00
2021-05-29 21:17:19 -04:00
def rename_vm(zkhandler, domain, new_domain):
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Verify that the VM is in a stopped state; renaming is not supported otherwise
state = zkhandler.read(("domain.state", dom_uuid))
2023-01-30 11:48:43 -05:00
if state not in ["stop", "disable"]:
return (
False,
'ERROR: VM "{}" is not in stopped state; VMs cannot be renamed while running.'.format(
domain
),
)
# Parse and valiate the XML
2021-05-29 21:17:19 -04:00
vm_config = common.getDomainXML(zkhandler, dom_uuid)
# Obtain the RBD disk list using the common functions
ddisks = common.getDomainDisks(vm_config, {})
pool_list = []
rbd_list = []
for disk in ddisks:
if disk["type"] == "rbd":
pool_list.append(disk["name"].split("/")[0])
rbd_list.append(disk["name"].split("/")[1])
# Rename each volume in turn
for idx, rbd in enumerate(rbd_list):
rbd_new = re.sub(r"{}".format(domain), new_domain, rbd)
# Skip renaming if nothing changed
if rbd_new == rbd:
continue
2021-05-29 21:17:19 -04:00
ceph.rename_volume(zkhandler, pool_list[idx], rbd, rbd_new)
# Replace the name in the config
vm_config_new = (
lxml.etree.tostring(vm_config, encoding="ascii", method="xml")
.decode()
.replace(domain, new_domain)
)
# Get VM information
2021-05-29 21:17:19 -04:00
_b, dom_info = get_info(zkhandler, dom_uuid)
# Undefine the old VM
2021-05-29 21:17:19 -04:00
undefine_vm(zkhandler, dom_uuid)
# Define the new VM
define_vm(
zkhandler,
vm_config_new,
dom_info["node"],
dom_info["node_limit"],
dom_info["node_selector"],
dom_info["node_autostart"],
migration_method=dom_info["migration_method"],
profile=dom_info["profile"],
tags=dom_info["tags"],
initial_state="stop",
)
# If the VM is migrated, store that
if dom_info["migrated"] != "no":
zkhandler.write([(("domain.last_node", dom_uuid), dom_info["last_node"])])
return True, 'Successfully renamed VM "{}" to "{}".'.format(domain, new_domain)
2021-05-29 21:17:19 -04:00
def undefine_vm(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Shut down the VM
current_vm_state = zkhandler.read(("domain.state", dom_uuid))
if current_vm_state != "stop":
change_state(zkhandler, dom_uuid, "stop")
# Gracefully terminate the class instances
change_state(zkhandler, dom_uuid, "delete")
# Delete the configurations
zkhandler.delete([("domain", dom_uuid)])
return True, 'Undefined VM "{}" from the cluster.'.format(domain)
2021-05-29 21:17:19 -04:00
def remove_vm(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
2021-05-29 21:17:19 -04:00
disk_list = common.getDomainDiskList(zkhandler, dom_uuid)
# Shut down the VM
current_vm_state = zkhandler.read(("domain.state", dom_uuid))
if current_vm_state != "stop":
change_state(zkhandler, dom_uuid, "stop")
2021-05-29 21:17:19 -04:00
# Wait for 1 second to allow state to flow to all nodes
time.sleep(1)
# Remove disks
for disk in disk_list:
# vmpool/vmname_volume
try:
disk_pool, disk_name = disk.split("/")
except ValueError:
continue
retcode, message = ceph.remove_volume(zkhandler, disk_pool, disk_name)
if not retcode:
if re.match("^ERROR: No volume with name", message):
continue
else:
return False, message
# Gracefully terminate the class instances
change_state(zkhandler, dom_uuid, "delete")
# Wait for 1/2 second to allow state to flow to all nodes
time.sleep(0.5)
# Delete the VM configuration from Zookeeper
zkhandler.delete([("domain", dom_uuid)])
return True, 'Removed VM "{}" and its disks from the cluster.'.format(domain)
2021-05-29 21:17:19 -04:00
def start_vm(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Set the VM to start
change_state(zkhandler, dom_uuid, "start")
return True, 'Starting VM "{}".'.format(domain)
2021-05-29 21:17:19 -04:00
def restart_vm(zkhandler, domain, wait=False):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and verify we're OK to proceed
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state != "start":
return False, 'ERROR: VM "{}" is not in "start" state!'.format(domain)
retmsg = 'Restarting VM "{}".'.format(domain)
# Set the VM to restart
change_state(zkhandler, dom_uuid, "restart")
if wait:
while zkhandler.read(("domain.state", dom_uuid)) == "restart":
2021-06-13 13:41:21 -04:00
time.sleep(0.5)
retmsg = 'Restarted VM "{}"'.format(domain)
return True, retmsg
2021-05-29 21:17:19 -04:00
def shutdown_vm(zkhandler, domain, wait=False):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and verify we're OK to proceed
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state != "start":
return False, 'ERROR: VM "{}" is not in "start" state!'.format(domain)
retmsg = 'Shutting down VM "{}"'.format(domain)
# Set the VM to shutdown
change_state(zkhandler, dom_uuid, "shutdown")
if wait:
while zkhandler.read(("domain.state", dom_uuid)) == "shutdown":
2021-06-13 13:41:21 -04:00
time.sleep(0.5)
2020-02-19 10:04:58 -05:00
retmsg = 'Shut down VM "{}"'.format(domain)
return True, retmsg
2021-05-29 21:17:19 -04:00
def stop_vm(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
2021-05-29 21:17:19 -04:00
# Set the VM to stop
change_state(zkhandler, dom_uuid, "stop")
return True, 'Forcibly stopping VM "{}".'.format(domain)
def disable_vm(zkhandler, domain, force=False):
2019-10-23 23:37:42 -04:00
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
2019-10-23 23:37:42 -04:00
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and perform a shutdown/stop if VM is online
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state in ["start"]:
if force:
change_state(zkhandler, dom_uuid, "stop")
# Wait for the command to be registered by the node
time.sleep(0.5)
else:
change_state(zkhandler, dom_uuid, "shutdown")
# Wait for the shutdown to complete
while zkhandler.read(("domain.state", dom_uuid)) != "stop":
time.sleep(0.5)
2019-10-23 23:37:42 -04:00
2021-05-29 21:17:19 -04:00
# Set the VM to disable
change_state(zkhandler, dom_uuid, "disable")
2019-10-23 23:37:42 -04:00
return True, 'Disabled VM "{}".'.format(domain)
2019-10-23 23:37:42 -04:00
def update_vm_sriov_nics(zkhandler, dom_uuid, source_node, target_node):
# Update all the SR-IOV device states on both nodes, used during migrations but called by the node-side
vm_config = zkhandler.read(("domain.xml", dom_uuid))
parsed_xml = lxml.objectify.fromstring(vm_config)
dnetworks = common.getDomainNetworks(parsed_xml, {})
retcode = True
retmsg = ""
for network in dnetworks:
if network["type"] in ["direct", "hostdev"]:
# Check if the network is already in use
is_used = zkhandler.read(
("node.sriov.vf", target_node, "sriov_vf.used", network["source"])
)
if is_used == "True":
used_by_name = searchClusterByUUID(
zkhandler,
zkhandler.read(
(
"node.sriov.vf",
target_node,
"sriov_vf.used_by",
network["source"],
)
),
)
if retcode:
retcode_this = False
retmsg = 'Attempting to use SR-IOV network "{}" which is already used by VM "{}"'.format(
network["source"], used_by_name
)
else:
retcode_this = True
# We must update the "used" section
if retcode_this:
# This conditional ensure that if we failed the is_used check, we don't try to overwrite the information of a VF that belongs to another VM
set_sriov_vf_vm(
zkhandler,
dom_uuid,
target_node,
network["source"],
network["mac"],
network["type"],
)
# ... but we still want to free the old node in an case
unset_sriov_vf_vm(zkhandler, source_node, network["source"])
if not retcode_this:
retcode = retcode_this
return retcode, retmsg
2021-05-29 21:17:19 -04:00
def move_vm(zkhandler, domain, target_node, wait=False, force_live=False):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and verify we're OK to proceed
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state != "start":
# If the current state isn't start, preserve it; we're not doing live migration
target_state = current_state
else:
if force_live:
target_state = "migrate-live"
else:
target_state = "migrate"
current_node = zkhandler.read(("domain.node", dom_uuid))
2019-06-24 13:37:56 -04:00
if not target_node:
2021-05-29 21:17:19 -04:00
target_node = common.findTargetNode(zkhandler, dom_uuid)
else:
2019-06-24 13:25:24 -04:00
# Verify node is valid
2021-05-29 21:17:19 -04:00
valid_node = common.verifyNode(zkhandler, target_node)
2019-06-24 13:25:24 -04:00
if not valid_node:
2019-10-12 01:50:15 -04:00
return False, 'ERROR: Specified node "{}" is invalid.'.format(target_node)
2019-06-24 13:25:24 -04:00
# Check if node is within the limit
node_limit = zkhandler.read(("domain.meta.node_limit", dom_uuid))
if node_limit and target_node not in node_limit.split(","):
return (
False,
'ERROR: Specified node "{}" is not in the allowed list of nodes for VM "{}".'.format(
target_node, domain
),
)
2019-06-24 13:25:24 -04:00
# Verify if node is current node
if target_node == current_node:
last_node = zkhandler.read(("domain.last_node", dom_uuid))
if last_node:
zkhandler.write([(("domain.last_node", dom_uuid), "")])
return True, 'Making temporary migration permanent for VM "{}".'.format(
domain
)
return False, 'ERROR: VM "{}" is already running on node "{}".'.format(
domain, current_node
)
if not target_node:
return (
False,
'ERROR: Could not find a valid migration target for VM "{}".'.format(
domain
),
)
retmsg = 'Permanently migrating VM "{}" to node "{}".'.format(domain, target_node)
lock = zkhandler.exclusivelock(("domain.state", dom_uuid))
2021-05-29 21:17:19 -04:00
with lock:
zkhandler.write(
[
(("domain.state", dom_uuid), target_state),
(("domain.node", dom_uuid), target_node),
(("domain.last_node", dom_uuid), ""),
]
)
2021-05-29 21:17:19 -04:00
# Wait for 1/2 second for migration to start
time.sleep(0.5)
# Update any SR-IOV NICs
update_vm_sriov_nics(zkhandler, dom_uuid, current_node, target_node)
if wait:
while zkhandler.read(("domain.state", dom_uuid)) == target_state:
2021-06-13 13:41:21 -04:00
time.sleep(0.5)
retmsg = 'Permanently migrated VM "{}" to node "{}"'.format(domain, target_node)
return True, retmsg
def migrate_vm(
zkhandler, domain, target_node, force_migrate, wait=False, force_live=False
):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and verify we're OK to proceed
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state != "start":
# If the current state isn't start, preserve it; we're not doing live migration
target_state = current_state
else:
if force_live:
target_state = "migrate-live"
else:
target_state = "migrate"
current_node = zkhandler.read(("domain.node", dom_uuid))
last_node = zkhandler.read(("domain.last_node", dom_uuid))
2019-06-24 13:37:56 -04:00
if last_node and not force_migrate:
2020-02-19 10:18:41 -05:00
return False, 'ERROR: VM "{}" has been previously migrated.'.format(domain)
2019-06-24 13:37:56 -04:00
if not target_node:
2021-05-29 21:17:19 -04:00
target_node = common.findTargetNode(zkhandler, dom_uuid)
else:
2019-06-24 13:25:24 -04:00
# Verify node is valid
2021-05-29 21:17:19 -04:00
valid_node = common.verifyNode(zkhandler, target_node)
2019-06-24 13:25:24 -04:00
if not valid_node:
2019-10-12 01:50:15 -04:00
return False, 'ERROR: Specified node "{}" is invalid.'.format(target_node)
2019-06-24 13:25:24 -04:00
# Check if node is within the limit
node_limit = zkhandler.read(("domain.meta.node_limit", dom_uuid))
if node_limit and target_node not in node_limit.split(","):
return (
False,
'ERROR: Specified node "{}" is not in the allowed list of nodes for VM "{}".'.format(
target_node, domain
),
)
2019-06-24 13:25:24 -04:00
# Verify if node is current node
if target_node == current_node:
return False, 'ERROR: VM "{}" is already running on node "{}".'.format(
domain, current_node
)
if not target_node:
return (
False,
'ERROR: Could not find a valid migration target for VM "{}".'.format(
domain
),
)
# Don't overwrite an existing last_node when using force_migrate
real_current_node = current_node # Used for the SR-IOV update
if last_node and force_migrate:
current_node = last_node
retmsg = 'Migrating VM "{}" to node "{}".'.format(domain, target_node)
lock = zkhandler.exclusivelock(("domain.state", dom_uuid))
2021-05-29 21:17:19 -04:00
with lock:
zkhandler.write(
[
(("domain.state", dom_uuid), target_state),
(("domain.node", dom_uuid), target_node),
(("domain.last_node", dom_uuid), current_node),
]
)
2021-05-29 21:17:19 -04:00
# Wait for 1/2 second for migration to start
time.sleep(0.5)
# Update any SR-IOV NICs
update_vm_sriov_nics(zkhandler, dom_uuid, real_current_node, target_node)
if wait:
while zkhandler.read(("domain.state", dom_uuid)) == target_state:
2021-06-13 13:41:21 -04:00
time.sleep(0.5)
retmsg = 'Migrated VM "{}" to node "{}"'.format(domain, target_node)
return True, retmsg
2021-05-29 21:17:19 -04:00
def unmigrate_vm(zkhandler, domain, wait=False, force_live=False):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get state and verify we're OK to proceed
current_state = zkhandler.read(("domain.state", dom_uuid))
if current_state != "start":
2019-03-20 10:19:01 -04:00
# If the current state isn't start, preserve it; we're not doing live migration
target_state = current_state
else:
if force_live:
target_state = "migrate-live"
else:
target_state = "migrate"
current_node = zkhandler.read(("domain.node", dom_uuid))
target_node = zkhandler.read(("domain.last_node", dom_uuid))
if target_node == "":
return False, 'ERROR: VM "{}" has not been previously migrated.'.format(domain)
retmsg = 'Unmigrating VM "{}" back to node "{}".'.format(domain, target_node)
lock = zkhandler.exclusivelock(("domain.state", dom_uuid))
2021-05-29 21:17:19 -04:00
with lock:
zkhandler.write(
[
(("domain.state", dom_uuid), target_state),
(("domain.node", dom_uuid), target_node),
(("domain.last_node", dom_uuid), ""),
]
)
2021-05-29 21:17:19 -04:00
# Wait for 1/2 second for migration to start
time.sleep(0.5)
# Update any SR-IOV NICs
update_vm_sriov_nics(zkhandler, dom_uuid, current_node, target_node)
if wait:
while zkhandler.read(("domain.state", dom_uuid)) == target_state:
2021-06-13 13:41:21 -04:00
time.sleep(0.5)
retmsg = 'Unmigrated VM "{}" back to node "{}"'.format(domain, target_node)
return True, retmsg
2021-05-29 21:17:19 -04:00
def get_console_log(zkhandler, domain, lines=1000):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Get the data from ZK
console_log = zkhandler.read(("domain.console.log", dom_uuid))
if console_log is None:
return True, ""
# Shrink the log buffer to length lines
shrunk_log = console_log.split("\n")[-lines:]
loglines = "\n".join(shrunk_log)
return True, loglines
2021-05-29 21:17:19 -04:00
def get_info(zkhandler, domain):
# Validate that VM exists in cluster
2021-05-29 21:17:19 -04:00
dom_uuid = getDomainUUID(zkhandler, domain)
2019-05-20 22:15:28 -04:00
if not dom_uuid:
return False, 'ERROR: No VM named "{}" is present in the cluster.'.format(
domain
)
2019-05-20 22:15:28 -04:00
# Gather information from XML config and print it
2021-05-29 21:17:19 -04:00
domain_information = common.getInformationFromXML(zkhandler, dom_uuid)
2019-06-24 13:37:56 -04:00
if not domain_information:
2019-05-20 22:15:28 -04:00
return False, 'ERROR: Could not get information about VM "{}".'.format(domain)
return True, domain_information
def get_list(
zkhandler, node=None, state=None, tag=None, limit=None, is_fuzzy=True, negate=False
):
if node is not None:
# Verify node is valid
2021-05-29 21:17:19 -04:00
if not common.verifyNode(zkhandler, node):
return False, 'Specified node "{}" is invalid.'.format(node)
if state is not None:
valid_states = [
"start",
"restart",
"shutdown",
"stop",
"disable",
"fail",
"migrate",
"unmigrate",
"provision",
]
if state not in valid_states:
2019-03-20 11:31:54 -04:00
return False, 'VM state "{}" is not valid.'.format(state)
full_vm_list = zkhandler.children("base.domain")
full_vm_list.sort()
2019-03-12 23:52:59 -04:00
# Set our limit to a sensible regex
if limit is not None:
# Check if the limit is a UUID
is_limit_uuid = False
2019-03-12 23:52:59 -04:00
try:
uuid_obj = UUID(limit, version=4)
limit = str(uuid_obj)
is_limit_uuid = True
except ValueError:
pass
if is_fuzzy and not is_limit_uuid:
try:
# Implcitly assume fuzzy limits
if not re.match(r"\^.*", limit):
limit = ".*" + limit
if not re.match(r".*\$", limit):
limit = limit + ".*"
except Exception as e:
return False, "Regex Error: {}".format(e)
2019-03-12 23:52:59 -04:00
get_vm_info = dict()
2018-09-25 02:20:32 -04:00
for vm in full_vm_list:
name = zkhandler.read(("domain", vm))
is_limit_match = False
is_tag_match = False
is_node_match = False
is_state_match = False
# Check on limit
if limit is not None:
# Try to match the limit against the UUID (if applicable) and name
try:
if is_limit_uuid and re.fullmatch(limit, vm):
is_limit_match = True
if re.fullmatch(limit, name):
is_limit_match = True
except Exception as e:
return False, "Regex Error: {}".format(e)
else:
is_limit_match = True
if tag is not None:
vm_tags = zkhandler.children(("domain.meta.tags", vm))
if negate and tag not in vm_tags:
is_tag_match = True
if not negate and tag in vm_tags:
is_tag_match = True
else:
is_tag_match = True
# Check on node
if node is not None:
vm_node = zkhandler.read(("domain.node", vm))
if negate and vm_node != node:
is_node_match = True
if not negate and vm_node == node:
is_node_match = True
else:
is_node_match = True
# Check on state
if state is not None:
vm_state = zkhandler.read(("domain.state", vm))
if negate and vm_state != state:
is_state_match = True
if not negate and vm_state == state:
is_state_match = True
else:
is_state_match = True
get_vm_info[vm] = (
True
if is_limit_match and is_tag_match and is_node_match and is_state_match
else False
)
# Obtain our VM data in a thread pool
# This helps parallelize the numerous Zookeeper calls a bit, within the bounds of the GIL, and
# should help prevent this task from becoming absurdly slow with very large numbers of VMs.
# The max_workers is capped at 32 to avoid creating an absurd number of threads especially if
# the list gets called multiple times simultaneously by the API, but still provides a noticeable
# speedup.
vm_execute_list = [vm for vm in full_vm_list if get_vm_info[vm]]
vm_data_list = list()
with ThreadPoolExecutor(max_workers=32, thread_name_prefix="vm_list") as executor:
futures = []
for vm_uuid in vm_execute_list:
futures.append(
executor.submit(common.getInformationFromXML, zkhandler, vm_uuid)
)
for future in futures:
try:
vm_data_list.append(future.result())
except Exception:
pass
return True, sorted(vm_data_list, key=lambda d: d["name"])
def backup_vm(
zkhandler, domain, target_path, incremental_parent=None, retain_snapshots=False
):
2023-10-17 12:39:37 -04:00
tstart = time.time()
2023-10-17 12:32:40 -04:00
# 0. Validations
# Validate that VM exists in cluster
dom_uuid = getDomainUUID(zkhandler, domain)
if not dom_uuid:
return False, 'ERROR: Could not find VM "{}" in the cluster!'.format(domain)
# Validate that the target path exists
if not re.match(r"^/", target_path):
return (
False,
f"ERROR: Target path {target_path} is not a valid absolute path on the primary coordinator!",
)
# Ensure that target_path (on this node) exists
if not os.path.isdir(target_path):
return False, f"ERROR: Target path {target_path} does not exist!"
# 1. Get information about VM
2023-10-17 11:51:03 -04:00
vm_detail = get_list(zkhandler, limit=dom_uuid, is_fuzzy=False)[1][0]
if not isinstance(vm_detail, dict):
return False, f"ERROR: VM listing returned invalid data: {vm_detail}"
vm_volumes = [
tuple(d["name"].split("/")) for d in vm_detail["disks"] if d["type"] == "rbd"
]
# 2a. Validate that all volumes exist (they should, but just in case)
for pool, volume in vm_volumes:
if not ceph.verifyVolume(zkhandler, pool, volume):
return (
False,
f"ERROR: VM defines a volume {pool}/{volume} which does not exist!",
)
# 2b. Validate that, if an incremental_parent is given, it is valid
# The incremental parent is just a datestring
if incremental_parent is not None:
for pool, volume in vm_volumes:
if not ceph.verifySnapshot(
zkhandler, pool, volume, f"backup_{incremental_parent}"
):
return (
False,
2023-10-17 12:17:29 -04:00
f"ERROR: Incremental parent {incremental_parent} given, but no snapshots were found; cannot export an incremental backup.",
)
export_fileext = "rbddiff"
else:
export_fileext = "rbdimg"
# 3. Set datestring in YYYYMMDDHHMMSS format
now = datetime.now()
2023-10-17 12:05:45 -04:00
datestring = now.strftime("%Y%m%d%H%M%S")
snapshot_name = f"backup_{datestring}"
# 4. Create destination directory
vm_target_root = f"{target_path}/{domain}"
vm_target_backup = f"{target_path}/{domain}/.{datestring}"
if not os.path.isdir(vm_target_backup):
try:
os.makedirs(vm_target_backup)
except Exception as e:
return False, f"ERROR: Failed to create backup directory: {e}"
# 5. Take snapshot of each disks with the name @backup_{datestring}
is_snapshot_create_failed = False
which_snapshot_create_failed = list()
msg_snapshot_create_failed = list()
for pool, volume in vm_volumes:
retcode, retmsg = ceph.add_snapshot(zkhandler, pool, volume, snapshot_name)
if not retcode:
is_snapshot_create_failed = True
which_snapshot_create_failed.append(f"{pool}/{volume}")
msg_snapshot_create_failed.append(retmsg)
if is_snapshot_create_failed:
for pool, volume in vm_volumes:
if ceph.verifySnapshot(zkhandler, pool, volume, snapshot_name):
ceph.remove_snapshot(zkhandler, pool, volume, snapshot_name)
return (
False,
f'ERROR: Failed to create snapshot for volume(s) {", ".join(which_snapshot_create_failed)}: {", ".join(msg_snapshot_create_failed)}',
)
# 6. Dump snapshot to folder with `rbd export` (full) or `rbd export-diff` (incremental)
is_snapshot_export_failed = False
which_snapshot_export_failed = list()
msg_snapshot_export_failed = list()
for pool, volume in vm_volumes:
if incremental_parent is not None:
incremental_parent_snapshot_name = f"backup_{incremental_parent}"
retcode, stdout, stderr = common.run_os_command(
f"rbd export-diff --from-snap {incremental_parent_snapshot_name} {pool}/{volume}@{snapshot_name} {vm_target_backup}/{volume}.{export_fileext}"
)
if retcode:
is_snapshot_export_failed = True
which_snapshot_export_failed.append(f"{pool}/{volume}")
msg_snapshot_export_failed.append(stderr)
else:
retcode, stdout, stderr = common.run_os_command(
f"rbd export --export-format 2 {pool}/{volume}@{snapshot_name} {vm_target_backup}/{volume}.{export_fileext}"
)
if retcode:
is_snapshot_export_failed = True
which_snapshot_export_failed.append(f"{pool}/{volume}")
msg_snapshot_export_failed.append(stderr)
if is_snapshot_export_failed:
for pool, volume in vm_volumes:
if ceph.verifySnapshot(zkhandler, pool, volume, snapshot_name):
ceph.remove_snapshot(zkhandler, pool, volume, snapshot_name)
return (
False,
f'ERROR: Failed to export snapshot for volume(s) {", ".join(which_snapshot_export_failed)}: {", ".join(msg_snapshot_export_failed)}',
)
# 7. Create and dump VM backup information
2023-10-17 12:10:55 -04:00
backup_type = "incremental" if incremental_parent is not None else "full"
vm_backup = {
2023-10-17 12:10:55 -04:00
"type": backup_type,
"datestring": datestring,
"incremental_parent": incremental_parent,
"vm_detail": vm_detail,
"backup_files": [f".{datestring}/{v}.{export_fileext}" for p, v in vm_volumes],
}
with open(f"{vm_target_root}/{domain}.{datestring}.pvcbackup", "w") as fh:
2023-10-17 12:02:24 -04:00
jdump(vm_backup, fh)
# 8. Remove snapshots if retain_snapshot is False
2023-10-17 12:32:40 -04:00
is_snapshot_remove_failed = False
which_snapshot_remove_failed = list()
msg_snapshot_remove_failed = list()
if not retain_snapshots:
for pool, volume in vm_volumes:
if ceph.verifySnapshot(zkhandler, pool, volume, snapshot_name):
retcode, retmsg = ceph.remove_snapshot(
zkhandler, pool, volume, snapshot_name
)
if not retcode:
is_snapshot_remove_failed = True
which_snapshot_remove_failed.append(f"{pool}/{volume}")
msg_snapshot_remove_failed.append(retmsg)
2023-10-17 12:39:37 -04:00
tend = time.time()
ttot = round(tend - tstart, 2)
2023-10-17 12:32:40 -04:00
if is_snapshot_remove_failed:
2023-10-17 12:53:08 -04:00
retmsg = (
f"WARNING: Successfully backed up VM '{domain}' ({backup_type} @ {datestring}) to '{target_path}' in {ttot} seconds, but failed to remove snapshot as requested for volume(s) {', '.join(which_snapshot_remove_failed)}: {', '.join(msg_snapshot_remove_failed)}",
)
elif retain_snapshots:
retmsg = (
f"Successfully backed up VM '{domain}' ({backup_type} @ {datestring}, snapshots retained) to '{target_path}' in {ttot} seconds.",
)
else:
retmsg = (
f"Successfully backed up VM '{domain}' ({backup_type} @ {datestring}) to '{target_path}' in {ttot} seconds.",
2023-10-17 12:32:40 -04:00
)
2023-10-17 12:53:08 -04:00
return (True, retmsg)